@chengyenhsieh wesley hsieh

wesley hsieh posts on X about bytedance, ai, seedance, this is the most. They currently have [---] followers and [--] posts still getting attention that total [-------] engagements in the last [--] hours.

Engagements: [-------] #

[--] Week [-----] -80%
[--] Month [------] +3,005%
[--] Months [------] +50%
[--] Year [------] +53,550%

Mentions: [--] #

[--] Week [--] +28%
[--] Month [--] +1,467%
[--] Months [--] +195%
[--] Year [--] +7,400%

Followers: [---] #

[--] Week [---] +3.70%
[--] Month [---] +51%
[--] Months [---] +73%

CreatorRank: [---------] #

Social Influence

Social category influence technology brands stocks countries celebrities

Social topic influence bytedance #92, ai, seedance, this is, the first, science, seed, data, design, anime

Top accounts mentioned or mentioned by @zaixiangzheng @ylecuns @xinyounju @karpathy @drfeifei @adamdangelo @rauchg @scottbelsky @quanquangu @ycombinator @yuningshen1 @dugu9sword @feijkelboom @daanroos @sciencemag @theias @vanderbiltu @cambridgeuni @harvard @isomorphiclabs

Top assets mentioned Alphabet Inc Class A (GOOGL)

Top Social Posts

Top posts by engagements in the last [--] hours

"DPLM-2: A Multimodal Diffusion Protein Language Model LMs can be used for science. Multimodal LM with discrete diffusion enables the co-generation of both protein sequence and 3D structures. Page: Paper: https://arxiv.org/abs/2410.13782project https://bytedance.github.io/dplm/dplm-2 https://arxiv.org/abs/2410.13782project https://bytedance.github.io/dplm/dplm-2"
X Link 2024-10-18T23:14Z [---] followers, [---] engagements

"Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models 📢 ICLR Oral [----] 📌 Block diffusion overcomes key limitations of autoregressive/diffusion models by supporting flexible-length generation and improving inference efficiency with KV caching and parallel token sampling. Exciting to see another successful scheme for integrating autoregressive and diffusion models. 📜 Paper: 📟 Project: #ML #GenerativeAI #WesDailyExploration https://m-arriola.com/bd3lms/ https://openreview.net/forumid=tyEyYT267x https://m-arriola.com/bd3lms/"
X Link 2025-03-13T04:00Z [--] followers, [---] engagements

"🚀 Revolutionizing Visual Generation with Autoregressive Models 🔥 NeurIPS [----] Best Paper from ByteDance introduces a game-changing approach to visual generationrethinking Autoregressive (AR) modeling. Lets revist it 👇 #ML #ComputerVision #VisualGeneration #GenerativeAI #WesDailyExploration"
X Link 2025-03-18T18:59Z [--] followers, [---] engagements

"OpenAI just dropped GPT-4o image generation 🎨 🔥 Whats impressive 🔹 Precise Instruction Following GPT-4o nails detailed prompts with incredible attention to detail. 🔹 Multi-turn Image Editing I particularly love this feature. Upload your own image & iteratively refine it Just like the cat-to-game-character demo below 🐱➡🎮. "Give this cat a detective hat and a monocle" 🔓 Who can use it Starting today it's rolling out to Plus Pro Team and Free users as the default image generator in ChatGPT. 🤔 Hidden Bonus: OpenAIs website might just be giving us a sneak peek at their model designs. #ML"
X Link 2025-03-25T22:17Z [--] followers, [---] engagements

"🚨 ByteDance Research at #ICLR2025 Swing by the booth on April [--] 15:0015:30 @ Booth L07 🗓📍 @zaixiang_zheng will be sharing the latest breakthroughs in Generative AI for Science Chat with the team and see what weve been cooking up #ML #ICLR2025 Excited to be at #ICLR2025 in Singapore Come check out the latest from ByteDance Research's Generative AI for Science team: 🧬CryoFM: A Flow-based Foundation Model for Cryo-EM Densities (Apr [--] 15:00-17:30 Poster #5) https://t.co/bawX2IsNJC 🧬ProteinBench: A Holistic https://t.co/kwEUwwk2G1 Excited to be at #ICLR2025 in Singapore Come check out the"
X Link 2025-04-23T18:54Z [---] followers, [---] engagements

"📢 #ICML2025 Spotlight from ByteDance Research 🚀 Elucidating the Design Space of Multimodal Protein Language Models Multimodal protein Language Models (PLMs) such as ESM3 and DPLM-2 often struggle at protein structure modeling. We uncover why and how to fix it. Our designs enable the 650M multimodal PLM named DPLM-2.1 to beat 3B baselines and specialized folding models 👇 (1/n) 🧵"
X Link 2025-05-15T20:04Z [---] followers, 15.1K engagements

"🧬 Structure Data Exploration (Multimer) Multimer and monomer structures are more interconnected than expected. Incorporating multimer data boosts folding accuracy for both types unlocking broader generalization. We dive deeper into bridging their differences in the paper"
X Link 2025-05-15T20:04Z [---] followers, [---] engagements

"This work was done @bytedance (ByteDance Research) with the amazing team: Xinyou Wang (@Xinyou_NJU) as well as Daiheng Zhang (@dz2266) Dongyu Xue (@dngyxu1) Fei Ye (@FeiYE00844289) Shujian Huang and @zaixiang_zheng & @QuanquanGu"
X Link 2025-05-15T20:04Z [---] followers, [---] engagements

"The story behind @elonmusk founding Zip2: "It wasn't actually from the standpoint of I want to start a company. I just want to be part of building the internet. Since I couldn't get a job at an internet company I had to start an internet company" Love this "I'll do it myself" spirit. Big thanks to @ycombinator for the video footage of this event. https://www.youtube.com/watchv=cFIlta1GkiE&ab_channel=YCombinator https://www.youtube.com/watchv=cFIlta1GkiE&ab_channel=YCombinator"
X Link 2025-06-21T16:00Z [---] followers, [---] engagements

"👏 YCs AI Startup School with Dr. Fei-Fei Li Dr. Fei Fei Li the godmother of AI shared key moments from her research journey during the AI Startup School hosted by @ycombinator Below are some stories that I found particularly interesting. 🔹 The creation of ImageNet: In an era when data-driven algorithms were limited by a lack of large-scale datasets Dr. Li and the team set out to build a dataset that reflected the complexity of the real world. Using Amazon Mechanical Turk to collect and label images from the internet they published ImageNet in [----] which fundamentally shifted the direction"
X Link 2025-06-21T17:00Z [---] followers, [---] engagements

"📌Notes on Boltz-2 Just watched the video talk led by @GabriCorso @jeremyWohlwend and Saro Passaro that introduced Boltz-2 a structural biology foundation models. I summarized some learning notes below 🧵"
X Link 2025-06-28T02:00Z [---] followers, 16.7K engagements

"📌 Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction Boltz-2 developed by MIT researchers is a new structural biology foundation model. While AlphaFold3 and Boltz-1 excel in structure prediction they struggle with binding affinity. Boltz-2 aims to predict both accurately"
X Link 2025-06-28T02:00Z [---] followers, [----] engagements

"📌Improvements on Capturing Local Dynamics Boltz-2 MD ensembles generally obtain stronger correlations with the ground truth simulation and lower errors than Boltz-1 BioEmu and AlphaFlow. A significant gap with other baselines however lies in the diversity (Diversity IDDT)"
X Link 2025-06-28T02:00Z [---] followers, [---] engagements

"📌Physical Quality of Poses Issues in AF3: Chirality Bond Distances Stereochemistry and Steric Clashes Solution: Feynman-Kac steering inference-time potentials (A sampling trick for diffusion models) The high level idea is: [--]. Define function to guide diffusion paths that are more amenable to the potential function. [--]. Make small adjustments to the final x0 with respect to the potential function Results: Physical validity jumps to near perfect even though RMSD is nearly the same"
X Link 2025-06-28T02:00Z [---] followers, [---] engagements

"🎉 Just arrived at #ICML2025 Well be presenting our DPLM-2.1 poster on Tuesday. Come chat with me and @zaixiang_zheng about protein modeling and the next generation of diffusion protein language models (DPLM). 📍 Location: West Exhibition Hall B2-B3 Poster #W-115 🕓 Time: Tuesday July [--] 4:307:00 p.m. PDT Wed love to connect and answer questions. Learng more about DPLM: #ProteinModeling #AI4Science #GenerativeModel https://bytedance.github.io/dplm/ 📢 #ICML2025 Spotlight from ByteDance Research 🚀 Elucidating the Design Space of Multimodal Protein Language Models Multimodal protein Language"
X Link 2025-07-14T19:35Z [---] followers, [---] engagements

"🎉 DPLM-2.1 at #ICML2025 It's happening today (7/15). Well be presenting our DPLM-2.1 poster. Come chat with me and @zaixiang_zheng about protein modeling and the next generation of diffusion protein language models (DPLM). 📍 Location: West Exhibition Hall B2-B3 Poster #W-115 🕓 Time: Tuesday July [--] 4:307:00 p.m. PDT Wed love to connect and answer questions. Learng more about DPLM: #ProteinModeling #AI4Science #GenerativeModel https://bytedance.github.io/dplm/ 📢 #ICML2025 Spotlight from ByteDance Research 🚀 Elucidating the Design Space of Multimodal Protein Language Models Multimodal"
X Link 2025-07-15T17:00Z [---] followers, [---] engagements

"Our paper is now featured on the official ByteDance SEED publications page. Elucidating the Design Space of Multimodal Protein Language Models (ICML [----] Spotlight) explores the intersection of structure and sequence for multimodal protein foundation models. Check it out here: https://seed.bytedance.com/en/public_papers/elucidating-the-design-space-of-multimodal-protein-language-models https://seed.bytedance.com/en/public_papers/elucidating-the-design-space-of-multimodal-protein-language-models"
X Link 2025-07-31T19:00Z [---] followers, [--] engagements

"Great article. I've always wondered how language models generate text when the context length exceeds the training context window. You could naively increase the kv cache size but the performance would degrade dramatically. Attention sinks shows that the first few tokens (four tokens) in a long sentence typically retain most information. At inference this property can be used to manage long conversations without increasing the KV cache size. I've written the full story of Attention Sinks a technical deep-dive into how the mechanism was developed and how our research ended up being used in"
X Link 2025-08-10T04:02Z [---] followers, [---] engagements

"LLM Training in a Nutshell: Pre-training vs. Post-training Training LLMs usually comprises two core phases: pre-training and post-training. Pre-training: We train the model on massive amounts of internet-scale data and assume that knowledge across various domains is encoded and hidden within the LM. However this usually does not suffice because the model wasn't trained to "answer the question". Post-training: This is where we align the LLM to follow our prompts. Post-training typically comes in two flavors: [--]. Supervised fine-tuning (SFT) [--]. Reinforcement learning (RL) Lately Ive been diving"
X Link 2025-09-13T16:00Z [---] followers, [--] engagements

"I have to admit that Ive never really systematically differentiated these LaTeX math fonts and their typical use cases until recently (despite their importance in paper writing) Thank you ChatGPT 😅"
X Link 2025-10-16T00:30Z [---] followers, [--] engagements

"📌 [--]. Tracing the thoughts of a large language model by @AnthropicAI I'm so fascinated by the explainability study of LLM recently. The study investigates the internal processes of Claude [---] Haiku revealing that the model operates in a conceptual space shared across multiple languages. 🔗 https://www.anthropic.com/research/tracing-thoughts-language-model https://www.anthropic.com/research/tracing-thoughts-language-model"
X Link 2025-04-16T22:01Z [---] followers, [--] engagements

"📌 [--]. OpenAI o3 & o4-mini by @OpenAI The o3 model excels in complex tasks such as coding mathematics and visual analysis while o4-mini offers efficient performance for use cases demanding high efficiency. These new models enable sophisticated reasoning across text images and tools. 🔗 https://openai.com/index/introducing-o3-and-o4-mini/ https://openai.com/index/introducing-o3-and-o4-mini/"
X Link 2025-04-16T22:01Z [---] followers, [---] engagements

"Attending #NeurIPS [----]. Happy to chat about diffusion protein language models (DPLM) from #ByteDanceSeed and multimodal generative models. Feel free to reach out if you also work in a related space. Our team is also recruiting [----] summer research interns. Feel free to chat if you are curious about generative models for proteins"
X Link 2025-12-03T23:39Z [---] followers, [---] engagements

"DPLM: Generative AI for Science (2026 Research Intern): https://job.toutiao.com/s/3ep5etF_SFc https://bytedance.github.io/dplm/ https://job.toutiao.com/s/3ep5etF_SFc https://bytedance.github.io/dplm/"
X Link 2025-12-03T23:39Z [---] followers, [---] engagements

"Confrover #1709 Essentially built a video model to infer protein dynamics @YuningShen1"
X Link 2025-12-05T00:44Z [---] followers, [---] engagements

"Interesting position paper: Machine learning conferences should establish a refutation and critiques track. By @yegordb. Have you ever struggled to reproduce prior work only for the authors to never reply to your email This paper proposes that conferences create a refutation track allowing the community to critique and improve existing work"
X Link 2025-12-05T02:35Z [---] followers, [---] engagements

"Yearly review from DeepMind. Very captivating. Well motivated to do some solid work next year. [----] was a year of AI agents reasoning and scientific discovery. Congratulations to all the teams and partners across @google and beyond 🚀 Heres our full recap 👇 https://t.co/mTqhQnB0Tr [----] was a year of AI agents reasoning and scientific discovery. Congratulations to all the teams and partners across @google and beyond 🚀 Heres our full recap 👇 https://t.co/mTqhQnB0Tr"
X Link 2025-12-24T10:06Z [---] followers, [---] engagements

"I often feel like I've only leveraged 10% of the efficiency gains that AI coding agents could offer. Coding has gradually shifted from being "efficiency-bound" (building what you want) to "knowledge-bound" (knowing what to build). In my case the least efficient scenarios occur when I hesitate between several implementation choices while integrating features into an existing pipeline. This stems from limited understanding of best design principles (e.g. modularization) or tools (e.g. reinventing the wheel). That said engineering and system-level knowledge become even more important with"
X Link 2025-12-27T05:36Z [---] followers, [---] engagements

"Congrats Boltz Lab is on its mission to make AI for biology more accessible and useful for science. Big news from Boltz today: were launching Boltz Lab a new platform with new small-molecule + protein design agents announcing Boltz PBC and a $28M seed round and sharing a multi-year partnership with Pfizer. More below 🚀 https://t.co/FJnT4gDgn4 Big news from Boltz today: were launching Boltz Lab a new platform with new small-molecule + protein design agents announcing Boltz PBC and a $28M seed round and sharing a multi-year partnership with Pfizer. More below 🚀 https://t.co/FJnT4gDgn4"
X Link 2026-01-08T14:42Z [---] followers, [---] engagements

"@polynoamial @kellerjordan0 The reality is that most people dont have enough computing resources to build stuff that might be of interest to frontier labs. What do you suggest"
X Link 2026-01-22T15:09Z [---] followers, [---] engagements

"👀Seedance [---] is out. Wow. DeepMind has Veo3. ByteDance Seed has Seedance. ByteDance Seeds research velocity is real and Seedance makes that hard to deny. One day this research branding will be too strong to ignore"
X Link 2026-02-07T15:41Z [---] followers, 32.3K engagements

"🧬 Protein Autoregressive Modeling (PAR) ByteDance Seed We introduce PAR the first multi-scale AR framework for protein backbone generation. Autoregressive modeling of 3D protein structures has long been considered difficult. Heres how we make it possible with PAR. 👇 https://par-protein.github.io/ https://par-protein.github.io/"
X Link 2026-02-05T20:23Z [---] followers, 13.9K engagements

"🛡 Exposure Bias: To solve the train-inference mismatch we adapted Noisy Context Learning (NCL) and Scheduled Sampling (SS). By training on corrupted contexts and model predictions we enable robust generation and sampling quality. 👉Result: Subtantially improved the degraded sampling designability in our preliminary study on a 60M PAR"
X Link 2026-02-05T20:23Z [---] followers, [---] engagements

"📊 Unconditional Backbone Generation: PAR effectively captures the true protein data distribution achieving a Frchet Protein Structure Distance (FPSD) of [-----] against the PDB. It maintains competitive designability reaching 96.6%. https://twitter.com/i/web/status/2019507295399604249 https://twitter.com/i/web/status/2019507295399604249"
X Link 2026-02-05T20:23Z [---] followers, [---] engagements

"Someone please remake One Punch Man with Seedance Gojo vs Naruto. Saitama vs Genos. Now all of this is possible with Seedance [---]. A Chinese model with almost no censorship and its about to take over. In the coming days youll see your favorite anime characters fighting with studio-level quality spectacular animation and https://t.co/CG4AHdOkwT Gojo vs Naruto. Saitama vs Genos. Now all of this is possible with Seedance [---]. A Chinese model with almost no censorship and its about to take over. In the coming days youll see your favorite anime characters fighting with studio-level quality"
X Link 2026-02-08T00:02Z [---] followers, [----] engagements

"Interesting assumption: iterative denoising at the same noise level can keep the video on a low-dimensional manifold where the original data lies thereby improving physical realism. What if your video generator could refine itselfat inference time ❌No new models. ❌No retraining. ❌No external verifier. 💡 Introducing Self-Refining Video Sampling By reinterpreting a pretrained generator (Wan2.2 Cosmos) as a denoising autoencoder we enable iterative https://t.co/tNENBmMDa0 What if your video generator could refine itselfat inference time ❌No new models. ❌No retraining. ❌No external verifier. 💡"
X Link 2026-01-30T05:13Z [---] followers, [--] engagements

"Energy Demands of AI: Current advances in AI are dependent on large terrestrial data centers. This in the long-term impose hardship on communities and the environment. Some statistics: 🌍 Earth's energy today Global electricity consumption: [-----] TWh/year Average power: [---] TW In [----] U.S. data centers consumed [---] TWh of electricity in [----] more than 4% of the countrys total consumption. https://www.pewresearch.org/short-reads/2025/10/24/what-we-know-about-energy-use-at-us-data-centers-amid-the-ai-boom/#how-much-energy-do-data-centers-use"
X Link 2026-02-03T02:29Z [---] followers, [--] engagements

"How to solve this Simply put the idea is to transform these resource-intensive facilities to a location with vast space and energy i.e. space. https://twitter.com/i/web/status/2018512112147177680 https://twitter.com/i/web/status/2018512112147177680"
X Link 2026-02-03T02:29Z [---] followers, [--] engagements

"🚀 Space AI Potential 📌Short-term milestone: 1M tons per year [--] million tons / satellites per year x [---] kW/ ton = [---] GW (0.1 TW) About 3% of global electricity. 📌Mid-term milestone: meet civilization's power production. Simply launch more satellites - [--] TW/year Comparable to the entire Earth civilizations instantaneous power. 📌Long-term milestone: Lunar factory Factories on the Moon can take advantage of lunar resources to manufacture satellites and deploy them further into space - put [---] to [----] TW/year of AI satellites into deep space. [------] Earth's current electricity consumption"
X Link 2026-02-03T02:29Z [---] followers, [--] engagements

"PAR exhibits several striking empirical behaviors: 📊 SOTA Generative Power: Achieves a SOTA FPSD of [-----] against the PDB capturing the true protein distribution with high fidelity. 🎨 Interpretable Process: Like sculpting a statue it first forms the global topology and then progressively refines structural details across scales. ✨ Zero-Shot Versatility: Supports 16-point prompt-based generation and motif scaffolding without fine-tuning mirroring the flexibility of LLMs. ⚡ Efficient Sampling: Orchestrated multi-scale strategies deliver a 2.5x speedup over single-scale baselines. 📈 Favorable"
X Link 2026-02-05T20:23Z [---] followers, [---] engagements

"💡 Motivation: Why haven't traditional AR models worked for proteins [--]. Discretization: Converting 3D atoms to discrete tokens loses structural fidelity. [--]. Unidirectional dependency: Standard AR is one-way but protein residues have strong bidirectional dependencies. [--]. Exposure bias: Trained on ground truth context + inference based on model predictions - training-inference mismatch. https://twitter.com/i/web/status/2019507289229795621 https://twitter.com/i/web/status/2019507289229795621"
X Link 2026-02-05T20:23Z [---] followers, [---] engagements

"✍ Zero-shot Prompted Generation: Because PAR handles multiple granularities a human can specify a coarse layout (e.g. just [--] points) and the model will generate a complete structurally consistent protein backbone. No fine-tuning required"
X Link 2026-02-05T20:23Z [---] followers, [----] engagements

"🧬 Zero-shot Motif Scaffolding: PAR achieves out-of-the-box generalization by teacher-forcing motif coordinates at each scale. Unlike diffusion models it requires no fine-tuning or special masks to produce diverse high-fidelity scaffolds. Detailed benchmarking analysis is available in the appendix https://twitter.com/i/web/status/2019507302995558695 https://twitter.com/i/web/status/2019507302995558695"
X Link 2026-02-05T20:23Z [---] followers, [---] engagements

"🧪 Appendix Experiments: We pushed PAR even further Multi-scale Downsampling: sequence-based downsampling preserves pairwise spatial relationships. Zero-shot Motif-scaffolding: Zero-shot PAR on RFDiffusion benchmark. Long Proteins: We finetuned PAR on proteins up to [---] residues. Self-conditioning: Proven to be complementary to our multi-scale framework for even better quality. Cluster Diversity: Comprehensive fold-diversity analysis"
X Link 2026-02-05T20:23Z [---] followers, [---] engagements

"🤝 Acknowledgement: This is a joint work by Yanru Qu @chengyenhsieh (Project Lead) @zaixiang_zheng @GeLiuSaber and @QuanquanGu from ByteDance Seed and UIUC. Special thanks to our colleagues @Xinyou_NJU Jiasheng Ye Liang Hong @YuningShen1 @li_yilai @dugu9sword Fei YE at ByteDance Seed for their support. https://twitter.com/i/web/status/2019507316366926143 https://twitter.com/i/web/status/2019507316366926143"
X Link 2026-02-05T20:23Z [---] followers, [---] engagements

"🔗 Paper & Project Page: 📄 Paper: 🌐 Project Page: https://par-protein.github.io https://arxiv.org/abs/2602.04883 https://par-protein.github.io https://arxiv.org/abs/2602.04883 https://par-protein.github.io https://arxiv.org/abs/2602.04883 https://par-protein.github.io https://arxiv.org/abs/2602.04883"
X Link 2026-02-05T20:24Z [---] followers, [---] engagements

"🚀 More from ByteDance Seed AI4S: Check out our team's other cutting-edge work: DPLM (@zaixiang_zheng @chengyenhsieh @Xinyou_NJU @jsye588986 Liang Hong): SeedFold (@dugu9sword ): SeedProteo: (Wei Qu). Star-MD: (@nimawrites ). https://bytedance-seed.github.io/ConfRover/starmd https://arxiv.org/abs/2512.24192 https://seedfold.github.io/ https://bytedance.github.io/dplm/ https://seedfold.github.io/ https://bytedance.github.io/dplm/ https://bytedance-seed.github.io/ConfRover/starmd https://arxiv.org/abs/2512.24192 https://seedfold.github.io/ https://bytedance.github.io/dplm/"
X Link 2026-02-05T20:24Z [---] followers, [---] engagements

"@iScienceLuvr cuz ByteDance values research 🧐"
X Link 2026-02-07T05:09Z [---] followers, [---] engagements

"@Waymo @GoogleDeepMind Lets see whether this could bridge Teslas 2D image scaling advantage"
X Link 2026-02-07T05:10Z [---] followers, [--] engagements

"World model for ultrasound medical videos. JEPA seems to have some nice properties e.g. robust to noisy data and data-efficient. EchoJEPA discards whats unpredictable and locks onto what matters clinically: ➡ chamber geometry ➡ wall motion ➡ valve dynamics @ylecun's vision: machines should learn like humans by building internal world models not reconstructing every pixel. We just validated this idea at the largest scale ever attempted in cardiac ultrasound. Introducing EchoJEPA the first world model for medical video. 🫀 https://t.co/DZssCL2UgR @ylecun's vision: machines should learn like"
X Link 2026-02-08T17:40Z [---] followers, [---] engagements

"I feel sad for LeBron. Since ByteDance released Seedance he has been angle-broken by tons of tech bros and little kids 😭"
X Link 2026-02-10T01:31Z [---] followers, [----] engagements

"Wow agentic research is moving to the next stage. Btw Stanford is such a hatching ground of startups Introducing Simile. Simulating human behavior is one of the most consequential and technically difficult problems of our time. We raised $100M from Index Hanabi A* BCV @karpathy @drfeifei @adamdangelo @rauchg @scottbelsky among others. https://t.co/lwEH6a6BYk Introducing Simile. Simulating human behavior is one of the most consequential and technically difficult problems of our time. We raised $100M from Index Hanabi A* BCV @karpathy @drfeifei @adamdangelo @rauchg @scottbelsky among others."
X Link 2026-02-12T21:41Z [---] followers, [---] engagements

"1-step discrete diffusion You like discrete diffusion but it's too slow 🥀 You like test-time inference but it's for continuous methods 😩 We fixed it. Introducing Categorical Flow Maps: continuously sample discrete data in a single step 🚀💫 How 🧵 💪 Co-led with @FEijkelboom @daan_roos_ https://t.co/ET3f3Qvx3I You like discrete diffusion but it's too slow 🥀 You like test-time inference but it's for continuous methods 😩 We fixed it. Introducing Categorical Flow Maps: continuously sample discrete data in a single step 🚀💫 How 🧵 💪 Co-led with @FEijkelboom @daan_roos_"
X Link 2026-02-17T03:30Z [---] followers, [----] engagements

"You like discrete diffusion but it's too slow 🥀 You like test-time inference but it's for continuous methods 😩 We fixed it. Introducing Categorical Flow Maps: continuously sample discrete data in a single step 🚀💫 How 🧵 💪 Co-led with @FEijkelboom @daan_roos_ https://twitter.com/i/web/status/2023412112052158597 https://twitter.com/i/web/status/2023412112052158597"
X Link 2026-02-16T15:00Z [---] followers, 83.5K engagements

"Dario Amodei just said the quiet part out loud: The real AI moats aren't in chatbots. They're in medicine and the physical world. Anyone can wrap a model in a pretty UI. Very few can navigate FDA trials biological complexity and regulatory mazes. The biggest AI companies won't be the ones building addictive apps. They'll be the ones quietly extending human life. This is why Anthropic is betting on Claude in healthcare. Why DeepMind spun off Isomorphic Labs. Why every major lab has a "biology" team now. The consumer AI race is a feature war. The real race is understanding protein structures"
X Link 2026-02-17T02:41Z 17.5K followers, 115.4K engagements

"QT45 A major work from our lab is out in Science @ScienceMag another key step towards a true demonstration of RNA self-replication. https://t.co/2Fe8XonYst (1/6) https://t.co/1E1VJXBUdF A major work from our lab is out in Science @ScienceMag another key step towards a true demonstration of RNA self-replication. https://t.co/2Fe8XonYst (1/6) https://t.co/1E1VJXBUdF"
X Link 2026-02-15T01:53Z [---] followers, [---] engagements

"A major work from our lab is out in Science @ScienceMag another key step towards a true demonstration of RNA self-replication. (1/6) https://www.science.org/doi/10.1126/science.adt2760 https://www.science.org/doi/10.1126/science.adt2760"
X Link 2026-02-13T16:30Z [----] followers, 32.1K engagements

"This is actually insane. 🤯 Seedance [---] basically means were about to make live action versions of our favorite anime ourselves. Fan fiction just evolved into fan cinema"
X Link 2026-02-14T05:26Z 19K followers, 2.1M engagements

"Six patients with complete spinal cord injuries regained movement. A Brazilian scientist worked in silence for [--] years on something medicine said was impossible: regenerating the spinal cord. Dr. Tatiana Sampaio extracted a protein from placentas that acts as "biological glue" recreating the conditions that let embryonic neurons connect. https://t.co/SlaiNQ7C3P A Brazilian scientist worked in silence for [--] years on something medicine said was impossible: regenerating the spinal cord. Dr. Tatiana Sampaio extracted a protein from placentas that acts as "biological glue" recreating the"
X Link 2026-02-14T22:20Z [---] followers, [---] engagements

"A Brazilian scientist worked in silence for [--] years on something medicine said was impossible: regenerating the spinal cord. Dr. Tatiana Sampaio extracted a protein from placentas that acts as "biological glue" recreating the conditions that let embryonic neurons connect. Six patients with complete spinal cord injuries regained movement. Bruno Drummond was tetraplegic after a car accident. Two weeks after treatment he moved his toe. Today he walks climbs stairs dances. Her quote when asked why she finally went public: "I no longer have the right to be conservative." [--] years. No social"
X Link 2026-02-13T23:26Z 17.5K followers, 651.9K engagements

"I love how AI is helping science GPT-5.2 derived a new result in theoretical physics. Were releasing the result in a preprint with researchers from @the_IAS @VanderbiltU @Cambridge_Uni and @Harvard. It shows that a gluon interaction many physicists expected would not occur can arise under specific GPT-5.2 derived a new result in theoretical physics. Were releasing the result in a preprint with researchers from @the_IAS @VanderbiltU @Cambridge_Uni and @Harvard. It shows that a gluon interaction many physicists expected would not occur can arise under specific"
X Link 2026-02-13T21:43Z [---] followers, [---] engagements

"GPT-5.2 derived a new result in theoretical physics. Were releasing the result in a preprint with researchers from @the_IAS @VanderbiltU @Cambridge_Uni and @Harvard. It shows that a gluon interaction many physicists expected would not occur can arise under specific conditions. https://openai.com/index/new-result-theoretical-physics/ https://openai.com/index/new-result-theoretical-physics/"
X Link 2026-02-13T19:19Z 4.6M followers, 4.3M engagements

"The Iso team has cooked something incredible: our new technical report unveils the latest results from our drug design engine the IsoDDE progressing far beyond AlphaFold [--]. This breaks new ground compared to AF and other similar methods by a significant degree across all key benchmarks. 1/7 https://twitter.com/i/web/status/2021170265242173677 https://twitter.com/i/web/status/2021170265242173677"
X Link 2026-02-10T10:31Z 18.4K followers, 160.2K engagements

"Amazing paper from @PhilHolliger https://www.science.org/doi/10.1126/science.adt2760 https://www.science.org/doi/10.1126/science.adt2760"
X Link 2026-02-13T00:25Z 18.6K followers, 58.3K engagements

"Introducing Simile. Simulating human behavior is one of the most consequential and technically difficult problems of our time. We raised $100M from Index Hanabi A* BCV @karpathy @drfeifei @adamdangelo @rauchg @scottbelsky among others"
X Link 2026-02-12T19:00Z 18.9K followers, 2.2M engagements

"RT @demishassabis: The drug design engine were building at @IsomorphicLabs is extending the SOTA further across key benchmarks showing hu"
X Link 2026-02-10T16:40Z [---] followers, [---] engagements

"The drug design engine were building at @IsomorphicLabs is extending the SOTA further across key benchmarks showing huge progress in accuracy and capabilities critical for in-silico drug discovery. Incredible work from @maxjaderberg and the entire team at Isomorphic Labs Today we share a technical report demonstrating how our drug design engine achieves a step-change in accuracy for predicting biomolecular structures more than doubling the performance of AlphaFold [--] on key benchmarks and unlocking rational drug design even for examples it has https://t.co/rTzmHenPWR Today we share a technical"
X Link 2026-02-10T14:03Z 647.9K followers, 358.6K engagements

"Today we share a technical report demonstrating how our drug design engine achieves a step-change in accuracy for predicting biomolecular structures more than doubling the performance of AlphaFold [--] on key benchmarks and unlocking rational drug design even for examples it has never seen before. Head to the comments to read our blog"
X Link 2026-02-10T10:00Z 40.2K followers, 1.3M engagements

"Youre not learning if youre not confused half the time"
X Link 2026-02-09T14:47Z 1.4M followers, 191K engagements

"I feel sad for LeBron. Since ByteDance released Seedance he has been angle-broken by tons of tech bros and little kids 😭"
X Link 2026-02-10T01:31Z [---] followers, [----] engagements

"NEW research from FAIR at Meta Cornell and CMU. This paper is a bigger deal than it seems. Apparently you don't need billions of parameters to teach an AI model to reason. The default approach to post-training language models for reasoning today remains finetuning millions or even billions of parameters. But what if the signal needed for reasoning is far sparser than we assume This new research introduces TinyLoRA a method that scales low-rank adapters down to as few as a single trainable parameter. Using TinyLoRA with RL they trained Qwen2.5-7B to 91% accuracy on GSM8K with only 13"
X Link 2026-02-08T14:48Z 289.4K followers, 51.3K engagements

"@ylecun's vision: machines should learn like humans by building internal world models not reconstructing every pixel. We just validated this idea at the largest scale ever attempted in cardiac ultrasound. Introducing EchoJEPA the first world model for medical video. 🫀 18M echocardiograms 👥 300K patients 🧠 Learns heart dynamics not imaging noise The problem: Ultrasound is messy. Speckle shadows attenuation. Most pretraining objectives end up modeling the scanner not the heart. The idea: Stop reconstructing pixels. Predict latent structure instead. EchoJEPA discards whats unpredictable and"
X Link 2026-02-06T20:01Z 17.5K followers, 90.8K engagements

"RT @DeryaTR_: Another spectacular MedAI model from Bo and his team This is one of the hardest problems in radiology for AI to learn and i"
X Link 2026-02-08T17:30Z [---] followers, [--] engagements

"Another spectacular MedAI model from Bo and his team This is one of the hardest problems in radiology for AI to learn and it would have a massive impact on heart disease diagnosis. Congrats guys @ylecun's vision: machines should learn like humans by building internal world models not reconstructing every pixel. We just validated this idea at the largest scale ever attempted in cardiac ultrasound. Introducing EchoJEPA the first world model for medical video. 🫀 https://t.co/DZssCL2UgR @ylecun's vision: machines should learn like humans by building internal world models not reconstructing every"
X Link 2026-02-06T21:21Z 334.6K followers, 16.1K engagements

"The Midjourney for Antibody Design: Review No Priors Podcast In the No Priors podcast featuring Joshua Meier and Jack Dent from Chai Discovery they discussed how Chai-2 was posed to fuel antibody discoveries. This conversation offers insights on the http://x.com/i/article/2020324004456984587 http://x.com/i/article/2020324004456984587"
X Link 2026-02-08T03:44Z [---] followers, [----] engagements

"Gojo vs Naruto. Saitama vs Genos. Now all of this is possible with Seedance [---]. A Chinese model with almost no censorship and its about to take over. In the coming days youll see your favorite anime characters fighting with studio-level quality spectacular animation and results that outperform Sora [--] and Vidu Q3 in animation. This isnt a one-off test its a shift in the entire landscape. Seedance [---] delivers action acting impact and fluidity like no other model available today. Im genuinely excited for the official release. I cant wait to see how people will use Seedance [---] because this"
X Link 2026-02-07T14:26Z [----] followers, 90.1K engagements

"Excited to share World-Gymnast: Training Robots with RL in a World Model. Training a VLA policy in a world model with RL transfers to much improved real-robot success (according to third-party robot AutoEval). Website: Paper: http://arxiv.org/abs/2602.02454 http://world-gymnast.github.io http://arxiv.org/abs/2602.02454 http://world-gymnast.github.io"
X Link 2026-02-05T18:49Z [----] followers, 41K engagements

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