[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] #  @akshay_pachaar Akshay 🚀 Akshay 🚀 posts on X about ai, $googl, solve, llm the most. They currently have XXXXXXX followers and 1559 posts still getting attention that total XXXXXX engagements in the last XX hours. ### Engagements: XXXXXX [#](/creator/twitter::703601972/interactions)  - X Week XXXXXXX -XX% - X Month XXXXXXXXX -XX% - X Months XXXXXXXXXX +38% - X Year XXXXXXXXXX +82% ### Mentions: XX [#](/creator/twitter::703601972/posts_active)  - X Week XX -XX% - X Month XXX -XX% - X Months XXX +111% - X Year XXX +148% ### Followers: XXXXXXX [#](/creator/twitter::703601972/followers)  - X Week XXXXXXX +0.61% - X Month XXXXXXX +2.20% - X Months XXXXXXX +14% - X Year XXXXXXX +42% ### CreatorRank: XXXXXXX [#](/creator/twitter::703601972/influencer_rank)  ### Social Influence **Social category influence** [technology brands](/list/technology-brands) [stocks](/list/stocks) #4797 [finance](/list/finance) **Social topic influence** [ai](/topic/ai) #5800, [$googl](/topic/$googl) #556, [solve](/topic/solve) #227, [llm](/topic/llm) #11, [microsoft](/topic/microsoft) #117, [if you](/topic/if-you) #6033, [rag](/topic/rag) #7, [token](/topic/token), [agentic](/topic/agentic) #310, [snowflake](/topic/snowflake) #48 **Top accounts mentioned or mentioned by** [@akshaypachaar](/creator/undefined) [@avichawla](/creator/undefined) [@lightningai](/creator/undefined) [@llamaindex](/creator/undefined) [@crewaiinc](/creator/undefined) [@ollama](/creator/undefined) [@mlspring](/creator/undefined) [@firecrawldev](/creator/undefined) [@zepai](/creator/undefined) [@qdrantengine](/creator/undefined) [@cometml](/creator/undefined) [@milvusio](/creator/undefined) [@abacusai](/creator/undefined) [@assemblyai](/creator/undefined) [@tinztwins](/creator/undefined) [@streamlit](/creator/undefined) [@cometmls](/creator/undefined) [@cleanlabai](/creator/undefined) [@dailydoseofds](/creator/undefined) [@activeloopai](/creator/undefined) **Top assets mentioned** [Alphabet Inc Class A (GOOGL)](/topic/$googl) [Microsoft Corp. (MSFT)](/topic/microsoft) ### Top Social Posts Top posts by engagements in the last XX hours "I have been training neural networks for XX years now. Here are XX ways I actively use to optimize model training: (detailed explanation .🧵)" [X Link](https://x.com/akshay_pachaar/status/1996928899008417900) 2025-12-05T13:05Z 239.1K followers, 6346 engagements "So let's dive in and understand how KV caching works.👇" [X Link](https://x.com/akshay_pachaar/status/1980250256362315841) 2025-10-20T12:30Z 238.9K followers, 8567 engagements "Massive update for AI Engineers Training diffusion models just got a lot easier. dLLM is an open-source library that does for diffusion models what Hugging Face did for transformers. Here's why this matters: Traditional autoregressive models generate text left-to-right one token at a time. Diffusion models work differently - they refine the entire sequence iteratively giving you better control over generation quality and more flexible editing capabilities. The problem Building and training these models required stitching together scattered tools and reimplementing research papers from" [X Link](https://x.com/akshay_pachaar/status/1989677979069550657) 2025-11-15T12:52Z 238.9K followers, 64.1K engagements "Claude Scientific Skills. Turn Claude into your AI research assistant capable of executing complex multi-step scientific workflows across maths biology chemistry medicine and beyond. XXX% open-source (123+ skills)" [X Link](https://x.com/akshay_pachaar/status/1994391281641165300) 2025-11-28T13:01Z 238.9K followers, 57.1K engagements "Top AI Engineers never do manual prompt engineering. Today I'll show you how to automatically find the best prompts for any agentic workflow you're building. We'll use @Cometml's XXX% open-source Opik to do so. Let's go 🚀" [X Link](https://x.com/akshay_pachaar/status/1995840591947857942) 2025-12-02T13:00Z 239.1K followers, 4354 engagements "Microsoft did it again Building with AI agents almost never works on the first try. You spend days tweaking prompts adding examples hoping it gets better. Nothing systematic just guesswork. This is exactly what Microsoft's Agent Lightning solves. It's an open-source framework that trains ANY AI agent with reinforcement learning. Works with LangChain AutoGen CrewAI OpenAI SDK or plain Python. Here's how it works: Your agent runs normally with whatever framework you're using. Just add a lightweight agl.emit() helper or let the tracer auto-collect everything. Agent Lightning captures every" [X Link](https://x.com/akshay_pachaar/status/1984236460510519631) 2025-10-31T12:30Z 239K followers, 204.6K engagements "This is a goldmine of AI resources by MongoDB (free and geared towards real-world AI engineering) Building AI prototypes locally is fun. You can experiment quickly push code and try out different models with almost zero setup. Building AI for real users is where things get tricky. You need to handle storage retrieval performance security and scalable context management. MongoDBs AI resource hub solves that learning gap beautifully. It gives you a complete ecosystem of guides demos and learning tracks designed for developers who want to build production-grade AI applications with reliable data" [X Link](https://x.com/akshay_pachaar/status/1994312601757733069) 2025-11-28T07:49Z 239K followers, 12.4K engagements "Mistral just dropped their biggest release ever. And the real story isn't the benchmarks. It's what they're giving away for free. Here's what you need to know: The Release: Four new models. All Apache XXX. All open source. - Ministral 3B 8B 14B for edge devices - Mistral Large 3: 675B total parameters 41B active That last one A sparse mixture-of-experts beast trained from scratch on 3000 H200 GPUs. Why It Matters: Most companies release "open" models with asterisks everywhere. Mistral just handed you: Base models (train your own) Instruct models (deploy immediately) Reasoning variants" [X Link](https://x.com/akshay_pachaar/status/1995938201991840014) 2025-12-02T19:28Z 238.9K followers, 22.8K engagements "You're in an ML Engineer interview at Apple. The interviewer asks: "Two models are XX% accurate. - Model A is XX% confident. - Model B is XX% confident. Which one would you pick" You: "Any would work since both have same accuracy." Interview over. Here's what you missed: Modern neural networks can be misleading. They are overconfident in their predictions. For instance I saw an experiment that used the CIFAR-100 dataset to compare LeNet with ResNet. LeNet produced: - Accuracy = XXXX - Average confidence = XXXX ResNet produced: - Accuracy = XXX - Average confidence = XXX Despite being more" [X Link](https://x.com/akshay_pachaar/status/1994020936488734823) 2025-11-27T12:30Z 239K followers, 126.6K engagements "If you found it insightful reshare with your network. Find me @akshay_pachaar ✔ For more insights and tutorials on LLMs AI Agents and Machine Learning" [X Link](https://x.com/akshay_pachaar/status/1998628862146785640) 2025-12-10T05:40Z 239.1K followers, 2133 engagements "Build a XXX% local private and secure MCP client. You can connect it to any MCP server. Step-by-step guide:" [X Link](https://x.com/akshay_pachaar/status/1998695004571840811) 2025-12-10T10:03Z 239.1K followers, 3669 engagements "Google just dropped a new LLM You can run it locally on just XXX GB RAM. Let's fine-tune this on our own data (100% locally):" [X Link](https://x.com/akshay_pachaar/status/1956334618715800005) 2025-08-15T12:38Z 239.1K followers, 2M engagements "8 key skills to become a full-stack AI Engineer:" [X Link](https://x.com/akshay_pachaar/status/1964680446446567822) 2025-09-07T13:21Z 239.1K followers, 501.8K engagements "Meta just solved the biggest problem in RAG Most RAG systems waste your money. They retrieve XXX chunks when you only need XX. They force the LLM to process thousands of irrelevant tokens. You pay for compute you don't need. Meta AI just solved this. They built REFRAG a new RAG approach that compresses and filters context before it hits the LLM. The results are insane: - 30.85x faster time-to-first-token - 16x larger context windows - 2-4x fewer tokens processed - Outperforms LLaMA on XX RAG benchmarks Here's what makes REFRAG different: Traditional RAG dumps everything into the LLM. Every" [X Link](https://x.com/akshay_pachaar/status/1989327114303398379) 2025-11-14T13:38Z 239.1K followers, 104.8K engagements "Finally an open-source Python library for Context engineering Pixeltable is a unified declarative framework that handles your entire multimodal pipeline from data storage to model execution. The idea is simple: instead of stitching together a vector database a SQL database an embedding service and an agent framework everything lives in one system. Your documents embeddings conversation history and agent outputs are all just tables. Embeddings are computed columns that update automatically. Vector search works alongside your regular data operations. Built for end-to-end context engineering" [X Link](https://x.com/akshay_pachaar/status/1991409231040565379) 2025-11-20T07:32Z 239K followers, 66.8K engagements "Google just dropped "Attention is all you need (V2)" This paper could solve AI's biggest problem: Catastrophic forgetting. When AI models learn something new they tend to forget what they previously learned. Humans don't work this way and now Google Research has a solution. Nested Learning. This is a new machine learning paradigm that treats models as a system of interconnected optimization problems running at different speeds - just like how our brain processes information. Here's why this matters: LLMs don't learn from experiences; they remain limited to what they learned during training." [X Link](https://x.com/akshay_pachaar/status/1992507686802624701) 2025-11-23T08:17Z 239.1K followers, 515.6K engagements "Youre in an ML Engineer interview at Google. Interviewer: We need to train an LLM across 1000 GPUs. How would you make sure all GPUs share what they learn You: Use a central parameter server to aggregate and redistribute the weights. Interview over. Heres what you missed:" [X Link](https://x.com/akshay_pachaar/status/1992571349332804081) 2025-11-23T12:30Z 239.1K followers, 370.9K engagements "NVIDIA just dropped a paper that might solve the biggest trade-off in LLMs. Speed vs. Quality. Autoregressive models (like GPT) are smart but slow - they generate one token at a time leaving most of your GPU sitting idle. Diffusion models are fast but often produce incoherent outputs. TiDAR gets you both in a single forward pass. Here's the genius part: Modern GPUs can process way more tokens than we actually use. TiDAR exploits these "free slots" by: X. Drafting multiple tokens at once using diffusion (the "thinking" phase) X. Verifying them using autoregression (the "talking" phase) Both" [X Link](https://x.com/akshay_pachaar/status/1993658535679508599) 2025-11-26T12:30Z 239K followers, 79.6K engagements "8 key skills to become a full-stack AI Engineer: (free/open-source resources below)" [X Link](https://x.com/akshay_pachaar/status/1994759678245712089) 2025-11-29T13:25Z 239.1K followers, 45.6K engagements "RAG was never the end goal. Memory in AI agents is where everything is heading. Let me break down this evolution in the simplest way possible. RAG (2020-2023): - Retrieve info once generate response - No decision-making just fetch and answer - Problem: Often retrieves irrelevant context Agentic RAG: - Agent decides *if* retrieval is needed - Agent picks *which* source to query - Agent validates *if* results are useful - Problem: Still read-only can't learn from interactions AI Memory: - Read AND write to external knowledge - Learns from past conversations - Remembers user preferences past" [X Link](https://x.com/akshay_pachaar/status/1995108099007693206) 2025-11-30T12:30Z 239.1K followers, 98.2K engagements "Everyone building AI agents keeps making the same database mistake They give agents direct read/write access to their production database thinking that rate limits and permissions will keep things safe. Here's what actually happens: The Agent runs XX parallel queries to optimize something. Each query seems reasonable in isolation. Together they create a cascade of locks updates and resource consumption that brings down your entire system. Or worse: The agent decides to test something and to do that: - It creates a new index. - Drops it. - Creates another. - Tests a migration. - Rolls it back." [X Link](https://x.com/akshay_pachaar/status/1995451244874510527) 2025-12-01T11:13Z 239.1K followers, 67.9K engagements "I boosted my AI Agent's performance by XXX% Using a fully open-source technique. Here's the full breakdown (with code):" [X Link](https://x.com/akshay_pachaar/status/1995840565695635892) 2025-12-02T13:00Z 239K followers, 52.7K engagements "DeepSeek cracked the O(L) attention bottleneck. Their new V3.2 model introduces DeepSeek Sparse Attention (DSA) and it's the only architectural change they made. That tells you how important this is. What does it solve: Standard attention scales quadratically. Double your context length quadruple the compute. This is why long-context inference gets expensive fast. DSA brings complexity down from O(L) to O(Lk) where k is fixed. How it works: A lightweight Lightning Indexer scores which tokens actually matter for each query. Small number of heads runs in FP8 computationally cheap. Then a" [X Link](https://x.com/akshay_pachaar/status/1996195253490192579) 2025-12-03T12:30Z 239.1K followers, 59.6K engagements "8 AI model architectures visually explained: Everyone talks about LLMs but there's a whole family of specialized models doing incredible things. Here's a quick breakdown: X. LLM (Large Language Models) Text goes in gets tokenized into embeddings processed through transformers and text comes out. ChatGPT Claude Gemini Llama. X. LCM (Large Concept Models) Works at concept level not tokens. Input is segmented into sentences passed through SONAR embeddings then uses diffusion before output. Meta's LCM is the pioneer. X. LAM (Large Action Models) Turns intent into action. Input flows through" [X Link](https://x.com/akshay_pachaar/status/1996569562658337218) 2025-12-04T13:17Z 239.1K followers, 67.2K engagements "You're in a Research Scientist interview at Google. Interviewer: We have a base LLM that's terrible at maths. How would you turn it into a maths & reasoning powerhouse You: I'll get some problems labeled and fine-tune the model. Interview over. Here's what you missed:" [X Link](https://x.com/akshay_pachaar/status/1997284196306633111) 2025-12-06T12:37Z 239.1K followers, 359.3K engagements "Google just dropped X new papers at NeurIPS. It could fundamentally change how AI handles memory. Here's the problem they're solving: Transformers are powerful but expensive. The longer the context the slower and costlier they get. That's a real limitation when you need to process entire documents or genomic sequences. The research community tried fixes like Mamba and other linear models. They're fast but compress everything into a fixed-size state. And as the context gets longer they start losing important details. Google introduced Titans and MIRAS to fix this. Titans is the architecture." [X Link](https://x.com/akshay_pachaar/status/1997654015631651194) 2025-12-07T13:06Z 239.1K followers, 55.9K engagements "HuggingFace just made fine-tuning 10x easier One line of English to fine-tune any open-source LLM. They released a new "skill" you can plug into Claude or any coding agent. It doesn't just write training scripts but actually submits jobs to cloud GPUs monitors progress and pushes finished models to the Hub. Here's how it works: You say something like: "Fine-tune Qwen3-0.6B on the open-r1/codeforces-cots dataset" And Claude will: Validate your dataset format Select appropriate GPU hardware Submit the job to Hugging Face Jobs Monitor training progress Push the finished model to the Hub The" [X Link](https://x.com/akshay_pachaar/status/1997946287556321359) 2025-12-08T08:28Z 239.1K followers, 81.7K engagements "NaiveRAG is fast but dumb. GraphRAG is smart but slow. This open-source solution fixes both. RAG systems have a fundamental problem: They treat documents as isolated chunks. No connections. No context. No understanding of how things relate. Graph RAG addresses this but traditional graph databases become painfully slow for real-time applications. What if you could combine the speed of vector search with the intelligence of knowledge graphs That's exactly what I built. A real-time AI Avatar that uses a knowledge graph as its memory. You talk to it directly everything happens in real-time. Watch" [X Link](https://x.com/akshay_pachaar/status/1998007894458155100) 2025-12-08T12:32Z 239.1K followers, 51.9K engagements "Find all the code here: Graphiti GitHub repo: (don't forget to star 🌟)" [X Link](https://x.com/akshay_pachaar/status/1998013294817444138) 2025-12-08T12:54Z 239.1K followers, 8710 engagements "Microsoft. Google. AWS. Everyone's trying to solve the same problem for AI Agents: How to connect your agents to enterprise data without duct-taping a dozen tools together Your data lives in Postgres Snowflake MongoDB Gmail etc scattered across dozens of apps. Your AI logic lives in Python scripts and vector databases. Building manual RAG pipelines with custom connectors for every data source means you're already set up for failure. Here's an open-source project tackling this differently: MindsDB treats AI models as virtual tables. Instead of moving data to AI it brings AI to the data. The" [X Link](https://x.com/akshay_pachaar/status/1998454729475764491) 2025-12-09T18:08Z 239.1K followers, 16.6K engagements "Every company I talk to is literally trying to solve this problem: How to let AI handle DevOps without risking a production wipeout. The typical DevOps workflow today involves: - Hours of debugging server configs - Manually writing Terraform scripts - Searching scattered docs and forums - Copy-pasting CI/CD pipeline setups - Scanning deployment logs line by line AI could automate much of this but the fear of just one hallucinated kubectl delete command that can wipe out an entire production cluster is real. For instance in July 2025 Replit's Agent wiped out a company's entire production DB." [X Link](https://x.com/akshay_pachaar/status/1998732160140980724) 2025-12-10T12:30Z 239.1K followers, 1770 engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@akshay_pachaar Akshay 🚀
Akshay 🚀 posts on X about ai, $googl, solve, llm the most. They currently have XXXXXXX followers and 1559 posts still getting attention that total XXXXXX engagements in the last XX hours.
Engagements: XXXXXX #
- X Week XXXXXXX -XX%
- X Month XXXXXXXXX -XX%
- X Months XXXXXXXXXX +38%
- X Year XXXXXXXXXX +82%
Mentions: XX #
- X Week XX -XX%
- X Month XXX -XX%
- X Months XXX +111%
- X Year XXX +148%
Followers: XXXXXXX #
- X Week XXXXXXX +0.61%
- X Month XXXXXXX +2.20%
- X Months XXXXXXX +14%
- X Year XXXXXXX +42%
CreatorRank: XXXXXXX #
Social Influence
Social category influence technology brands stocks #4797 finance
Social topic influence ai #5800, $googl #556, solve #227, llm #11, microsoft #117, if you #6033, rag #7, token, agentic #310, snowflake #48
Top accounts mentioned or mentioned by @akshaypachaar @avichawla @lightningai @llamaindex @crewaiinc @ollama @mlspring @firecrawldev @zepai @qdrantengine @cometml @milvusio @abacusai @assemblyai @tinztwins @streamlit @cometmls @cleanlabai @dailydoseofds @activeloopai
Top assets mentioned Alphabet Inc Class A (GOOGL) Microsoft Corp. (MSFT)
Top Social Posts
Top posts by engagements in the last XX hours
"I have been training neural networks for XX years now. Here are XX ways I actively use to optimize model training: (detailed explanation .🧵)"
X Link 2025-12-05T13:05Z 239.1K followers, 6346 engagements
"So let's dive in and understand how KV caching works.👇"
X Link 2025-10-20T12:30Z 238.9K followers, 8567 engagements
"Massive update for AI Engineers Training diffusion models just got a lot easier. dLLM is an open-source library that does for diffusion models what Hugging Face did for transformers. Here's why this matters: Traditional autoregressive models generate text left-to-right one token at a time. Diffusion models work differently - they refine the entire sequence iteratively giving you better control over generation quality and more flexible editing capabilities. The problem Building and training these models required stitching together scattered tools and reimplementing research papers from"
X Link 2025-11-15T12:52Z 238.9K followers, 64.1K engagements
"Claude Scientific Skills. Turn Claude into your AI research assistant capable of executing complex multi-step scientific workflows across maths biology chemistry medicine and beyond. XXX% open-source (123+ skills)"
X Link 2025-11-28T13:01Z 238.9K followers, 57.1K engagements
"Top AI Engineers never do manual prompt engineering. Today I'll show you how to automatically find the best prompts for any agentic workflow you're building. We'll use @Cometml's XXX% open-source Opik to do so. Let's go 🚀"
X Link 2025-12-02T13:00Z 239.1K followers, 4354 engagements
"Microsoft did it again Building with AI agents almost never works on the first try. You spend days tweaking prompts adding examples hoping it gets better. Nothing systematic just guesswork. This is exactly what Microsoft's Agent Lightning solves. It's an open-source framework that trains ANY AI agent with reinforcement learning. Works with LangChain AutoGen CrewAI OpenAI SDK or plain Python. Here's how it works: Your agent runs normally with whatever framework you're using. Just add a lightweight agl.emit() helper or let the tracer auto-collect everything. Agent Lightning captures every"
X Link 2025-10-31T12:30Z 239K followers, 204.6K engagements
"This is a goldmine of AI resources by MongoDB (free and geared towards real-world AI engineering) Building AI prototypes locally is fun. You can experiment quickly push code and try out different models with almost zero setup. Building AI for real users is where things get tricky. You need to handle storage retrieval performance security and scalable context management. MongoDBs AI resource hub solves that learning gap beautifully. It gives you a complete ecosystem of guides demos and learning tracks designed for developers who want to build production-grade AI applications with reliable data"
X Link 2025-11-28T07:49Z 239K followers, 12.4K engagements
"Mistral just dropped their biggest release ever. And the real story isn't the benchmarks. It's what they're giving away for free. Here's what you need to know: The Release: Four new models. All Apache XXX. All open source. - Ministral 3B 8B 14B for edge devices - Mistral Large 3: 675B total parameters 41B active That last one A sparse mixture-of-experts beast trained from scratch on 3000 H200 GPUs. Why It Matters: Most companies release "open" models with asterisks everywhere. Mistral just handed you: Base models (train your own) Instruct models (deploy immediately) Reasoning variants"
X Link 2025-12-02T19:28Z 238.9K followers, 22.8K engagements
"You're in an ML Engineer interview at Apple. The interviewer asks: "Two models are XX% accurate. - Model A is XX% confident. - Model B is XX% confident. Which one would you pick" You: "Any would work since both have same accuracy." Interview over. Here's what you missed: Modern neural networks can be misleading. They are overconfident in their predictions. For instance I saw an experiment that used the CIFAR-100 dataset to compare LeNet with ResNet. LeNet produced: - Accuracy = XXXX - Average confidence = XXXX ResNet produced: - Accuracy = XXX - Average confidence = XXX Despite being more"
X Link 2025-11-27T12:30Z 239K followers, 126.6K engagements
"If you found it insightful reshare with your network. Find me @akshay_pachaar ✔ For more insights and tutorials on LLMs AI Agents and Machine Learning"
X Link 2025-12-10T05:40Z 239.1K followers, 2133 engagements
"Build a XXX% local private and secure MCP client. You can connect it to any MCP server. Step-by-step guide:"
X Link 2025-12-10T10:03Z 239.1K followers, 3669 engagements
"Google just dropped a new LLM You can run it locally on just XXX GB RAM. Let's fine-tune this on our own data (100% locally):"
X Link 2025-08-15T12:38Z 239.1K followers, 2M engagements
"8 key skills to become a full-stack AI Engineer:"
X Link 2025-09-07T13:21Z 239.1K followers, 501.8K engagements
"Meta just solved the biggest problem in RAG Most RAG systems waste your money. They retrieve XXX chunks when you only need XX. They force the LLM to process thousands of irrelevant tokens. You pay for compute you don't need. Meta AI just solved this. They built REFRAG a new RAG approach that compresses and filters context before it hits the LLM. The results are insane: - 30.85x faster time-to-first-token - 16x larger context windows - 2-4x fewer tokens processed - Outperforms LLaMA on XX RAG benchmarks Here's what makes REFRAG different: Traditional RAG dumps everything into the LLM. Every"
X Link 2025-11-14T13:38Z 239.1K followers, 104.8K engagements
"Finally an open-source Python library for Context engineering Pixeltable is a unified declarative framework that handles your entire multimodal pipeline from data storage to model execution. The idea is simple: instead of stitching together a vector database a SQL database an embedding service and an agent framework everything lives in one system. Your documents embeddings conversation history and agent outputs are all just tables. Embeddings are computed columns that update automatically. Vector search works alongside your regular data operations. Built for end-to-end context engineering"
X Link 2025-11-20T07:32Z 239K followers, 66.8K engagements
"Google just dropped "Attention is all you need (V2)" This paper could solve AI's biggest problem: Catastrophic forgetting. When AI models learn something new they tend to forget what they previously learned. Humans don't work this way and now Google Research has a solution. Nested Learning. This is a new machine learning paradigm that treats models as a system of interconnected optimization problems running at different speeds - just like how our brain processes information. Here's why this matters: LLMs don't learn from experiences; they remain limited to what they learned during training."
X Link 2025-11-23T08:17Z 239.1K followers, 515.6K engagements
"Youre in an ML Engineer interview at Google. Interviewer: We need to train an LLM across 1000 GPUs. How would you make sure all GPUs share what they learn You: Use a central parameter server to aggregate and redistribute the weights. Interview over. Heres what you missed:"
X Link 2025-11-23T12:30Z 239.1K followers, 370.9K engagements
"NVIDIA just dropped a paper that might solve the biggest trade-off in LLMs. Speed vs. Quality. Autoregressive models (like GPT) are smart but slow - they generate one token at a time leaving most of your GPU sitting idle. Diffusion models are fast but often produce incoherent outputs. TiDAR gets you both in a single forward pass. Here's the genius part: Modern GPUs can process way more tokens than we actually use. TiDAR exploits these "free slots" by: X. Drafting multiple tokens at once using diffusion (the "thinking" phase) X. Verifying them using autoregression (the "talking" phase) Both"
X Link 2025-11-26T12:30Z 239K followers, 79.6K engagements
"8 key skills to become a full-stack AI Engineer: (free/open-source resources below)"
X Link 2025-11-29T13:25Z 239.1K followers, 45.6K engagements
"RAG was never the end goal. Memory in AI agents is where everything is heading. Let me break down this evolution in the simplest way possible. RAG (2020-2023): - Retrieve info once generate response - No decision-making just fetch and answer - Problem: Often retrieves irrelevant context Agentic RAG: - Agent decides if retrieval is needed - Agent picks which source to query - Agent validates if results are useful - Problem: Still read-only can't learn from interactions AI Memory: - Read AND write to external knowledge - Learns from past conversations - Remembers user preferences past"
X Link 2025-11-30T12:30Z 239.1K followers, 98.2K engagements
"Everyone building AI agents keeps making the same database mistake They give agents direct read/write access to their production database thinking that rate limits and permissions will keep things safe. Here's what actually happens: The Agent runs XX parallel queries to optimize something. Each query seems reasonable in isolation. Together they create a cascade of locks updates and resource consumption that brings down your entire system. Or worse: The agent decides to test something and to do that: - It creates a new index. - Drops it. - Creates another. - Tests a migration. - Rolls it back."
X Link 2025-12-01T11:13Z 239.1K followers, 67.9K engagements
"I boosted my AI Agent's performance by XXX% Using a fully open-source technique. Here's the full breakdown (with code):"
X Link 2025-12-02T13:00Z 239K followers, 52.7K engagements
"DeepSeek cracked the O(L) attention bottleneck. Their new V3.2 model introduces DeepSeek Sparse Attention (DSA) and it's the only architectural change they made. That tells you how important this is. What does it solve: Standard attention scales quadratically. Double your context length quadruple the compute. This is why long-context inference gets expensive fast. DSA brings complexity down from O(L) to O(Lk) where k is fixed. How it works: A lightweight Lightning Indexer scores which tokens actually matter for each query. Small number of heads runs in FP8 computationally cheap. Then a"
X Link 2025-12-03T12:30Z 239.1K followers, 59.6K engagements
"8 AI model architectures visually explained: Everyone talks about LLMs but there's a whole family of specialized models doing incredible things. Here's a quick breakdown: X. LLM (Large Language Models) Text goes in gets tokenized into embeddings processed through transformers and text comes out. ChatGPT Claude Gemini Llama. X. LCM (Large Concept Models) Works at concept level not tokens. Input is segmented into sentences passed through SONAR embeddings then uses diffusion before output. Meta's LCM is the pioneer. X. LAM (Large Action Models) Turns intent into action. Input flows through"
X Link 2025-12-04T13:17Z 239.1K followers, 67.2K engagements
"You're in a Research Scientist interview at Google. Interviewer: We have a base LLM that's terrible at maths. How would you turn it into a maths & reasoning powerhouse You: I'll get some problems labeled and fine-tune the model. Interview over. Here's what you missed:"
X Link 2025-12-06T12:37Z 239.1K followers, 359.3K engagements
"Google just dropped X new papers at NeurIPS. It could fundamentally change how AI handles memory. Here's the problem they're solving: Transformers are powerful but expensive. The longer the context the slower and costlier they get. That's a real limitation when you need to process entire documents or genomic sequences. The research community tried fixes like Mamba and other linear models. They're fast but compress everything into a fixed-size state. And as the context gets longer they start losing important details. Google introduced Titans and MIRAS to fix this. Titans is the architecture."
X Link 2025-12-07T13:06Z 239.1K followers, 55.9K engagements
"HuggingFace just made fine-tuning 10x easier One line of English to fine-tune any open-source LLM. They released a new "skill" you can plug into Claude or any coding agent. It doesn't just write training scripts but actually submits jobs to cloud GPUs monitors progress and pushes finished models to the Hub. Here's how it works: You say something like: "Fine-tune Qwen3-0.6B on the open-r1/codeforces-cots dataset" And Claude will: Validate your dataset format Select appropriate GPU hardware Submit the job to Hugging Face Jobs Monitor training progress Push the finished model to the Hub The"
X Link 2025-12-08T08:28Z 239.1K followers, 81.7K engagements
"NaiveRAG is fast but dumb. GraphRAG is smart but slow. This open-source solution fixes both. RAG systems have a fundamental problem: They treat documents as isolated chunks. No connections. No context. No understanding of how things relate. Graph RAG addresses this but traditional graph databases become painfully slow for real-time applications. What if you could combine the speed of vector search with the intelligence of knowledge graphs That's exactly what I built. A real-time AI Avatar that uses a knowledge graph as its memory. You talk to it directly everything happens in real-time. Watch"
X Link 2025-12-08T12:32Z 239.1K followers, 51.9K engagements
"Find all the code here: Graphiti GitHub repo: (don't forget to star 🌟)"
X Link 2025-12-08T12:54Z 239.1K followers, 8710 engagements
"Microsoft. Google. AWS. Everyone's trying to solve the same problem for AI Agents: How to connect your agents to enterprise data without duct-taping a dozen tools together Your data lives in Postgres Snowflake MongoDB Gmail etc scattered across dozens of apps. Your AI logic lives in Python scripts and vector databases. Building manual RAG pipelines with custom connectors for every data source means you're already set up for failure. Here's an open-source project tackling this differently: MindsDB treats AI models as virtual tables. Instead of moving data to AI it brings AI to the data. The"
X Link 2025-12-09T18:08Z 239.1K followers, 16.6K engagements
"Every company I talk to is literally trying to solve this problem: How to let AI handle DevOps without risking a production wipeout. The typical DevOps workflow today involves: - Hours of debugging server configs - Manually writing Terraform scripts - Searching scattered docs and forums - Copy-pasting CI/CD pipeline setups - Scanning deployment logs line by line AI could automate much of this but the fear of just one hallucinated kubectl delete command that can wipe out an entire production cluster is real. For instance in July 2025 Replit's Agent wiped out a company's entire production DB."
X Link 2025-12-10T12:30Z 239.1K followers, 1770 engagements