| 1 | Mar 16, '23 | I got to try GPT-4's multimodal capabilities and it's quite impressive! A quick thread of examples... Let's start out with solving a CAPTCHA, no big deal | 63× | 1.2M | 2.9k | 467 | 630 |
| 2 | Oct 22, '25 | This is actually kinda sad. Using ChatGPT in WhatsApp was a fun hack to use ChatGPT on flights for free. | 53.5× | 1M | 4.9k | 77 | 649 |
| 3 | Feb 17, '25 | Have you heard of Cleo? Cleo was an account on Math Stack Exchange that was infamous for dropping the answer to the most difficult integrals with no explanation... often mere minut | 51.1× | 986k | 5.6k | 320 | 2.5k |
| 4 | May 13, '24 | The livestream demo is not the only cool part about GPT-4o Remember, GPT-4o is an end-to-end trained multimodal model! No one is reading the GPT-4o blog post which highlights so ma | 40.2× | 776k | 1.5k | 143 | 952 |
| 5 | May 12, '23 | Very excited to share the news that I successfully defended my PhD research today! 🥳🎉 After 4 years 8 months in the @UCDavisGrad @UCDavisBMEGG graduate program, I am now Dr. Tani | 28× | 540k | 3.5k | 207 | 149 |
| 6 | Mar 25, '25 | Wait GPT-4o can just one-shot stuff like this?! That's impressive... | 27.3× | 527k | 4.4k | 125 | 1.1k |
| 7 | Sep 19, '25 | This paper claims LLMs are better at selecting successful founders than VCs "We introduce VCBench, the first benchmark for predicting founder success in venture capital (VC)" "most | 27× | 521k | 2.3k | 262 | 1.5k |
| 8 | Jan 25, '25 | DeepSeek CEO's thesis is that in AI, there isn't any moat in being closed-source, but rather in having a talented team that can keep innovating | 25.3× | 488k | 4.9k | 625 | 1.2k |
| 9 | Aug 28, '24 | Diffusion Models Are Real-Time Game Engines abs: project page: Google presents GameNGen, the first game engine powered entirely by a neural model that enables real-time interaction | 24.8× | 479k | 1.9k | 380 | 979 |
| 10 | Dec 28, '22 | Are you wondering how large language models like ChatGPT and InstructGPT actually work? One of the secret ingredients is RLHF - Reinforcement Learning from Human Feedback. Let's di | 24.1× | 465k | 2.5k | 504 | 1.6k |
| 11 | Oct 1, '25 | I unfortunately don't know much about mobile app and desktop software dev so I'm genuinely asking: WHY DO ALL COMPANIES RELEASE iOS/macOS PRODUCTS FIRST/ONLY??? How hard is it to r | 23.9× | 461k | 1.6k | 61 | 401 |
| 12 | Feb 17, '25 | Large Language Diffusion Models Introduces LLaDA-8B, a large language diffusion model that pretrained on 2.3 trillion tokens using 0.13 million H800 GPU hours, followed by SFT on 4 | 23.5× | 454k | 1.7k | 262 | 1.3k |
| 13 | Dec 1, '24 | how does someone solve Advent of Code problem in 9 seconds??!! | 22.7× | 438k | 2.5k | 53 | 361 |
| 14 | Nov 12, '25 | A NEW PAPER FROM YANN LECUN: LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics This could be one of LeCun's last papers at Meta (lol), but it's a really | 21.8× | 420k | 1.8k | 251 | 1.4k |
| 15 | Dec 10, '24 | Training Large Language Models to Reason in a Continuous Latent Space Introduces a new paradigm for LLM reasoning called Chain of Continuous Thought (COCONUT) Extremely simple chan | 19.6× | 378k | 1.9k | 289 | 1.7k |
| 16 | Jan 22, '25 | the contrast lol | 19.3× | 372k | 11k | 276 | 392 |
| 17 | Nov 20, '23 | "feel the AGI" | 18.6× | 358k | 879 | 93 | 175 |
| 18 | Feb 26, '25 | Diffusion language models are SO FAST!! A new startup, Inception Labs, has released Mercury Coder, "the first commercial-scale diffusion large language model" It's 5-10x faster tha | 18.4× | 354k | 2.4k | 248 | 1.8k |
| 19 | Jan 6, '25 | US has teams building frontier models (OpenAI, Anthropic, etc.) Europe has teams building frontier models (Mistral, DeepMind, etc.) China has teams building frontier models (DeepSe | 18.1× | 350k | 2.8k | 150 | 727 |
| 20 | Feb 10, '25 | Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach We study a novel language model architecture that is capable of scaling test-time computation by impl | 18.1× | 349k | 1.2k | 187 | 1.1k |
| 21 | Nov 5, '23 | xAI: Announcing Grok Grok-1 model card: The first model, Grok-0, was a 33B autoregressive LLM, and approached Llama-2-70b with half the training resources. Grok-1 surpasses that, a | 18× | 348k | 884 | 133 | 276 |
| 22 | Jan 20, '25 | Okay so this is so far the most important paper in AI of the year | 17.9× | 346k | 3k | 327 | 2k |
| 23 | Feb 28, '23 | Claude, @AnthropicAI's powerful ChatGPT alternative, was trained with "Constitutional AI". Constitutional AI is particularly interesting since it uses less human feedback than othe | 16.6× | 321k | 885 | 113 | 794 |
| 24 | Sep 24, '25 | SimpleFold: Folding Proteins is Simpler than You Think "we introduce SimpleFold, the first flow-matching based protein folding model that solely uses general purpose transformer bl | 14.9× | 288k | 1.3k | 135 | 911 |
| 25 | Oct 20, '25 | DeepSeek released an OCR model today. Their motivation is really interesting: they want to use visual modality as an efficient compression medium for textual information, and use t | 14.4× | 278k | 1.1k | 138 | 460 |
| 26 | Jan 29, '25 | Anyone who thinks DeepSeek just came out of nowhere should see this graph. For each model on this graph, weights, code, and detailed papers were released. This is a team with a str | 13.9× | 269k | 3.6k | 622 | 1k |
| 27 | Dec 26, '23 | Are you wondering how the new Mamba language model works? Mamba is based on state-space models (SSMs), a new competitor to the Transformer architecture. Here are 5 resources to hel | 11.7× | 226k | 1.2k | 183 | 1.8k |
| 28 | Dec 9, '24 | A new tutorial on RL by Kevin Patrick Murphy, a Research Scientist at Google DeepMind who also wrote several comprehensive, well-regarded textbooks on ML/DL. This ought to be a goo | 11.6× | 224k | 2.3k | 261 | 3.4k |
| 29 | Oct 22, '24 | Some people need to stop being so math-brained when writing their papers... Using 10 pages of math to explain 10 lines of code is super annoying. | 11.5× | 223k | 1.2k | 38 | 120 |
| 30 | May 20, '25 | Google is releasing a diffusion language model let's goooooooo! | 11.5× | 222k | 1.7k | 87 | 223 |
| 31 | Jan 2, '25 | Nostalgic for GPT-3 and its absolutely hilarious generated greentexts... | 11.4× | 221k | 6.2k | 202 | 565 |
| 32 | Mar 17, '24 | xAI releases Grok-1 blog: code: Base model trained on a large amount of text data, not fine-tuned for any particular task. 314B parameter Mixture-of-Experts model with 25% of the w | 11.3× | 219k | 1.1k | 253 | 397 |
| 33 | Mar 31, '25 | I have EXCITING news: I've started a company! Introducing Sophont We’re building open multimodal foundation models for the future of healthcare. We need a DeepSeek for medical AI, | 11.2× | 217k | 1.2k | 107 | 308 |
| 34 | Feb 28, '25 | Thinking about creating a medical AI group chat here on Twitter... If you're a researcher/engineer/clinician/etc. working in AI or medical AI, let me know if you're interested in j | 11× | 212k | 1.7k | 78 | 487 |
| 35 | Jan 23, '25 | Yesterday was my last day at Stability AI. I had a great time over the past 2.5 years working with amazing colleagues on the cutting edge of AI research and development but now it | 10.2× | 197k | 1.4k | 37 | 226 |
| 36 | Feb 4, '25 | Language Models Use Trigonometry to Do Addition "We first discover that numbers are represented in these LLMs as a generalized helix, which is strongly causally implicated for the | 10.1× | 195k | 2.3k | 332 | 1.9k |
| 37 | Mar 27, '25 | "A Manga Guide to DeepSeek-V3 Technical Report" from now on this is how I will post all papers 🤣 | 9.8× | 190k | 1.8k | 167 | 793 |
| 38 | Feb 8, '25 | Meta researchers used AI to predict the text a person was typing just from non-invasive brain recording! With EEG, their "Brain2Qwerty" model gets 67% of the characters wrong, but | 9.4× | 182k | 1.4k | 243 | 868 |
| 39 | Feb 4, '25 | It's so over for Ireland | 9.4× | 181k | 3.9k | 83 | 205 |
| 40 | Nov 18, '23 | View post on X | 9.3× | 179k | 977 | 150 | 124 |
| 41 | Oct 21, '25 | Maybe computer vision is all you need lol To be fair that's how humans work, it isn't like language is a separate sense lol | 9.3× | 179k | 846 | 56 | 250 |
| 42 | Dec 10, '24 | The inventors of flow matching have released a comprehensive guide going over the math & code of flow matching! Also covers variants like non-Euclidean & discrete flow matc | 9.0× | 174k | 1.7k | 272 | 2k |
| 43 | Dec 12, '23 | Spotted at #NeurIPS2023: Disney Research demos a RL-trained remote control robot! Super impressive work! | 8.8× | 170k | 1.1k | 140 | 149 |
| 44 | Sep 8, '25 | Having seen a few different blog posts like this, I am convinced that with a small team and limited funding it's possible to build a half decent, actually useful, text search engin | 8.4× | 162k | 1.1k | 68 | 1.2k |
| 45 | Nov 16, '22 | I will attempt to explain the basic idea of how diffusion models work! ... in only 15 tweets! 😲 Let's get started ↓ | 8.4× | 0 | 3.8k | 668 | 2.8k |
| 46 | Dec 13, '24 | Microsoft Phi-4 is announced! It's a 14B parameter LM trained heavily on synthetic data, with very strong performance, even exceeding GPT-4o on GPQA and MATH benchmarks! Currently | 8.4× | 161k | 1.3k | 181 | 491 |
| 47 | Feb 17, '25 | Vladimir confirmed to an investigator he indeed was Cleo, providing more details about the whole saga. He was coming up with various hard integrals to solve and he had an idea of w | 8.3× | 161k | 1.2k | 18 | 86 |
| 48 | Nov 9, '25 | Recently, a tutorial/textbook on diffusion models was released on arXiv. Looks like a great resource, starts from the beginning, explaining diffusion models from variational, score | 8.1× | 157k | 1.2k | 180 | 1.3k |
| 49 | Dec 22, '24 | People are discovering how autoregressive image generation works lol | 7.9× | 152k | 1.4k | 22 | 157 |
| 50 | Feb 6, '25 | If someone says RL is so simple and easy, ask them if they've read these blog posts 😄 | 7.6× | 148k | 1.5k | 149 | 2k |
| 51 | Sep 15, '25 | so they lied to us 😭😭😭 | 7.6× | 146k | 2.2k | 28 | 68 |
| 52 | Apr 30, '24 | Google announces Med-Gemini, a family of Gemini models fine-tuned for medical tasks! 🔬 Achieves SOTA on 10 of the 14 benchmarks, spanning text, multimodal & long-context applicati | 7.4× | 142k | 1k | 200 | 745 |
| 53 | May 8, '24 | The devil works fast but lucidrains works faster | 6.8× | 132k | 1.2k | 73 | 218 |
| 54 | May 10, '25 | I would pay hundreds of dollars per month to cut down or eliminate the amount of sleep my body requires more people need to be working on this fr | 6.7× | 129k | 989 | 43 | 98 |
| 55 | Feb 19, '25 | YOLO v12???? you've got to be kidding me, why is there a new YOLO model like every 6 months lol | 6.6× | 127k | 1.2k | 64 | 317 |
| 56 | Dec 15, '24 | View post on X | 6.6× | 127k | 2.1k | 166 | 623 |
| 57 | Mar 14, '23 | GPT-4 release Med-PaLM2 announcement PaLM API release Claude API release | 6.5× | 125k | 1.1k | 137 | 115 |
| 58 | Jan 20, '25 | I appreciate DeepSeek providing examples of failure, especially since these are ideas that have been widely discussed for achieving o1-style models. This is very rare to see in AI | 6.4× | 124k | 1.4k | 138 | 502 |
| 59 | Dec 17, '24 | Every day, you guys... 🤣🤣🤣 | 6.4× | 123k | 2.9k | 101 | 70 |
| 60 | Dec 28, '24 | This YouTube video by @welchlabs is a very approachable, concise explanation of LLM mechanistic interpretability and sparse autoencoders (SAEs). I highly recommend checking it out! | 6.3× | 122k | 1.8k | 208 | 2k |
| 61 | Jul 4, '25 | do you guys remember this lol | 6.2× | 120k | 2.6k | 36 | 149 |
| 62 | Oct 12, '25 | Wow, I just reached a new milestone: 1000 CITATIONS! 🥳 I've been waiting for this one... Onto publishing more research! | 6.0× | 116k | 1.3k | 16 | 80 |
| 63 | Apr 27, '22 | Awesome and surprising things you can do with Jupyter Notebooks ⬇ | 6.0× | 0 | 3k | 586 | 1.8k |
| 64 | Apr 26, '24 | The @PyTorch team is developing a library for large model training called torchtitan 👀 They have scripts to train Llama-3 from scratch The library went public today on GitHub but | 5.9× | 115k | 1.1k | 189 | 615 |
| 65 | Mar 14, '25 | um what 😭 | 5.8× | 112k | 1.3k | 6 | 54 |
| 66 | Mar 20, '25 | Course material for an MIT class "Introduction to Flow Matching and Diffusion Models", looks great if you want a principled and hands on understanding of diffusion models/flow matc | 5.4× | 103k | 1.5k | 212 | 1.9k |
| 67 | Dec 15, '21 | This is currently the most important network in deep learning! From helping to power search for billions of users to better understanding proteins, it does it all! Here are 10 of t | 5.3× | 0 | 2.5k | 549 | 1.6k |
| 68 | Jan 28, '25 | Part of the reason the public is shocked is that most of them have only played with ChatGPT 4o (in the free plan) so when they try a reasoning model like R1 they think China has ma | 5.2× | 101k | 1.4k | 76 | 113 |
| 69 | Jun 19, '25 | Is there an RL group chat where people talk about latest methods, papers, practitioner advice, etc.? If not maybe I should make one? | 5.2× | 100k | 890 | 15 | 356 |
| 70 | May 16, '25 | They're blaming it on a single rogue employee lol | 5.1× | 99k | 1.5k | 20 | 46 |
| 71 | Aug 15, '25 | I can't believe Grok Imagine is this bad... This is like 2023 level of quality... The xAI team definitely can do better 😭 And why is Elon proudly sharing this, come on... | 5.0× | 96k | 1k | 27 | 37 |
| 72 | Apr 9, '25 | He is one of the best AI researchers of all time. A good reminder that you shouldn't over-index on signals like GitHub contributions | 4.7× | 91k | 1.6k | 55 | 204 |
| 73 | Mar 15, '25 | A new YouTube video by Welch Labs that gives an awesome walk-through of how multi-latent attention works (one of the innovations of DeepSeek)! Check it out! | 4.7× | 91k | 1.6k | 192 | 1.3k |
| 74 | Jun 9, '25 | I don't even agree with the Apple paper but this is an extremely midwit take | 4.6× | 89k | 1.6k | 27 | 102 |
| 75 | Aug 23, '21 | Just discovered this AMAZING website - "Deep Learning Drizzle" It is a constantly-updated list of machine learning/deep learning course materials that are taught by domain experts | 4.6× | 0 | 2.1k | 663 | 1.2k |
| 76 | Dec 10, '24 | [MASK] is All You Need New paper from CompVis group, introduces a new method called Discrete Interpolants that builds on top of discrete flow matching, and connects it to masked au | 4.5× | 87k | 1.1k | 127 | 744 |
| 77 | Aug 12, '25 | RLVR/RLHF libraries: • verl - ByteDance • TRL - HuggingFace • slime - Zhipu AI • prime-rl - Prime Intellect • ROLL - Alibaba • Nemo-RL - NVIDIA • AReaL - Ant Research • SkyRL - UC | 4.2× | 82k | 981 | 105 | 1.1k |
| 78 | Jun 2, '25 | How much do language models memorize? "We formally separate memorization into two components: unintended memorization, the information a model contains about a specific dataset, an | 4.1× | 79k | 1.1k | 163 | 989 |
| 79 | May 31, '22 | Have you seen #dalle2 and #Imagen and wondered how it works? Both models utilize diffusion models, a new class of generative models that have overtaken GANs in terms of visual qual | 4.0× | 0 | 1.9k | 388 | 1.3k |
| 80 | Jun 20, '25 | This is pretty cool, a DeepSeek researcher open-sourced "nano-vLLM", a lightweight implementation of vLLM in ~1,200 lines of Python code benchmark with Qwen3-0.6B matches throughpu | 4.0× | 77k | 1k | 135 | 848 |
| 81 | Nov 4, '25 | Google DeepMind release: Towards Robust Mathematical Reasoning Introduces IMO-Bench, a suite of advanced reasoning benchmarks that played a crucial role in GDM's IMO-gold journey. | 3.8× | 73k | 937 | 146 | 504 |
| 82 | Oct 6, '25 | It's actually insane that ChatGPT has hit 800 million weekly active users. That's 10% of the entire world's population! 🤯 | 3.3× | 64k | 895 | 48 | 49 |
| 83 | Aug 26, '22 | Research in AI is surprisingly more accessible to people with different backgrounds compared to other fields. Anyone (w/ relevant experience) can contribute to impactful research. | 3.3× | 0 | 1.5k | 320 | 1k |
| 84 | Apr 16, '25 | I am telling you guys if you really want to truly grasp diffusion models you MUST read all of @sedielem's blog posts!!! | 3.3× | 63k | 1k | 113 | 1.6k |
| 85 | Sep 29, '22 | ICLR 2023 (a top ML/AI conference) submissions have been released, and do you know what that means? Time for mind-blowing papers! 🤯↓ | 3.2× | 0 | 1.7k | 342 | 833 |
| 86 | Jul 21, '25 | Kimi K2 paper dropped! describes: - MuonClip optimizer - large-scale agentic data synthesis pipeline that systematically generates tool-use demonstrations via simulated and real-wo | 3.1× | 60k | 949 | 163 | 577 |
| 87 | Feb 18, '25 | @12exyz tbf o3 isn't public and those benchmarks can't be verified yet so I think it's understandable that they haven't included it yet | 3.1× | 60k | 1.5k | 7 | 21 |
| 88 | Apr 8, '24 | One of my favorite YouTubers, @3blue1brown, has put out an incredible video explainer about the attention mechanism! I highly recommend checking it out! | 2.9× | 55k | 1.1k | 129 | 425 |
| 89 | Jan 13, '25 | born just in time to build ChatGPT wrappers | 2.6× | 50k | 998 | 65 | 53 |
| 90 | Aug 30, '21 | After you train a machine learning model, the BEST way to showcase it to the world is to make a demo for others to try your model! Here is a quick thread🧵on two of the easiest way | 2.5× | 0 | 1.3k | 222 | 746 |
| 91 | Aug 31, '22 | So @StableDiffusion has various options and controls and one of the main ones is the sampler used for generation. Let's talk a little bit about these samplers since this has some i | 2.4× | 0 | 1.4k | 220 | 619 |
| 92 | Feb 4, '22 | What matters most when training a neural network is how well it generalizes to unseen data. For neural networks, it turns out there's a simple principle that can allow you to under | 2.3× | 0 | 1.3k | 245 | 612 |
| 93 | Dec 18, '23 | Perspectives on the State and Future of Deep Learning -- 2023 abs: Leading AI researchers write about their thoughts on the future of deep learning | 2.3× | 44k | 344 | 108 | 254 |
| 94 | May 20, '25 | Google coming after OpenAI, Meta, Apple, pretty much everyone today | 2.3× | 43k | 1.4k | 50 | 60 |
| 95 | Sep 5, '24 | Ilya: | 2.2× | 43k | 885 | 23 | 24 |
| 96 | Aug 16, '21 | Annotated PyTorch Paper Implementations by @labmlai is an AMAZING resource: • Deep learning papers explained in-depth with code side-by-side • Constantly updated with some of the l | 1.9× | 0 | 1k | 216 | 509 |
| 97 | Aug 20, '21 | The Tesla team discussed how they are using AI to crack Full Self Driving (FSD) at their Tesla AI Day event. They introduced many cool things: - HydraNets - Dojo Processing Units - | 1.8× | 0 | 1.3k | 236 | 271 |
| 98 | Apr 6, '22 | How the machine learning community feels after PaLM and DALL·E 2 during this week: | 1.6× | 0 | 1.6k | 155 | 98 |
| 99 | Apr 5, '22 | I had a great time chatting with @karpathy yesterday! We discussed a range of topics, from synthetic biology to ML conferences. What's funny is that we are about 2 hrs from each ot | 1.4× | 0 | 1.7k | 41 | 43 |
| 100 | Mar 26, '21 | These new #DeepLearning models are getting huge!😅 | 1.0× | 0 | 988 | 116 | 11 |
@karpathy
@levelsio
@paulg
@naval
@marclou
@illyism