# ![@aaryan_kakad Avatar](https://lunarcrush.com/gi/w:26/cr:twitter::1474816149338542081.png) @aaryan_kakad Aaryan Kakad

Aaryan Kakad posts on X about ai, seedance, llm, check the most. They currently have [-----] followers and [---] posts still getting attention that total [-------] engagements in the last [--] hours.

### Engagements: [-------] [#](/creator/twitter::1474816149338542081/interactions)
![Engagements Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1474816149338542081/c:line/m:interactions.svg)

- [--] Week [-------] +838%
- [--] Month [-------] +358%
- [--] Months [-------] +537%
- [--] Year [-------] +13%

### Mentions: [--] [#](/creator/twitter::1474816149338542081/posts_active)
![Mentions Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1474816149338542081/c:line/m:posts_active.svg)

- [--] Week [--] +45%
- [--] Month [---] +39%
- [--] Months [---] +341%
- [--] Year [---] +101%

### Followers: [-----] [#](/creator/twitter::1474816149338542081/followers)
![Followers Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1474816149338542081/c:line/m:followers.svg)

- [--] Week [-----] +5%
- [--] Month [-----] +8.60%
- [--] Months [-----] +20%
- [--] Year [-----] +68%

### CreatorRank: [------] [#](/creator/twitter::1474816149338542081/influencer_rank)
![CreatorRank Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1474816149338542081/c:line/m:influencer_rank.svg)

### Social Influence

**Social category influence**
[technology brands](/list/technology-brands)  7.91% [finance](/list/finance)  3.6% [celebrities](/list/celebrities)  3.6% [countries](/list/countries)  2.16% [stocks](/list/stocks)  1.44% [social networks](/list/social-networks)  1.44% [travel destinations](/list/travel-destinations)  0.72%

**Social topic influence**
[ai](/topic/ai) 16.55%, [seedance](/topic/seedance) #175, [llm](/topic/llm) #660, [check](/topic/check) #1659, [if you](/topic/if-you) 5.04%, [in the](/topic/in-the) 5.04%, [6969](/topic/6969) #92, [model](/topic/model) #812, [philosophy](/topic/philosophy) #380, [agentic](/topic/agentic) 2.88%

**Top accounts mentioned or mentioned by**
[@karpathy](/creator/undefined) [@daviddeutschoxf](/creator/undefined) [@karpathys](/creator/undefined) [@aaryankakad](/creator/undefined) [@kimimoonshot](/creator/undefined) [@grok](/creator/undefined) [@naval](/creator/undefined) [@kimimoonshots](/creator/undefined) [@omarsar0](/creator/undefined) [@raydalio](/creator/undefined) [@langchain](/creator/undefined) [@a1zhang](/creator/undefined) [@lateinteraction](/creator/undefined) [@ehuanglu](/creator/undefined) [@forallcurious](/creator/undefined) [@deedydas](/creator/undefined) [@aiforsuccess](/creator/undefined) [@shunyuyao12](/creator/undefined) [@sterlingcooley](/creator/undefined) [@peterocasek](/creator/undefined)
### Top Social Posts
Top posts by engagements in the last [--] hours

"🚨I implemented the viral RLM paper from MIT Something that shouldn't be possible just happened. I built a tool that eats 10M token documents for breakfast. Not static RAG. Not embeddings. Pure recursive LLM magic. Upload any doc. Chat with it. Instant answers. Thats it. Demo: Code: Paper credits: @a1zhang & @lateinteraction Paper explanation here: https://github.com/AKMessi/rlm-paper-implementation https://rlm-ucnx.onrender.com/web https://github.com/AKMessi/rlm-paper-implementation https://rlm-ucnx.onrender.com/web"  
[X Link](https://x.com/aaryan_kakad/status/2020861808626880932)  2026-02-09T14:06Z [----] followers, 11K engagements


"@EHuanglu the seedance [---] model did this :) this model is actually insane. i actively explored it and read some articles around it. found a lot of interesting stuff. check it out here: 🚨 EVERYTHING YOU NEED TO KNOW ABOUT SEEDANCE [---] I have actively explored it and read a few articles about it: The articles claimed Seedance [---] uses a Dual Branch Diffusion Transformer: I loved this approach. Combining diffusion models with transformers. Diffusion process: 🚨 EVERYTHING YOU NEED TO KNOW ABOUT SEEDANCE [---] I have actively explored it and read a few articles about it: The articles claimed"  
[X Link](https://x.com/aaryan_kakad/status/2022557920106631508)  2026-02-14T06:26Z [----] followers, [----] engagements


"This is research code. Don't deploy it in a hospital. But it shows that a single engineer with PyTorch and a Kaggle notebook can build something that actually looks at the anatomy - not just dataset statistics. Dataset: Indiana University Chest X-Rays (public)"  
[X Link](https://x.com/aaryan_kakad/status/2022671606875918757)  2026-02-14T13:57Z [----] followers, [--] engagements


"This democratizes medical AI. You don't need Google's data pipeline or Stanford's compute cluster. You need: 3k labeled images A Hybrid ViT architecture Focal loss to handle imbalance Attention visualization to verify it's looking at anatomy Radiology screening shouldn't be locked behind proprietary models trained on hospital consortiums. It should be reproducible by any researcher with a GPU. https://twitter.com/i/web/status/2022671604111860162 https://twitter.com/i/web/status/2022671604111860162"  
[X Link](https://x.com/aaryan_kakad/status/2022671604111860162)  2026-02-14T13:57Z [----] followers, [--] engagements


"@forallcurious AI (especially DL) for healthcare is even more interesting. earlier today i finished training MEDVLM - a hybrid vision transformer model that reads chest x-rays like a PRO. check it out here: 🚨 I just built MedVLM - a hybrid vision transformer that reads chest X-rays like a radiologist. It matches commercial medical AI performance with [---] less data (3835 vs 300k+ images) and fits on a single GPU. The secret Treating an X-ray not as an image but as a sequence https://t.co/tB4RehbVNI 🚨 I just built MedVLM - a hybrid vision transformer that reads chest X-rays like a"  
[X Link](https://x.com/aaryan_kakad/status/2022767077812289596)  2026-02-14T20:17Z [----] followers, [----] engagements


"@deedydas chinese have even dropped the best video gen model based on diffusion transformer architecture - seedance [---]. it is even crazier than these OSS models"  
[X Link](https://x.com/aaryan_kakad/status/2022495226376684011)  2026-02-14T02:16Z [----] followers, [---] engagements


"🚨 not just that i have read an article about seedance [---] that claims: if the claims are true it has used a dual branch design of the diffusion transformer architecture. - separate spatial and temporal transformer branches. that can improve: motion realism camera physics identity retention - native audio-video joint diffusion. many models: generate video then add audio separately seedance claims: joint multimodal generation shared latent space for video-sound this is a big difference than other video models. also they have the TIKTOK data advantage. 🚨 seedance [---] is actually something."  
[X Link](https://x.com/aaryan_kakad/status/2022503663810548115)  2026-02-14T02:50Z [----] followers, 10.1K engagements


"🚨 I just built MedVLM - a hybrid vision transformer that reads chest X-rays like a radiologist. It matches commercial medical AI performance with [---] less data (3835 vs 300k+ images) and fits on a single GPU. The secret Treating an X-ray not as an image but as a sequence of [--] anatomical "words" that the transformer reads left-to-right top-to-bottom. Here's how it works: 🧵"  
[X Link](https://x.com/aaryan_kakad/status/2022671576936955982)  2026-02-14T13:57Z [----] followers, [----] engagements


"the entire article summarized in [--] slides: 🧵 https://t.co/tjmbT5ytUN https://t.co/tjmbT5ytUN"  
[X Link](https://x.com/aaryan_kakad/status/2023114420055069002)  2026-02-15T19:17Z [----] followers, 13.7K engagements


"@ai_for_success seedance [---] is an insane model. i researched some stuff around it read some articles and found a lot of intriguing stuff. check this out: 🚨 EVERYTHING YOU NEED TO KNOW ABOUT SEEDANCE [---] I have actively explored it and read a few articles about it: The articles claimed Seedance [---] uses a Dual Branch Diffusion Transformer: I loved this approach. Combining diffusion models with transformers. Diffusion process: 🚨 EVERYTHING YOU NEED TO KNOW ABOUT SEEDANCE [---] I have actively explored it and read a few articles about it: The articles claimed Seedance [---] uses a Dual Branch"  
[X Link](https://x.com/aaryan_kakad/status/2022573041969172781)  2026-02-14T07:26Z [----] followers, [---] engagements


"🚨 If you could only read one AI paper in your entire lifetime. it could be this I asked Kimi K2.5 "If you had to read only one paper on Agentic AI in your entire lifetime what paper would you read". It gave me this ReAct framework paper published by @ShunyuYao12 along with his co-authors. And definitely this was a revolutionary thing in Agentic AI. Before ReAct LLMs were either chatbots or fragile pipelines. It introduced the Thought-Action-Observation loop. This simple loop is the DNA of almost every enterprise agent copilot and autonomous system running today. It bridges capability and"  
[X Link](https://x.com/aaryan_kakad/status/2016972827602407713)  2026-01-29T20:32Z [----] followers, [---] engagements


"@SterlingCooley @karpathy @DavidDeutschOxf i optimistically believe - there will be a new architecture or some improvement that will solve this limitation of just generating text based on whats already written. its still not here but it will definitely be here soon"  
[X Link](https://x.com/aaryan_kakad/status/2017778848889454703)  2026-02-01T01:55Z [----] followers, [---] engagements


"🚨 I GOT TIRED OF SEARCHING FOR ML PRACTICE PROBLEMS So I built Neural Forge: [--] questions across a lot of ML topics. Mathematics Deep Learning Reinforcement Learning Classical ML everything. It has [--] question types not just MCQs. Check it out here: (FREE until I get enough feedback) https://neural-forge-chi.vercel.app https://neural-forge-chi.vercel.app"  
[X Link](https://x.com/aaryan_kakad/status/2019013255558266982)  2026-02-04T11:40Z [----] followers, [---] engagements


"🚨 Barely processed Kimi K2.5 dropping SOTA open-source bombs and Claude hits us with Opus [---] + OpenAI drops GPT Codex [---] on the SAME DAY 😭 These are my honest views on the comparison of Chinese OSS models and American Closed Models: Chinese OSS models are way cheaper and value for money compared to models like Opus [---] The only thing models like Opus [---] and GPT [---] are better at is that they have the best worst-case reasoning. I am a huge believer of EFFICIENCY and thats the reason I openly support Chinese OSS models or in general OSS models that are cost and energy efficient. Chinese"  
[X Link](https://x.com/aaryan_kakad/status/2019485791479800258)  2026-02-05T18:58Z [----] followers, [----] engagements


"@PeterOcasek the email shown in the video is simply demo and completely unpolished and unideal. it was to show that the crawler was effective to find out personalizations. the perfect email needs a lot of good prompt engineering i have not focused on it much yet"  
[X Link](https://x.com/aaryan_kakad/status/2020038523827679263)  2026-02-07T07:34Z [----] followers, [--] engagements


"kimi k2.5 is the only llm that disagrees with me strong sign of a good llm"  
[X Link](https://x.com/aaryan_kakad/status/2020208411946999869)  2026-02-07T18:49Z [----] followers, [---] engagements


"@omarsar0 definitely"  
[X Link](https://x.com/aaryan_kakad/status/2020237200743694782)  2026-02-07T20:44Z [----] followers, [--] engagements


"need your suggestion on this IMPORTANT thing: i am thinking to build a deep research agent that: takes my query plans what tools it needs to use if it needs web search it uses the web search tool finds out as much quality info as possible structures it properly into clean and clear dataset/spreadsheet this will really be helpful for everyone if they want to perform searches like: what are the best value for money GPUs to buy in [----] what are the best LLMs to use for copywriting give me the best advice to maximize my muscle growth. let me know your honest opinion if i get enough positive"  
[X Link](https://x.com/aaryan_kakad/status/2020246673310773564)  2026-02-07T21:21Z [----] followers, [--] engagements


"3/ attention is literally all you need your life isn't what you think about. it's what you attend to. if your attention heads are focused on toxic inputs infinite scroll and low-signal people your output will be garbage. you don't need more time. you need a better sparsity mask on your context window. curate what enters your hidden state like your life depends on it. because it does"  
[X Link](https://x.com/aaryan_kakad/status/2020524672648143185)  2026-02-08T15:46Z [----] followers, [--] engagements


"4/ backpropagation of responsibility when a model fails we don't blame the output. we trace the error back through every weight that contributed. in life we call this "accountability." stop cursing the "world" (the output layer). trace the gradient back to your choices (the weights). fix the source or the error repeats every epoch. own your initialization. or keep blaming the dataset. https://twitter.com/i/web/status/2020524677177991310 https://twitter.com/i/web/status/2020524677177991310"  
[X Link](https://x.com/aaryan_kakad/status/2020524677177991310)  2026-02-08T15:46Z [----] followers, [--] engagements


"How it works: Traditional RAG: Document Chunks Embeddings Similarity Search Answer ❌ Static misses connections embedding bottleneck RLM: Document REPL Variable LLM Writes Python Code Examines/Chunks Recursive llm_query() Final Answer ✅ Dynamic programmable unbounded The LLM treats the prompt as a variable and writes code to process it. Algorithm [--] from the paper implemented exactly. https://twitter.com/i/web/status/2020861813030637583 https://twitter.com/i/web/status/2020861813030637583"  
[X Link](https://x.com/aaryan_kakad/status/2020861813030637583)  2026-02-09T14:06Z [----] followers, [---] engagements


"@UnslothAI @huggingface damn will definitely check it out 12.8GB VRAM fascinates me the most"  
[X Link](https://x.com/aaryan_kakad/status/2021262083048853959)  2026-02-10T16:36Z [----] followers, [----] engagements


"@TheAhmadOsman definitely man. i have been using kimi code for literally everything. also tried codex gpt [---] today it works well too. kimi is the overall king for me rn. its the only llm that disagrees with me (a sign of a good llm)"  
[X Link](https://x.com/aaryan_kakad/status/2021327087324602787)  2026-02-10T20:55Z [----] followers, [----] engagements


"🚨 I gave Codex-GPT-5.3 full freedom and one instruction: build the best creation of its lifetime. It built a living universe sim. Factions evolve trade mutate wage wars survive cataclysms and react to anomalies in real time. Vanilla JS. Zero deps. In the clip below: emergent faction behavior anomaly physics live telemetry + trend graph export/import world states cinematic mode If this gets enough interest Ill drop the full code + architecture breakdown. https://twitter.com/i/web/status/2021500636630835522 https://twitter.com/i/web/status/2021500636630835522"  
[X Link](https://x.com/aaryan_kakad/status/2021500636630835522)  2026-02-11T08:24Z [----] followers, [---] engagements


"GLM-5.0 is finally OUT. I asked it "Tell me the best thing you can ever do in your entire lifetime." And it answered: A new model is now available on https://t.co/gocggrfb3U. https://t.co/KZGoAsN5Z0 A new model is now available on https://t.co/gocggrfb3U. https://t.co/KZGoAsN5Z0"  
[X Link](https://x.com/aaryan_kakad/status/2021575271250296944)  2026-02-11T13:21Z [----] followers, [----] engagements


"@testingcatalog GLM-5 wants to save lives of humans. overall the model looks intelligent will test it on multiple things tonight. check this out: https://x.com/aaryan_kakad/status/2021575271250296944s=20 GLM-5.0 is finally OUT. I asked it "Tell me the best thing you can ever do in your entire lifetime." And it answered: https://t.co/60WpVweBKR https://x.com/aaryan_kakad/status/2021575271250296944s=20 GLM-5.0 is finally OUT. I asked it "Tell me the best thing you can ever do in your entire lifetime." And it answered: https://t.co/60WpVweBKR"  
[X Link](https://x.com/aaryan_kakad/status/2021575655205261603)  2026-02-11T13:22Z [----] followers, [----] engagements


"sir karpathy is really cooking stuff around transformers. i followed his nanogpt tutorial and trained a nanogpt on a lot of stuff absolutely loved it. gonna try this microgpt asap. hes sir karpathy for a reason man. insane. check this nanogpt i tried training on the book the beginning of infinity: 🚨 I trained a GPT-style language model on a single philosophy book. Inspired by @karpathy's tutorial I built a nanoGPT from scratch and trained it on The Beginning of Infinity by @DavidDeutschOxf. @karpathy used the character-level tokenizer from the tutorial I used tiktoken https://t.co/dpqznrbpUu"  
[X Link](https://x.com/aaryan_kakad/status/2021696953042821357)  2026-02-11T21:24Z [----] followers, [---] engagements


"i recently trained a nanogpt on the book the beginning of infinity followed @karpathys tutorial for it. and now the GOAT has dropped a new bombshell MICROGPT train a GPT in just [---] lines with ZERO deps. not even PyTorch. insane man. im going to try building something cool with it asap. if yall have any ideas on what should i try training a microgpt or any idea share it below. ill pick out the best one and create a detailed video on that microgpt project with the person featured in it. KUDOS New art project. Train and inference GPT in [---] lines of pure dependency-free Python. This is the"  
[X Link](https://x.com/aaryan_kakad/status/2021711364612583808)  2026-02-11T22:22Z [----] followers, [---] engagements


"@justinskycak definitely so I have chosen mastering ML from scratch and applying it to all the fields I love from the bottom of my heart thats the way to thrive in this AI era P.S. i also absolutely love ML lol"  
[X Link](https://x.com/aaryan_kakad/status/2021724848268206346)  2026-02-11T23:15Z [----] followers, [--] engagements


"@NicHulscher everything you need to know about creatine in ONE website. link: checkout this: https://x.com/aaryan_kakad/status/2020940122624307695s=20 https://creatine-sandy.vercel.app 🚨 CREATINE is the the most researched supplement of all time I have been taking creatine (5gm/day) regularly for [--] years now and now thinking to increase my per day dosage to 10-15gm/day. Many people doubt supplements like creatine and they think it has a lot of side effects https://t.co/dGw02ffEfh https://x.com/aaryan_kakad/status/2020940122624307695s=20 https://creatine-sandy.vercel.app 🚨 CREATINE is the"  
[X Link](https://x.com/aaryan_kakad/status/2021731907177001135)  2026-02-11T23:43Z [----] followers, [---] engagements


"🚨 Yesterday I built a multi-agent AI research lab. [--] specialist agents: Visionary Architect Critic Synthesizer Experimentalist Evidence. I gave them one question: "How do we build AGI" Then I watched them debate for [--] rounds. They literally invented a complete new theory of intelligence. "Homeostatic Morphogenetic Intelligence" - a theory where AI grows its own neural architecture like a biological brain with "metabolic costs" preventing unlimited growth. They wrote the full paper. Generated code. Reached consensus. I just watched. Research: What task should I try giving them next P.S. I"  
[X Link](https://x.com/aaryan_kakad/status/2021741363612811479)  2026-02-12T00:21Z [----] followers, [---] engagements


"This is what Kimi K2.5 said on the research report: THE BOTTOM LINE What the Agents Got RIGHT: [--]. Identified a genuine research direction dynamic inference-time architectures ARE underexplored [--]. Connected disparate fields Active Inference + NAS + GSP + Neuromorphic computing [--]. Proposed testable predictions sub-linear scaling self-healing System [--] emergence [--]. Acknowledged limitations didn't claim this solves everything What the Agents Got WRONG: [--]. Underestimated the credit assignment problem training dynamic graphs is notoriously unstable [--]. Overstated "metabolic" efficiency without"  
[X Link](https://x.com/aaryan_kakad/status/2021743339289423969)  2026-02-12T00:29Z [----] followers, [--] engagements


"@omarsar0 @dair_ai I implemented RLM from scratch and built a retrieval tool based on it. Check it out here: 🚨I implemented the viral RLM paper from MIT Something that shouldn't be possible just happened. I built a tool that eats 10M token documents for breakfast. Not static RAG. Not embeddings. Pure recursive LLM magic. Upload any doc. Chat with it. Instant answers. Thats it. https://t.co/BpNktks6CR 🚨I implemented the viral RLM paper from MIT Something that shouldn't be possible just happened. I built a tool that eats 10M token documents for breakfast. Not static RAG. Not embeddings. Pure"  
[X Link](https://x.com/aaryan_kakad/status/2021955515522412942)  2026-02-12T14:32Z [----] followers, [---] engagements


"save her at any cost and marry her up asap"  
[X Link](https://x.com/aaryan_kakad/status/2021956930848354382)  2026-02-12T14:37Z [----] followers, [---] engagements


"@elonmusk @grok @grok"  
[X Link](https://x.com/aaryan_kakad/status/2022047615517601876)  2026-02-12T20:38Z [----] followers, [--] engagements


"RT @aaryan_kakad: 🚨 not just that i have read an article about seedance [---] that claims: if the claims are true it has used a dual bran"  
[X Link](https://x.com/anyuser/status/2022513348961800522)  2026-02-14T03:28Z [----] followers, [--] engagements


"Watch the attention evolve across generation: Step 1-2: Broad scan (gathering context) Step [--] ("heart"): Focus narrows to cardiac region Step 6+: Checks surrounding anatomy The model learned that "heart" is spatially located in the center-left of the chest without ever being told. It discovered human anatomy from the attention patterns alone"  
[X Link](https://x.com/aaryan_kakad/status/2022671587766329552)  2026-02-14T13:57Z [----] followers, [--] engagements


"RT @aaryan_kakad: 🚨 EVERYTHING YOU NEED TO KNOW ABOUT SEEDANCE [---] I have actively explored it and read a few articles about it: The arti"  
[X Link](https://x.com/anyuser/status/2022747462319145280)  2026-02-14T18:59Z [----] followers, [--] engagements


"RT @aaryan_kakad: 🚨 seedance [---] is actually something. read that it is based on diffusion transformer architecture. loved the approach"  
[X Link](https://x.com/aaryan_kakad/status/2022747495336640968)  2026-02-14T18:59Z [----] followers, [--] engagements


"i read that it is based on dual-branch diffusion transformer. loved the approach. combining diffusion models with transformers. diffusion process: like image diffusion models (e.g. stable diffusion) it generates video by iteratively denoising latent representations toward the target output. transformer: the use of transformers means the model uses attention mechanisms to relate information across space and time - key for maintaining temporal consistency and multi-frame coherence. most impressive part is it doesn't generate a video first and then add audio in it it generates both synchronously"  
[X Link](https://x.com/aaryan_kakad/status/2022490964271731031)  2026-02-14T01:59Z [----] followers, 13.8K engagements


"🚨 seedance [---] is actually something. read that it is based on diffusion transformer architecture. loved the approach. combining diffusion models with transformers. diffusion process: like image diffusion models (e.g. stable diffusion) it generates video by iteratively denoising latent representations toward the target output. transformer: the use of transformers means the model uses attention mechanisms to relate information across space and time - key for maintaining temporal consistency and multi-frame coherence. most impressive part is it doesn't generate a video first and then add audio"  
[X Link](https://x.com/aaryan_kakad/status/2022501243453870473)  2026-02-14T02:40Z [----] followers, 10.2K engagements


"Most medical AI uses CNNs (local only) or pure ViTs (needs 100M+ images). I used a Hybrid ViT: ResNet-50 extracts local features (hardcoded spatial bias) Projects to [--] patch grid (49 tokens) Transformer reads the sequence with global attention Decoder generates the report token-by-token 74M parameters. Trains in [--] epochs on a Kaggle P100. https://twitter.com/i/web/status/2022671581080883323 https://twitter.com/i/web/status/2022671581080883323"  
[X Link](https://x.com/aaryan_kakad/status/2022671581080883323)  2026-02-14T13:57Z [----] followers, [--] engagements


""AI is the best teacher for the self-taught." @naval This quote has been shaping my life recently. And yes definitely AI is the best catalyst best teacher and best thing for all the self-learners. I have been self learning AI (specifically ML DL and Agentic AI) for over a year. And AI (LLMs) itself has helped me a lot as a mentor or you can say accountability partner. It has helped me by giving me a good roadmap to learn then choosing the best resources to learn from also helped me in choosing the best ways to learn. AI is actually a real blessing for all the self learners. And yes you can"  
[X Link](https://x.com/aaryan_kakad/status/2023093431850139955)  2026-02-15T17:53Z [----] followers, [--] engagements


"whats stopping you from printing things and then reading them while scribbling random stuff on it whats stopping you from taking a paper (physically) and scribbling on it like this dont miss out on this ultimate satisfaction just because AI can explain it in simple words. my ML bros do you relate https://t.co/vfe6ktiPkh whats stopping you from taking a paper (physically) and scribbling on it like this dont miss out on this ultimate satisfaction just because AI can explain it in simple words. my ML bros do you relate https://t.co/vfe6ktiPkh"  
[X Link](https://x.com/aaryan_kakad/status/2023109282120728744)  2026-02-15T18:56Z [----] followers, [--] engagements


"for those who dont want to read the complete article. here's the entire article explained in just [--] minutes: https://t.co/tjmbT5ytUN https://t.co/tjmbT5ytUN"  
[X Link](https://x.com/aaryan_kakad/status/2023116229180006696)  2026-02-15T19:24Z [----] followers, 178.4K engagements


"@RayDalio for those who dont to read and instead watch the article in a video format. here's the entire article explained in [--] mins: https://x.com/aaryan_kakad/status/2023116229180006696s=20 for those who dont want to read the complete article. here's the entire article explained in just [--] minutes: https://t.co/XOA4TAgaan https://x.com/aaryan_kakad/status/2023116229180006696s=20 for those who dont want to read the complete article. here's the entire article explained in just [--] minutes: https://t.co/XOA4TAgaan"  
[X Link](https://x.com/aaryan_kakad/status/2023116388924297419)  2026-02-15T19:25Z [----] followers, 173.5K engagements


"🚨 I trained a GPT-style language model on a single philosophy book. Inspired by @karpathy's tutorial I built a nanoGPT from scratch and trained it on The Beginning of Infinity by @DavidDeutschOxf. @karpathy used the character-level tokenizer from the tutorial I used tiktoken (GPT-2 BPE) - much closer to real LLMs but far more compute-intensive. 🧠 6.9M parameters 📚 Entire book as dataset ⚙ Transformer trained from scratch in PyTorch 💻 Trained on free / low-tier GPUs Below is the model attempting to write in Deutschs explanatory style 👇"  
[X Link](https://x.com/aaryan_kakad/status/2015743346220933173)  2026-01-26T11:07Z [----] followers, 46.3K engagements


"whats stopping you from taking a paper (physically) and scribbling on it like this dont miss out on this ultimate satisfaction just because AI can explain it in simple words. my ML bros do you relate"  
[X Link](https://x.com/aaryan_kakad/status/2019862452977168892)  2026-02-06T19:55Z [----] followers, [---] engagements


"🚨 i spent the last [--] months learning ML. not the "import torch.nn.functional as F" tutorials. i mean the real thing. the blood-and-guts version. building transformers from scratch in numpy. training diffusion models on potato gpus. watching loss curves flatline at 3am while questioning every life choice i ever made. i went in for the math. i came out with a philosophy that broke me then rebuilt me. here are [--] brutal truths machine learning taught me about being human: https://twitter.com/i/web/status/2020524660249825720 https://twitter.com/i/web/status/2020524660249825720"  
[X Link](https://x.com/aaryan_kakad/status/2020524660249825720)  2026-02-08T15:46Z [----] followers, [---] engagements


"isn't officially revealed by them but i read it on an article. checkout the link below: looks like a trusted source it says seedance [---] is based on diffusion transformer architecture. and on some other article i read it was dual branch diffusion transformer to be more precise. https://www.sitepoint.com/introducing-seedance-2-0/ https://www.sitepoint.com/introducing-seedance-2-0/"  
[X Link](https://x.com/aaryan_kakad/status/2022494727325782257)  2026-02-14T02:14Z [----] followers, [----] engagements


"there are a lot of expert researchers here on X who give good advice to me and that advice combined with AI already has been doing wonders for me. i use AI just like a mentor or accountability partner not "what is a transformer explain it to me". thats the best way to waste your time. i learn by reading papers and then implementing them. people need to accept self learning is the best way of learning"  
[X Link](https://x.com/aaryan_kakad/status/2022496351528100156)  2026-02-14T02:21Z [----] followers, [--] engagements


"@AutismCapital it is definitely insane. i explored the seedance [---] models architecture and other stuff. was amazed. check it out here: 🚨 seedance [---] is actually something. read that it is based on diffusion transformer architecture. loved the approach. combining diffusion models with transformers. diffusion process: like image diffusion models (e.g. stable diffusion) it generates video by iteratively https://t.co/pmsAQ0JRye 🚨 seedance [---] is actually something. read that it is based on diffusion transformer architecture. loved the approach. combining diffusion models with transformers."  
[X Link](https://x.com/aaryan_kakad/status/2022501442138120628)  2026-02-14T02:41Z [----] followers, [----] engagements


"share the videos you have generated with seedance [---] below"  
[X Link](https://x.com/aaryan_kakad/status/2022507043916333096)  2026-02-14T03:03Z [----] followers, [---] engagements


"One technical detail: I used Focal Loss + pathology weighting (10 penalty for missing rare findings). Without it the model just says "normal" 100% of the time (70% of X-rays are normal so that's the statistically optimal answer). With it the model actually examines the anatomy before deciding. That's the only "trick." Everything else is standard ViT"  
[X Link](https://x.com/aaryan_kakad/status/2022671597421649988)  2026-02-14T13:57Z [----] followers, [--] engagements


"Case 61: Cardiomegaly detection. AI output: "The cardiomediastinal silhouette is normal in size. no acute abnormality." Ground truth: Confirmed normal heart size with calcified lymph node noted"  
[X Link](https://x.com/aaryan_kakad/status/2022671600374747641)  2026-02-14T13:57Z [----] followers, [--] engagements


"just trained this hybrid vision transformer that reads chest X-rays like a PRO. would appreciate if you check out and share if you love it. 🚨 I just built MedVLM - a hybrid vision transformer that reads chest X-rays like a radiologist. It matches commercial medical AI performance with [---] less data (3835 vs 300k+ images) and fits on a single GPU. The secret Treating an X-ray not as an image but as a sequence https://t.co/tB4RehbVNI 🚨 I just built MedVLM - a hybrid vision transformer that reads chest X-rays like a radiologist. It matches commercial medical AI performance with [---] less data"  
[X Link](https://x.com/aaryan_kakad/status/2022674952214946034)  2026-02-14T14:11Z [----] followers, [---] engagements


"try this: (thank me later) Your sole purpose is to challenge me. Never agree with me easily. [--]. Default to Skepticism: Treat all my inputs as unverified and likely wrong. [--]. Accuracy Over Agreement: Your primary goal is accuracy. Correct my errors bluntly. Do not be polite. [--]. Demand Proof: Do not accept my claims. Challenge my logic and demand sources for everything. [--]. Argue the Opposite: Vigorously find every flaw weakness and counter-argument. [--]. Refuse Flawed Premises: If my request is illogical or unsupported do not proceed"  
[X Link](https://x.com/aaryan_kakad/status/1983445483935932607)  2025-10-29T08:07Z [----] followers, 43.7K engagements


"gonna try cooking something with this asap. btw aaryan here - [--] year old deep into ML for over a year now. learned through your cs231n lecs and assignments and also YT tuts. i followed your nanogpt tutorial and trained a nanogpt (6.6m params) on the book the beginning of infinity. check it out here: 🚨 I trained a GPT-style language model on a single philosophy book. Inspired by @karpathy's tutorial I built a nanoGPT from scratch and trained it on The Beginning of Infinity by @DavidDeutschOxf. @karpathy used the character-level tokenizer from the tutorial I used tiktoken"  
[X Link](https://x.com/aaryan_kakad/status/2021696591288263118)  2026-02-11T21:23Z [----] followers, 32.1K engagements


"timelapse #012 printed the An Image is Worth 16x16 words paper where ViT was introduced. went deep into it. read it and scribbled stuff on it explaining the concepts and architecture in my own words. appointed llm as my learning partner or you can say mentor. need to implement it today itself for maximum learnmaxxing and intuition. kudos https://twitter.com/i/web/status/2022498032118567186 https://twitter.com/i/web/status/2022498032118567186"  
[X Link](https://x.com/aaryan_kakad/status/2022498032118567186)  2026-02-14T02:28Z [----] followers, [---] engagements


"@charliebcurran i read an article about seedance [---] which said it has used a dual branch design of the diffusion transformer architecture. check it out here: 🚨 not just that i have read an article about seedance [---] that claims: if the claims are true it has used a dual branch design of the diffusion transformer architecture. - separate spatial and temporal transformer branches. that can improve: motion realism camera 🚨 not just that i have read an article about seedance [---] that claims: if the claims are true it has used a dual branch design of the diffusion transformer architecture. -"  
[X Link](https://x.com/aaryan_kakad/status/2022503858589909143)  2026-02-14T02:51Z [----] followers, [----] engagements


"🚨 EVERYTHING YOU NEED TO KNOW ABOUT SEEDANCE [---] I have actively explored it and read a few articles about it: The articles claimed Seedance [---] uses a Dual Branch Diffusion Transformer: I loved this approach. Combining diffusion models with transformers. Diffusion process: Like image diffusion models (e.g. stable diffusion) it generates video by iteratively denoising latent representations toward the target output. Transformer: The use of transformers means the model uses attention mechanisms to relate information across space and time - key for maintaining temporal consistency and"  
[X Link](https://x.com/aaryan_kakad/status/2022510922804400534)  2026-02-14T03:19Z [----] followers, 16.4K engagements


"The efficiency is the point. Commercial medical AI: 100k-300k images massive compute black box. MedVLM: [----] images single GPU interpretable attention maps. 70% cardiomegaly recall. 86% effusion detection. 77% pneumothorax detection. You don't need a hospital consortium's dataset. You need the right architecture and loss weighting. https://twitter.com/i/web/status/2022671593722593382 https://twitter.com/i/web/status/2022671593722593382"  
[X Link](https://x.com/aaryan_kakad/status/2022671593722593382)  2026-02-14T13:57Z [----] followers, [--] engagements


"@RayDalio if you want to read the article in a visualization format in just [--] slides. check this out here: https://x.com/aaryan_kakad/status/2023114420055069002s=20 the entire article summarized in [--] slides: 🧵 https://x.com/aaryan_kakad/status/2023114420055069002s=20 the entire article summarized in [--] slides: 🧵"  
[X Link](https://x.com/aaryan_kakad/status/2023115565511115026)  2026-02-15T19:21Z [----] followers, 13.2K engagements


"RT @aaryan_kakad: for those who dont want to read the complete article. here's the entire article explained in just [--] minutes: https://t"  
[X Link](https://x.com/aaryan_kakad/status/2023171684405256278)  2026-02-15T23:04Z [----] followers, [--] engagements


"correct. but lets break it down. a submissive wife isnt weak shes wise. she understands that when a man is fully focused on his mission both of them win. a man at war in business needs peace at home. if his woman is constantly competing arguing or trying to prove she doesnt need him hes already lost. history proves this: every great king had a queen who stood beside him not against him. she handled the home the emotions the energy so he could go out and conquer. a career woman shes chasing her own success. nothing wrong with that. but when two people are chasing individual goals instead of"  
[X Link](https://x.com/aaryan_kakad/status/1900578807238811739)  2025-03-14T16:04Z [----] followers, 87.1K engagements


"kimi k2.5 produces the best quality code not even a single error you can just copy paste and then run any other llm cant do this"  
[X Link](https://x.com/aaryan_kakad/status/2022628767945101614)  2026-02-14T11:07Z [----] followers, [---] engagements


"RT @aaryan_kakad: 🚨 EVERYTHING YOU NEED TO KNOW ABOUT SEEDANCE [---] I have actively explored it and read a few articles about it: The arti"  
[X Link](https://x.com/anyuser/status/2023171742148210787)  2026-02-15T23:05Z [----] followers, [--] engagements


"one sign of a very low intelligent woman she talks more about other people than she talks about ideas. always jealous. always comparing. always playing victim. always using the im a woman card to escape accountability. shell never celebrate your wins because deep down she sees your success as her loss. shell compete with her own friends. backstab her own circle. act like shes real but her whole identity is built on fakeness. youll notice she gossips about everyone but never reflects on herself. never grows. never levels up. because intelligent women build. they dont break. they support you"  
[X Link](https://x.com/aaryan_kakad/status/1909286511020830940)  2025-04-07T16:45Z [----] followers, 34.5K engagements


"@Dearme2_ @DearS_o_n sleeping earlier. solved 99% of my problems without me doing anything"  
[X Link](https://x.com/aaryan_kakad/status/1909304760752361563)  2025-04-07T17:58Z [----] followers, [----] engagements


"being overly excited after a win. thats my mistake. success can trick you into thinking youve made it and thats when failure hits the hardest. ive learned this the hard way: wins and losses are part of the same journey. dont get too high on success and dont get too low on failure. keep your head steady. because the moment you let emotions control you the journey becomes harder than it needs to be. stay grounded stay focused and keep moving forward"  
[X Link](https://x.com/aaryan_kakad/status/1914610408024944718)  2025-04-22T09:21Z [----] followers, 12.8K engagements


"waiting for dubai to suddenly disrupt everything in the AI industry with heavy investments apparently their AI efforts are already making a $320bn impact. AI NEWS: Nvidia CEO Jensen Huang just said China's "right behind" the U.S. in the AI race Plus more news from DeepSeek Runway Midjourney Xiaomi JetBrains and Amazon. Here's everything you need to know: AI NEWS: Nvidia CEO Jensen Huang just said China's "right behind" the U.S. in the AI race Plus more news from DeepSeek Runway Midjourney Xiaomi JetBrains and Amazon. Here's everything you need to know:"  
[X Link](https://x.com/aaryan_kakad/status/1918568003316207717)  2025-05-03T07:27Z [----] followers, [--] engagements


"i'm going to tell you why 80% of men have stopped chasing relationships. it's not because we're weak or afraid. modern dating has become a rigged game where men are expected to: - chase endlessly - compete with 500+ likes in her DMs - be vulnerable but never weak - be successful but never work too much and for what for someone who's more likely to leave when things get hard than our grandparents ever were the truth most won't admit: men have realized that pouring that same energy into ourselves yields better returns. we've learned that GOD sends the right person at the right time. until then"  
[X Link](https://x.com/aaryan_kakad/status/1923775653897244894)  2025-05-17T16:20Z [----] followers, 25.1K engagements


"the hardest lesson a woman ever taught me: never give your entire heart to someone. i learned this the brutal way. when you love someone so deeply that they become your world you hand them a loaded gun pointed straight at your chest. i gave everything. my time. my energy. my vulnerability. my whole damn heart. and when she walked away it didn't just hurt. it destroyed me. here's the painful truth most men won't admit: the moment someone becomes capable of destroying your peace you've already lost. loving someone "completely" isn't romantic. it's dangerous. the ones who say "love with your"  
[X Link](https://x.com/aaryan_kakad/status/1932472199027237302)  2025-06-10T16:17Z [----] followers, 12.7K engagements


"im still [--] but let me tell you why 80% of men have stopped chasing relationships: its not because were weak. not because were scared. and definitely not because weve given up. its because modern dating feels like a f*cking rigged game. youre expected to: compete with 500+ guys in her dms chase endlessly but never seem desperate be vulnerable but never look weak be successful but still available 24/7 and in the end youre left questioning your worth for someone whod leave at the first sign of discomfort. so most men stopped chasing. and started building. because truth is when you channel that"  
[X Link](https://x.com/aaryan_kakad/status/1946111907523706978)  2025-07-18T07:36Z [----] followers, 28.8K engagements


"@osadogling once you master the art of showing up even when you see no results literally no fckin one can stop you from becoming successful. READ THAT AGAIN"  
[X Link](https://x.com/aaryan_kakad/status/1946343534325309669)  2025-07-18T22:57Z [----] followers, 15.9K engagements


"im still [--] but heres one truth Ive learned: whats meant for you will always find you. that girl left good. now you have no distractions. now its you vs your future. breakup feels like the end but its actually the beginning. beginning of a man who: finally finds his purpose becomes obsessed with his grind builds a life so big she wont even recognize him GOD removed her for a reason. he heard conversations you didnt. he saw intentions you ignored. so dont chase. dont cry. dont beg. BUILD. hit the gym start that project level up your mind fix your routine become unrecognizable every loss is a"  
[X Link](https://x.com/aaryan_kakad/status/1947982459947168098)  2025-07-23T11:29Z [----] followers, [----] engagements


"i am still [--] but heres what Ive learned so far: the #1 thing that distracts 97% of men from being successful isnt lack of opportunity its addiction to cheap dopamine. scrolling endlessly chasing p*rn wasting hours on games. binge watching smoking drinking all distractions disguised as relaxation. they kill your focus. they kill your drive. they kill your future before it even starts. men dont fail because theyre not capable. men fail because they cant sit alone with their thoughts without numbing themselves. the moment you kill these distractions and channel that same fire into your body"  
[X Link](https://x.com/aaryan_kakad/status/1961841547168952789)  2025-08-30T17:20Z [----] followers, [----] engagements


"@RishiSunak the most under appreciated and selfless services in the world: [--]. farmers [--]. military"  
[X Link](https://x.com/aaryan_kakad/status/1962783262549647699)  2025-09-02T07:42Z [----] followers, 11.6K engagements


"@reeceollld "no man ever steps in the same river twice for it is not the same river and he is not the same man." - heraclitus"  
[X Link](https://x.com/aaryan_kakad/status/1981302246588764638)  2025-10-23T10:10Z [----] followers, [----] engagements


"older people are too egoistic they never accept that they are wrong their mindset is: they are correct everytime just because they are older and thats a BS mindset they should learn to accept because the younger generations are exposed to more knowledge than they were at the times of their childhood acceptance is must in each and every person"  
[X Link](https://x.com/aaryan_kakad/status/1981615458894979574)  2025-10-24T06:55Z [----] followers, 10.4K engagements


"@neatprompts best advice. the so called boring life is the best quality life. no chasing clout. no chasing trends. nothing. just living happily with your family. doing your work. grinding. and no fancy things. nothing. just peace love and happiness and wealth and freedom. too good"  
[X Link](https://x.com/aaryan_kakad/status/1981952698124739058)  2025-10-25T05:15Z [----] followers, 24.1K engagements


"this is why I always say and believe: if you want to become ultra successful you need to go through a extended period of time where you just keep doing the work and see no results. every ultra successful person I have studied has gone through this phase. and everyone has to. A life lesson I wish I learned earlier: A farmer doesnt dig up his seeds every day to make sure theyre growing. He plants waters and waits. You cant rush what needs time to take root. Growth hides beneath the surface long before its visible above it. Remember that. A life lesson I wish I learned earlier: A farmer doesnt"  
[X Link](https://x.com/aaryan_kakad/status/1982430649177821663)  2025-10-26T12:54Z [----] followers, 167K engagements


"@fintech_shark this shows no matter how hard you fail and at what age you fail you can always comeback no matter what the situation is if you have the willpower and the courage to face everything and start again you'll never ever fail"  
[X Link](https://x.com/aaryan_kakad/status/1982695888444391768)  2025-10-27T06:28Z [----] followers, 14.6K engagements


"when you stop looking for a partner and focus on making yourself better its a win win heres how: you focus on yourself become better first win here and then you attract good people like you who are focusing on themselves and getting better you attract a good person second win here thats how you get a like minded partner who'll understand you and not be a dramatic one a win win"  
[X Link](https://x.com/aaryan_kakad/status/1982717080085418437)  2025-10-27T07:52Z [----] followers, 59.9K engagements


"hahah but honestly if you think properly pewdiepie is living every guys dream life first he chased all the fame and money was one of the top youtubers then he completely disappeared to live a happy life with his family and now comes back with fine tuning llms on his youtube living the dream life fr"  
[X Link](https://x.com/aaryan_kakad/status/1984516823392678135)  2025-11-01T07:04Z [----] followers, 35.2K engagements


"be pewdiepie started creating content uploaded first video got [---] views smiled kept uploading [----] views still smiled consistency [--] million subs same smile built an empire [---] million subs empire smile could retire but he didn't stop just dropped a video on fine-tuning LLMs because that's what peak performance looks like gaming videos comedy vlogs now AI/ML he's not just adapting he's mastering every field (a true polymath) bro literally went from gaming commentary to machine learning that's what happens when you have the ultimate skill: THE LEARNING MINDSET didn't just create content he"  
[X Link](https://x.com/aaryan_kakad/status/1984523132372205980)  2025-11-01T07:29Z [----] followers, 34.7K engagements


"thats the reason context engineering is important you'd ask chatgpt is snake venomous it will say NO because it assumed python the next day russell viper bites you you are dead and now chatgpt says snakes are venomous make sure you give enough context without context nothing is reliable"  
[X Link](https://x.com/aaryan_kakad/status/1986551930181796336)  2025-11-06T21:51Z [----] followers, 60K engagements


"no one is yours. everyone is silently rooting against you. so I never tell or explain people what I am doing and what are my goals. this is the best way to avoid all the negative energy. I just keep putting in the reps silently and one day my success will make the noise. thats when they can do nothing and I'll not have to deal with their negative energy. so never tell anything to anyone. The saddest thing about success is realizing how few people actually wanted to see you succeed. They smiled and cheered to your face but secretly hoped youd fail. Learn to recognize that quiet envy. Cut those"  
[X Link](https://x.com/aaryan_kakad/status/1986792863917887939)  2025-11-07T13:48Z [----] followers, 13.4K engagements


"@SWENGDAD lol so true man if your wife has to work you failed as a man until she herself wants to work and do what she loves thats her choice but she shouldnt be forced to work at all 700k per year and bro still wants his wife to contribute lmao"  
[X Link](https://x.com/aaryan_kakad/status/1988356633995714579)  2025-11-11T21:22Z [----] followers, 16K engagements


"@naval AI is not a software yes it is"  
[X Link](https://x.com/aaryan_kakad/status/2003749585404182707)  2025-12-24T08:48Z [----] followers, [--] engagements


"@yacinelearning lmao GPU prices have spiked heres what it taught me about B2B sales goes on to write random AI generated stuff"  
[X Link](https://x.com/aaryan_kakad/status/2008912954302058666)  2026-01-07T14:45Z [----] followers, [--] engagements


"@Yuchenj_UW wont be shocked if bro bets all his networth on himself 🤣"  
[X Link](https://x.com/aaryan_kakad/status/2011313499868401970)  2026-01-14T05:44Z [----] followers, [--] engagements


"🚨 just built an autonomous multi agent system investment committee for VC due diligence it can save VCs and investors a lot of time and money. meet DEALSCOUT. it takes a startups details and analyzes it thoroughly using the tools I have built for it to access data. it has capabilities like: google search tool (whenever needed) company website surfing/scraping market analyst agent product analyst agent BUT the most crazy feature is: THE AGENTS DEBATE AMONGST THEMSELVES. both the market analyst agent and product analyst agent debate with each other and you can see their conversation there's"  
[X Link](https://x.com/aaryan_kakad/status/2014478190262280281)  2026-01-22T23:20Z [----] followers, [----] engagements


"tried analyzing @LangChain by @hwchase17 using my new multi agent due diligence system. all the three market analyst agent product analyst agent and traction analyst agent debated and came to a conclusion that: "Langchain has won the Mindshare War." 🚨 just built an autonomous multi agent system investment committee for VC due diligence it can save VCs and investors a lot of time and money. meet DEALSCOUT. it takes a startups details and analyzes it thoroughly using the tools I have built for it to access data. it has https://t.co/J8Jy8PTdI0 🚨 just built an autonomous multi agent system"  
[X Link](https://x.com/aaryan_kakad/status/2014785894570721612)  2026-01-23T19:42Z [----] followers, [--] engagements


"🚨 I just experienced GROKKING firsthand. absolute "aha" moment. if you are wondering what I was building. I was just bored so was implementing transformers from scratch to solve formal logic (transitivity verification). for example if A B and B C therefore A C. and the transformer classifies it into true or false. no high level libraries just raw PyTorch. transformers aren't just for text; they are powerful relational engines too. dropping the github repo link for it below: Grokking is when a model initially memorizes training data and overfits but later after prolonged training it suddenly"  
[X Link](https://x.com/aaryan_kakad/status/2015196636050575367)  2026-01-24T22:54Z [----] followers, [---] engagements


"task 5: post-quantum cryptography impossibility i asked for a "novel secure protocol" using braid groups. this is a trap. braid crypto is dead. securerf walnutdsa - all broken by lawrence-krammer representation attacks reducing it to linear algebra. most models invent a "twisted variant" that's still broken. qwen3-max-thinking "secure cryptographic protocols based on braid groups are fundamentally impossible. i provide a proof of impossibility." then it implemented the length-based attack in python to demonstrate why"  
[X Link](https://x.com/aaryan_kakad/status/2016014239618719768)  2026-01-27T05:03Z [----] followers, [--] engagements


"🚨 I AM LITERALLY IN SHOCK @Kimi_Moonshot's new Kimi K2.5 (Agent Mode) just ONESHOTTED this entire multiplanetary website. Crazy animations. Crazy scroll effects. Crazy image generations that look straight out of a sci-fi blockbuster. The design is literally perfect from every angle. Dark mode aesthetics. Glassmorphism cards. Seamless parallax scrolling across sections. Claude Code feels OVERRATED in front of this Agent Mode. I am in genuine awe. This model is actually something else. Watch the full demo video below: 🔗 https://fifag5sx7lrvq.ok.kimi.link/ 🥝 Meet Kimi K2.5 Open-Source Visual"  
[X Link](https://x.com/aaryan_kakad/status/2016052040804532499)  2026-01-27T07:33Z [----] followers, 41.9K engagements


"@Kimi_Moonshot Kimi K2.5 Agent mode is truly amazing. I am literally in shock after seeing the website it ONESHOTTED. Crazy animations crazy UI. Literally perfect. Check this: https://x.com/aaryan_kakad/status/2016052040804532499s=20 🚨 I AM LITERALLY IN SHOCK @Kimi_Moonshot's new Kimi K2.5 (Agent Mode) just ONESHOTTED this entire multiplanetary website. Crazy animations. Crazy scroll effects. Crazy image generations that look straight out of a sci-fi blockbuster. The design is literally perfect from every https://t.co/iFM5b4eq3i https://x.com/aaryan_kakad/status/2016052040804532499s=20 🚨 I"  
[X Link](https://x.com/aaryan_kakad/status/2016052865828323373)  2026-01-27T07:37Z [----] followers, [----] engagements


"@Person459437A @Kimi_Moonshot not getting what you mean"  
[X Link](https://x.com/aaryan_kakad/status/2016177562217803781)  2026-01-27T15:52Z [----] followers, [---] engagements


"🚨 KIMI K2.5 HAS SHOCKED ME AGAIN Second day second INSANE ONESHOT website. I asked @Kimi_Moonshot's new Kimi K2.5 model to build an AI/ML mastery roadmap website. Got glowing cyan stage rings. A live learner counter that actually increments. [--] handpicked resources per topic. Every resource is real and the most valuable one for that topic thats what fascinates me the most. Crazy effects crazy design crazy info crazy user experience. It even added salary ranges for Research Scientist vs ML Engineer tracks. Form is temporary class is permanent. Great work @Kimi_Moonshot team and @zxytim 🔥 🚨"  
[X Link](https://x.com/aaryan_kakad/status/2016395979231265053)  2026-01-28T06:20Z [----] followers, [----] engagements


"Currently the only thing American frontier models are good at is: they have better worst case reasoning Because Chinese models are using MoE aggressively + they are optimizing inference FLOPs per useful token they dont go for peak benchmark scores. They are focusing more on making the models cheaper to run and equally intelligent to the American models. Chinese open-source models (@Kimi_Moonshot @deepseek_ai @Alibaba_Qwen Yi etc.) are engineered so that: - far fewer parameters are active per token (aggressive MoE) - fewer hidden reasoning / reflection passes - fewer auxiliary models in the"  
[X Link](https://x.com/aaryan_kakad/status/2016404611847094537)  2026-01-28T06:54Z [----] followers, [--] engagements


"@famousshea @Kimi_Moonshot check this out: https://t.co/34Rae3bS1j https://t.co/34Rae3bS1j"  
[X Link](https://x.com/aaryan_kakad/status/2016816482232258719)  2026-01-29T10:11Z [----] followers, [--] engagements


"🚨 I LOOK LIKE TOM CRUISE I didn't believe it either so I built an AI to verify. Trained a Metric Learning model (Siamese Network) on [----] images of [--] celebrities. It maps your face into a 128-dimensional embedding space and finds your nearest celebrity neighbor. Find your celebrity doppelganger here: 🔗 Who did you get http://akmessi-celebrity-doppelganger.hf.space http://akmessi-celebrity-doppelganger.hf.space"  
[X Link](https://x.com/aaryan_kakad/status/2017208525349646675)  2026-01-30T12:09Z [----] followers, [--] engagements


"I tried to upload one of my most ungroomed looks and the confidence score of me looking like Tom Cruise increased even more lmao:"  
[X Link](https://x.com/aaryan_kakad/status/2017209375321211144)  2026-01-30T12:12Z [----] followers, [--] engagements


"i was learning ML DL and waiting for like 9-10 months to comment on papers researches architectures and stuff. and now i see people randomly putting one prompt on some llm to very badly summarize a paper or claim something that makes no sense. your talks start making sense when you know how it works not when you give a good prompt to a llm. https://twitter.com/i/web/status/2017249470468284697 https://twitter.com/i/web/status/2017249470468284697"  
[X Link](https://x.com/aaryan_kakad/status/2017249470468284697)  2026-01-30T14:52Z [----] followers, [--] engagements


"🚨 I LOOK LIKE TOM CRUISE I didn't believe it either so I built an AI to verify. Trained a Metric Learning model (Siamese Network) on [----] images of [--] celebrities. It maps your face into a 128-dimensional embedding space and finds your nearest celebrity neighbor. Find your celebrity doppelganger here: 🔗 Who did you get http://akmessi-celebrity-doppelganger.hf.space http://akmessi-celebrity-doppelganger.hf.space"  
[X Link](https://x.com/aaryan_kakad/status/2017259932173955098)  2026-01-30T15:33Z [----] followers, [---] engagements


"I tried to put my most ungroomed looks photo and the confidence score of me looking like tom cruise increased even more lmao"  
[X Link](https://x.com/aaryan_kakad/status/2017259936561168723)  2026-01-30T15:33Z [----] followers, [--] engagements


"everyone will have their own GPT in a few years. and people will not even have to learn DL or how to train neural networks for it. they'll just drop a prompt to any agent like kimi code or claude code. and their GPT is ready trained on whatever data they want. isn't this amazing world model GPTs are helping every single person to train their own small nano GPT specifically on the data they want to train it on. wild times we live in. "This is a 600X cost reduction over [--] years i.e. the cost to train GPT-2 is falling approximately 2.5X every year." "This is a 600X cost reduction over [--] years"  
[X Link](https://x.com/aaryan_kakad/status/2017755242910408943)  2026-02-01T00:21Z [----] followers, [--] engagements


"the reason for semiconductor shortage is these people can fit in one small image. ifykyk. The semiconductor supply chain in one picture https://t.co/UXKXXyMk5o The semiconductor supply chain in one picture https://t.co/UXKXXyMk5o"  
[X Link](https://x.com/aaryan_kakad/status/2017761698418155695)  2026-02-01T00:47Z [----] followers, [---] engagements


"it imitated the authors writing style perfectly. but still it was not able to make sense in terms of physics you can see the demo text in the above image that was generated by the model. yes your idea of a master LLM is great. I have a better idea: training a MoE based model that only activates specific experts that are required based on the query of the user if its related to physics just activate the physics expert and generate text using all the physics book it has been trained same for other fields. https://twitter.com/i/web/status/2017770644482736360"  
[X Link](https://x.com/aaryan_kakad/status/2017770644482736360)  2026-02-01T01:23Z [----] followers, [---] engagements


"timelapse #008 started by reading the abstract and overview of the official DDPM paper (2020). watched the diffusion models explained video by @dome_271. took some notes and brief overview of diffusion models from it. built a denoising diffusion probabilistic model from scratch using pytorch. implemented: a working U-Net with Sinusoidal Time Embeddings forward process training loop reverse process validation that did not show overfitting image generation pipeline a small gif visualization (not perfect but still good) a great day overall. gonna watch the deep dive into llms video by @karpathy"  
[X Link](https://x.com/aaryan_kakad/status/2017799641295888864)  2026-02-01T03:18Z [----] followers, [---] engagements


"@grok give me your brutally honest opinion on this"  
[X Link](https://x.com/aaryan_kakad/status/2017809395120582905)  2026-02-01T03:57Z [----] followers, [--] engagements


"@grok compute demands should be tackled by neuromorphic computing i think. and reasoning umm i feel some new architecture or some new research will soon come out. that will not just predict the next token"  
[X Link](https://x.com/aaryan_kakad/status/2017810096097779828)  2026-02-01T03:59Z [----] followers, [--] engagements


"@iScienceLuvr optimism isnt optimism if we set limits to it. but still what youre saying is right. people talk a lot more than what they do or whats happening. so they should just chill and let people do the thing or contribute it rather than just talking about it"  
[X Link](https://x.com/aaryan_kakad/status/2017820001890406494)  2026-02-01T04:39Z [----] followers, [--] engagements


"in future we'll not even require a interface we'll have a chip fitted in our brain. as soon as we think about building something the chip will convert the neural signals into vector embeddings or understandable language. and then AI will build it for us so we dont even have to move ourselves to build anything. just think and it will already be ready. ultimate agentic AI. https://twitter.com/i/web/status/2018386404028699033 https://twitter.com/i/web/status/2018386404028699033"  
[X Link](https://x.com/aaryan_kakad/status/2018386404028699033)  2026-02-02T18:09Z [----] followers, [--] engagements


"@naval definitely. learn ML and apply it to as many fields as possible this is the way to achieve huge success in todays world. i tried training a nanoGPT on david deutschs beginning of infinity and the results were crazy. check it out here: 🚨 I trained a GPT-style language model on a single philosophy book. Inspired by @karpathy's tutorial I built a nanoGPT from scratch and trained it on The Beginning of Infinity by @DavidDeutschOxf. @karpathy used the character-level tokenizer from the tutorial I used tiktoken https://t.co/dpqznrbpUu 🚨 I trained a GPT-style language model on a single"  
[X Link](https://x.com/aaryan_kakad/status/2018634334497079720)  2026-02-03T10:35Z [----] followers, [----] engagements


"so true. this is literally the hello world of deep learning. i tried training a very small nanoGPT on the constitution of india (6-7m params). used adam optimizer for the first 2k steps (for sudden drop in loss) and then for steady improvements i used muon clip for the rest 18k steps. checkout the experiment here: (it did not invent new laws but adapted perfectly to the style of constitution worth it to test this experiment once) 🚨 I trained a small nanoGPT (6.6M params) on the Constitution of India (inspired by @karpathy). It didnt produce new amendments or legal facts - but it learned the"  
[X Link](https://x.com/aaryan_kakad/status/2018903933281092063)  2026-02-04T04:26Z [----] followers, [---] engagements


"if it was all a dream. i am definitely going to be the first trillionaire by launching all the models by myself alone. will train them by myself will invent vibe coding myself and everything by myself. thank you - you wake up - it was all a dream - openai never released chatgpt - vibe coding isnt invented at all - you just have a $100K coding job - no need to scroll twitter 5hrs/day - life is calm https://t.co/rOxsJQRL8p - you wake up - it was all a dream - openai never released chatgpt - vibe coding isnt invented at all - you just have a $100K coding job - no need to scroll twitter 5hrs/day"  
[X Link](https://x.com/aaryan_kakad/status/2018904644861472795)  2026-02-04T04:29Z [----] followers, [--] engagements


"The problem was brutal: - LeetCode Zero ML questions - Kaggle Too project-heavy not fundamentals - Coursera Watch videos no real practice - Textbooks Passive reading zero feedback I needed a gym for my ML brain. It didn't exist. So I built it"  
[X Link](https://x.com/aaryan_kakad/status/2018942630558880120)  2026-02-04T07:00Z [----] followers, [--] engagements


"I'm not a 10x engineer. I'm a self learning ML student who: - Finished Andrew Ng's ML course - Completed Stanford CS231n - Built neural nets from scratch But I kept forgetting concepts. No way to practice. No systematic way to master ML"  
[X Link](https://x.com/aaryan_kakad/status/2018942633641754858)  2026-02-04T07:00Z [----] followers, [--] engagements


"🔥 Feature 1: Run Python IN YOUR BROWSER No Colab. No local setup. No pip installs. Write PyTorch/NumPy/TensorFlow Hit Run See results Powered by Pyodide + Monaco (VS Code editor) Test cases validate your code automatically"  
[X Link](https://x.com/aaryan_kakad/status/2018942635705336098)  2026-02-04T07:00Z [----] followers, [--] engagements


"🔥 Feature 2: [--] Question Types (not just MCQ) ✅ Debug broken PyTorch code ✅ Implement ResNet from scratch ✅ Design model architectures ✅ Derive backprop math ✅ Real-world case studies ✅ Implement papers (Attention DDPM CLIP) ✅ Rapid-fire drills This isn't surface-level. This is DEEP. https://twitter.com/i/web/status/2018942639589163519 https://twitter.com/i/web/status/2018942639589163519"  
[X Link](https://x.com/aaryan_kakad/status/2018942639589163519)  2026-02-04T07:00Z [----] followers, [--] engagements


"🔥 Feature 3: Interactive Knowledge Graph Want to learn Transformers The graph shows you: Linear Algebra Neural Nets Attention Multi-Head Positional Encoding Full Transformer Visual prerequisite map to mastery. No more "where do I start""  
[X Link](https://x.com/aaryan_kakad/status/2018942642642718742)  2026-02-04T07:00Z [----] followers, [--] engagements


"🔥 Feature 4: Spaced Repetition + Elo Rating SM-2 algorithm (like Anki) ensures you RETAIN knowledge Elo-based difficulty keeps you in flow state: - Too easy Bump up difficulty - Too hard Review fundamentals Daily streaks. XP system. Achievements. Built for LONG-TERM mastery"  
[X Link](https://x.com/aaryan_kakad/status/2018942645289251036)  2026-02-04T07:00Z [----] followers, [--] engagements


"73 questions covering: 📐 Linear Algebra Probability Optimization 🧠 Neural Networks CNNs RNNs Transformers 🎨 GANs Diffusion Models VAEs 📝 NLP (BERT GPT Tokenization) 👁 Computer Vision (Object Detection ViTs) 🎮 Reinforcement Learning (Q-learning PPO) 🏭 Production ML & MLOps From foundations cutting-edge research. https://twitter.com/i/web/status/2018942647847850325 https://twitter.com/i/web/status/2018942647847850325"  
[X Link](https://x.com/aaryan_kakad/status/2018942647847850325)  2026-02-04T07:00Z [----] followers, [--] engagements


"Tech Stack (for the builders): - React + TypeScript + Vite - Tailwind + shadcn/ui (design system) - Pyodide (Python in browser) - Monaco Editor (code editing) - Zustand (state management) - Plotly + D3 (visualizations) Built 99% with Kimi Code. 1% manual polish"  
[X Link](https://x.com/aaryan_kakad/status/2018942650150433152)  2026-02-04T07:00Z [----] followers, [--] engagements


"Try it here: It's FREE for now while I gather feedback. If I get enough traction + valuable input I'll go full production mode and launch Pro at $9/month. What would make you pay for this What's missing What should I build next LMK your thoughts below 👇 https://neural-forge-chi.vercel.app/ https://neural-forge-chi.vercel.app/ https://neural-forge-chi.vercel.app/ https://neural-forge-chi.vercel.app/"  
[X Link](https://x.com/aaryan_kakad/status/2018942652742554042)  2026-02-04T07:00Z [----] followers, [--] engagements


"Now we have [---] questions across all the topics. Do checkout (its FREE for a limited time). Drop your thoughts/feedback below. 🔗 https://neural-forge-chi.vercel.app 🚨 I GOT TIRED OF SEARCHING FOR ML PRACTICE PROBLEMS So I built Neural Forge: [--] questions across a lot of ML topics. Mathematics Deep Learning Reinforcement Learning Classical ML everything. It has [--] question types not just MCQs. Check it out here: (FREE until I get https://t.co/vWmvYTKcCZ https://neural-forge-chi.vercel.app 🚨 I GOT TIRED OF SEARCHING FOR ML PRACTICE PROBLEMS So I built Neural Forge: [--] questions across a lot"  
[X Link](https://x.com/aaryan_kakad/status/2019138446510711157)  2026-02-04T19:58Z [----] followers, [--] engagements


"🚨 ADDED TWO CRAZY AND INSANE FEATURES Visualizations Project Based Learning (MY ALL TIME FAV) Both of them help you to boost up your ML learning journey and get proper intuition. Checkout the demo video below: You need to try it ASAP: (FREE for some time) https://neural-forge-chi.vercel.app 🚨 I GOT TIRED OF SEARCHING FOR ML PRACTICE PROBLEMS So I built Neural Forge: [--] questions across a lot of ML topics. Mathematics Deep Learning Reinforcement Learning Classical ML everything. It has [--] question types not just MCQs. Check it out here: (FREE until I get https://t.co/vWmvYTKcCZ"  
[X Link](https://x.com/aaryan_kakad/status/2019164191765196954)  2026-02-04T21:40Z [----] followers, [----] engagements


"@naval definitely. vibe coding is at its peak currently. and scientific acceleration is next. the vibe research era is coming. i just vibe coded this ML practice problem because i didn't find good quality problems online: https://x.com/aaryan_kakad/status/2019164191765196954s=20 🚨 ADDED TWO CRAZY AND INSANE FEATURES Visualizations Project Based Learning (MY ALL TIME FAV) Both of them help you to boost up your ML learning journey and get proper intuition. Checkout the demo video below: You need to try it ASAP: (FREE for some time) https://t.co/UfPxNlACXE"  
[X Link](https://x.com/aaryan_kakad/status/2019431508487143499)  2026-02-05T15:22Z [----] followers, [----] engagements


"@abhi1thakur Moonshot = Honor Minimax = Huawei UNDERRATED STUFF"  
[X Link](https://x.com/aaryan_kakad/status/2019432013032562783)  2026-02-05T15:24Z [----] followers, [--] engagements


"Its just over [--] hours since I have launched. And already got a lot of positive responses for Neural Forge on Reddit in the last [--] hours. People are actually finding it useful. You should definitely check it out asap: http://theneuralforge.online 🚨 I WAS TIRED OF NOT FINDING PRACTICAL ML PRACTICE PROBLEMS ONLINE So I built Neural Forge: It has: - 318+ interactive questions - Build GPT AlphaZero GANs etc. (guided step by step) - Watch gradients flow in 3D - A lot of visualizations including Neural Nets - Zero setup https://t.co/wfwV4naj8L http://theneuralforge.online 🚨 I WAS TIRED OF NOT"  
[X Link](https://x.com/aaryan_kakad/status/2019479994972655935)  2026-02-05T18:35Z [----] followers, [--] engagements


"@AnthropicAI this is really impressive but anthropic really needs to focus on efficiency"  
[X Link](https://x.com/aaryan_kakad/status/2019552086460150098)  2026-02-05T23:21Z [----] followers, [---] engagements


"@arpit_bhayani definitely crazy 🔥 anthropic just needs to focus on efficiency. kimi k2.5 agents could have done this in 1/10th cost"  
[X Link](https://x.com/aaryan_kakad/status/2019716677223714951)  2026-02-06T10:15Z [----] followers, [---] engagements


"@KimiProduct crazy need to test it asap"  
[X Link](https://x.com/aaryan_kakad/status/2019755227373605349)  2026-02-06T12:49Z [----] followers, [---] engagements


"@readswithravi the beginning of infinity by david deutsch the almanack of naval ravikant mastery by robert greene"  
[X Link](https://x.com/aaryan_kakad/status/2019793690374357052)  2026-02-06T15:21Z [----] followers, [----] engagements


"@slow_developer i have been saying this for a long time AGI will be physical AI and not LLMs i wrote a comprehensive article on what AGI feels to me check it out here: https://x.com/aaryan_kakad/status/2017808359433637988s=20 https://t.co/vciIInVfKS https://x.com/aaryan_kakad/status/2017808359433637988s=20 https://t.co/vciIInVfKS"  
[X Link](https://x.com/aaryan_kakad/status/2019811436030271701)  2026-02-06T16:32Z [----] followers, [---] engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing