41554 blogs · [ { "id": "01a087df-e7ad-73cc-873b-85a64c19387a", "title": "Releasing StreamHope - Merge Duplicate Playlists in YouTube Music, transfer from Spotify", "url": "https://shital.com/blog/streamhop-release/", "published_at": "2025-12-13T05:30:07+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c861381fb9", "title": "How to Get and Interpret GPU Memory Profiling", "url": "https://shital.com/blog/gpu-memory-profiling/", "published_at": "2025-12-06T23:17:55+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86173c794", "title": "o3-mini Level Hard Problem", "url": "https://shital.com/blog/tweets/post/202502011036-o3-mini-prevails/", "published_at": "2025-02-01T10:36:40+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8625e3684", "title": "DeepSeek's Recipe Evolution", "url": "https://shital.com/blog/tweets/post/202501272301-deepseek-recipe-cot-2line/", "published_at": "2025-01-27T23:01:25+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86306e7cd", "title": "Can You Spot Error Proving 1 = -1?", "url": "https://shital.com/blog/tweets/post/202501120513-11yo-proof-1-equals-minus-1/", "published_at": "2025-01-12T05:13:47+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8639b6153", "title": "Phi-4 Now On HuggingFace with MIT Licence!", "url": "https://shital.com/blog/tweets/thread/202501082332-phi-4-official-release/", "published_at": "2025-01-08T23:32:21+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8648e95b8", "title": "Phi-4: The Most Powerful smol Model!", "url": "https://shital.com/blog/tweets/thread/202412131104-microsoft-phi-4-smol-marvel/", "published_at": "2024-12-13T11:04:46+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c864e89425", "title": "Announcement: PhD Internship for AI for Reasoning", "url": "https://shital.com/blog/tweets/thread/202412110820-summer-of-ai-microsoft-phd-interns/", "published_at": "2024-12-11T08:20:21+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c865289b52", "title": "Some Hand Wavy Estimates on FLOPs and Computational Reducibility", "url": "https://shital.com/blog/tweets/thread/202412091102-ai-speed-limit-putnam/", "published_at": "2024-12-09T11:02:57+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8654b8b98", "title": "Erdős Open Problems as ASI Benchmark", "url": "https://shital.com/blog/tweets/thread/202411291040-ai-vs-erdos-ultimate-math-showdown/", "published_at": "2024-11-29T10:40:57+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c865a8f851", "title": "Tokenization and Decoding", "url": "https://shital.com/blog/tweets/thread/202411212010-bandaids-cant-fix-transformers/", "published_at": "2024-11-21T20:10:22+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c865bf6035", "title": "Scaling Laws and Data Wall", "url": "https://shital.com/blog/tweets/thread/202411150043-scaling-token-quality/", "published_at": "2024-11-15T00:43:32+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8669de300", "title": "OpenCoder", "url": "https://shital.com/blog/tweets/thread/202411110738-opencoder-outsmarts-stackv2/", "published_at": "2024-11-11T07:38:36+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c866f60cc5", "title": "The WSD Learning Rate Schedule", "url": "https://shital.com/blog/tweets/post/202411061656-cooling-down-learning-rate-wsd-schedule/", "published_at": "2024-11-06T16:56:20+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8676b0e99", "title": "Healthier Flossing with Silk!", "url": "https://shital.com/blog/tweets/post/202411031601-switch-to-silk/", "published_at": "2024-11-03T16:01:35+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8680fe159", "title": "Tricky Validation Plot", "url": "https://shital.com/blog/tweets/post/202410271015-lost-in-validation/", "published_at": "2024-10-27T10:15:07+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c868445017", "title": "DeepMind's Chess Model", "url": "https://shital.com/blog/tweets/thread/202410210731-chess-not-checkers-deepmind-ai/", "published_at": "2024-10-21T07:31:29+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c868d8fbc7", "title": "Solving Chicken-Egg Dilemma", "url": "https://shital.com/blog/tweets/post/202410210434-cracking-chicken-egg-dilemma/", "published_at": "2024-10-21T04:34:27+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8696886ec", "title": "New Optimizer: Muon", "url": "https://shital.com/blog/tweets/thread/202410131001-adamw-who-new-optimizer/", "published_at": "2024-10-13T10:01:24+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c869942a14", "title": "Vocab Scaling", "url": "https://shital.com/blog/tweets/post/202410011628-oversized-model-undersized-vocab/", "published_at": "2024-10-01T16:28:40+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86a92013a", "title": "Using Cursor", "url": "https://shital.com/blog/tweets/thread/202409271839-cracking-the-cursor-code/", "published_at": "2024-09-27T18:39:34+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86b131c4e", "title": "Terence Tao Grades o1", "url": "https://shital.com/blog/tweets/post/202409140853-terence-tao-ai-grading/", "published_at": "2024-09-14T08:53:49+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86b6211db", "title": "Things on Big Bang Theory Writer's Table", "url": "https://shital.com/blog/tweets/thread/202408042009-big-bang-clutter-challenge/", "published_at": "2024-08-04T20:09:05+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86bc81527", "title": "Inference Scaling", "url": "https://shital.com/blog/tweets/post/202408020413-100x-larger-models-not-enough/", "published_at": "2024-08-02T04:13:08+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86cc03482", "title": "IMO 2024 Test", "url": "https://shital.com/blog/tweets/thread/202407220853-frontier-models-vs-mathletes/", "published_at": "2024-07-22T08:53:37+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86d3a5ac8", "title": "LLM Judges Prefers Fluff?", "url": "https://shital.com/blog/tweets/post/202407200414-llm-judges-prefer-fluff-just-like-humans/", "published_at": "2024-07-20T04:14:49+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86d84a7a6", "title": "Announcement: Phi-3 14B Rivaling Llama3 70B", "url": "https://shital.com/blog/tweets/post/202405221045-phi3-14b-challenges-llama3-70b/", "published_at": "2024-05-22T10:45:21+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86dc85418", "title": "Power of Inference Scaling", "url": "https://shital.com/blog/tweets/post/202405080950-go-big-or-go-home-scaling-up-go-models/", "published_at": "2024-05-08T09:50:22+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86ddef18f", "title": "Human Hours to FLOPs", "url": "https://shital.com/blog/tweets/thread/202405050823-100-tflops-vs-2-hours/", "published_at": "2024-05-05T08:23:04+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86e7948e6", "title": "Sample Efficiency", "url": "https://shital.com/blog/tweets/post/202404291142-paying-attention-sample-efficiency/", "published_at": "2024-04-29T11:42:20+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86e8f0fc2", "title": "Phi-3: \"ChatGPT\" Natively on Your Phone", "url": "https://shital.com/blog/tweets/thread/202404240042-phi3-chatgpt-phone/", "published_at": "2024-04-24T00:42:47+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86f46e4dd", "title": "Inference Memory", "url": "https://shital.com/blog/tweets/thread/202404220422-inference-gradient-free-diet/", "published_at": "2024-04-22T04:22:02+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86fa2e022", "title": "Llama3 Leaps Ahead with Data Alone", "url": "https://shital.com/blog/tweets/thread/202404191425-llama3-leaps-ahead/", "published_at": "2024-04-19T14:25:10+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c86fea51bd", "title": "Pretraining has Gotten 10x Faster in Past Year!", "url": "https://shital.com/blog/tweets/post/202404171019-llm-training-10x-faster/", "published_at": "2024-04-17T10:19:14+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c87096fc95", "title": "Great Papers", "url": "https://shital.com/blog/tweets/post/202404152027-does-your-paper-inspire/", "published_at": "2024-04-15T20:27:15+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c870ee67e0", "title": "Marvels of ENIAC", "url": "https://shital.com/blog/tweets/thread/202404151959-eniac-tubular-triumph/", "published_at": "2024-04-15T19:59:24+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8717dfcbb", "title": "Intelligence and FLOPs Efficiency", "url": "https://shital.com/blog/tweets/thread/202404141641-chess-engines-slow-motion-rematch/", "published_at": "2024-04-14T16:41:59+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8726a2dd8", "title": "CIFAR-10 Training to 94% Accuracy in 3.29s", "url": "https://shital.com/blog/tweets/post/202404051413-blazing-fast-cifar10-training/", "published_at": "2024-04-05T14:13:15+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c87327b833", "title": "Matmul on CPUs", "url": "https://shital.com/blog/tweets/post/202404011521-from-flop-to-flops/", "published_at": "2024-04-01T15:21:13+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8738db732", "title": "Using rusage", "url": "https://shital.com/blog/tweets/thread/202403271630-rusage-llama-memory/", "published_at": "2024-03-27T16:30:55+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c8744be2ea", "title": "When Uniform Isn't Uniform: A Bug Hunt Tale", "url": "https://shital.com/blog/tweets/thread/202403221524-non-uniform-bug-hunt/", "published_at": "2024-03-22T15:24:20+00:00" }, { "id": "01a087df-e7ae-72ea-9250-e7c87546b6b9", "title": "GPT-4 Training Time", "url": "https://shital.com/blog/tweets/thread/202403190400-gpt4-fitness-plan/", "published_at": "2024-03-19T04:00:02+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840df6583fa", "title": "Scarce to Plentiful", "url": "https://shital.com/blog/tweets/post/202403170650-machines-clock-in-scarcity-out/", "published_at": "2024-03-17T06:50:23+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e03fdb27", "title": "A Lifetime in 1T Tokens", "url": "https://shital.com/blog/tweets/post/202402221235-lifetime-in-1t-tokens/", "published_at": "2024-02-22T12:35:39+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e0ea9af0", "title": "Power of Two Choices", "url": "https://shital.com/blog/tweets/thread/202402081658-balls-in-bins-problem/", "published_at": "2024-02-08T16:58:00+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e1a8fdc0", "title": "PyTorch Distributed Debugging Tips", "url": "https://shital.com/blog/tweets/post/202401221918-pytorch-gold-rush-debugging-tips/", "published_at": "2024-01-22T19:18:53+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e24a8fa0", "title": "EMA Tuning", "url": "https://shital.com/blog/tweets/post/202401192120-paperception-post-hoc-ema-tuning/", "published_at": "2024-01-19T21:20:35+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e2b22d1e", "title": "Phixtral Fusion", "url": "https://shital.com/blog/tweets/thread/202401151722-phixtral-fusion/", "published_at": "2024-01-15T17:22:19+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e362dd23", "title": "Phi-2: The Most Powerful 2.7B Model Yet", "url": "https://shital.com/blog/tweets/post/202401070502-phi-2-unleashed/", "published_at": "2024-01-07T05:02:54+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e3807de5", "title": "Amazing Robotics Demo with Cheap Hardware", "url": "https://shital.com/blog/tweets/thread/202401041635-robot-on-a-budget/", "published_at": "2024-01-04T16:35:45+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e469ad4d", "title": "Hyperparams for 23 Architectures", "url": "https://shital.com/blog/tweets/thread/202312301641-model-mayhem-23-architectures/", "published_at": "2023-12-30T16:41:32+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e4eda546", "title": "Large Activation Outliers", "url": "https://shital.com/blog/tweets/thread/202312251915-llm-outliers-not-emergent/", "published_at": "2023-12-25T19:15:37+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e578dbde", "title": "One Percent, One Week", "url": "https://shital.com/blog/tweets/post/202312171919-one-percent-one-week/", "published_at": "2023-12-17T19:19:30+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e6500e74", "title": "Phi-2: Smol Is More Powerful", "url": "https://shital.com/blog/tweets/thread/202312131822-phi-2-when-less-is-more-powerful/", "published_at": "2023-12-13T18:22:51+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e6b7a0d7", "title": "Looped Transformers", "url": "https://shital.com/blog/tweets/thread/202312121802-one-instruction-looped-transformers/", "published_at": "2023-12-12T18:02:12+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e6fc075d", "title": "Mamba Paper", "url": "https://shital.com/blog/tweets/thread/202312092042-mamba-vs-attention/", "published_at": "2023-12-09T20:42:55+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e7d0bed4", "title": "Thermodynamics of Prediction", "url": "https://shital.com/blog/tweets/thread/202312062303-whats-the-universe-computing/", "published_at": "2023-12-06T23:03:46+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e7d9f00f", "title": "Scaling MLPs Paper", "url": "https://shital.com/blog/tweets/post/202312031849-when-images-go-flat-mlps-step-up/", "published_at": "2023-12-03T18:49:44+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e8c3ee17", "title": "Matmul and Strassen's Algo", "url": "https://shital.com/blog/tweets/post/202311261831-matmul-powers-two-strassen/", "published_at": "2023-11-26T18:31:28+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e91eff0a", "title": "LLMs and Planning", "url": "https://shital.com/blog/tweets/post/202311241916-llms-self-play-planning/", "published_at": "2023-11-24T19:16:46+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e9217c38", "title": "Announcement: PhD Internship for LLM Evaluation", "url": "https://shital.com/blog/tweets/post/202311221218-phd-internship-unbearable-lightness-of-evaluation/", "published_at": "2023-11-22T12:18:38+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840e9bb2ead", "title": "Phi-2: Smol but Mighty", "url": "https://shital.com/blog/tweets/post/202311160232-phi2-smol-mighty-msr/", "published_at": "2023-11-16T02:32:26+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840ea59b6a6", "title": "Thankuritto", "url": "https://shital.com/blog/tweets/post/202311090324-gobble-n-roll-thankuritto/", "published_at": "2023-11-09T03:24:40+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840eb4e51b6", "title": "MathVista Benchmark", "url": "https://shital.com/blog/tweets/post/202310181523-mathvista-gpt4v-visual-weaknesses/", "published_at": "2023-10-18T15:23:21+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840ec4a9724", "title": "Loss Scaling in FP16 Training", "url": "https://shital.com/blog/tweets/thread/202310151451-gradients-gone-wild/", "published_at": "2023-10-15T14:51:51+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840ed423e49", "title": "Stability's 3B Model Training Report", "url": "https://shital.com/blog/tweets/post/202310020548-stability-3b-model-wandb/", "published_at": "2023-10-02T05:48:56+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840ed9251a1", "title": "Architecture vs Auto-regressive Objective", "url": "https://shital.com/blog/tweets/post/202309171738-linear-thinking-simple-nets/", "published_at": "2023-09-17T17:38:25+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840ee484a06", "title": "Phi-1: The Tiny Model Outsmarting Giants", "url": "https://shital.com/blog/tweets/thread/202309121512-phi-1-tiny-model-outsmarting-giants/", "published_at": "2023-09-12T15:12:58+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840eef40746", "title": "Reasoning AWOL? Paper Claims Only ICL Emerged", "url": "https://shital.com/blog/tweets/post/202309071200-reasoning-awol-only-icl-emerged/", "published_at": "2023-09-07T12:00:28+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840ef3684e8", "title": "Weight Averaging: Have Your Cake and Save Your Model", "url": "https://shital.com/blog/tweets/post/202309041906-weight-averaging-save-your-model/", "published_at": "2023-09-04T19:06:12+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840ef7480f6", "title": "Phi-nomenal Code Gen: Phi-1-Base Beats August Models", "url": "https://shital.com/blog/tweets/thread/202309031019-phi-nomenal-code-gen/", "published_at": "2023-09-03T10:19:32+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840f0264645", "title": "Weaving Longer Contexts: Yarn Scaling for LLMs", "url": "https://shital.com/blog/tweets/post/202309020836-yarn-scaling-for-llms/", "published_at": "2023-09-02T08:36:32+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840f03d63ad", "title": "Tensor Shape-Shifting: See Shapes Before Values", "url": "https://shital.com/blog/tweets/post/202309011817-tensor-shape-shifting/", "published_at": "2023-09-01T18:17:05+00:00" }, { "id": "01a087df-e7af-73d7-9cb2-4840f054b8a3", "title": "Tiny Weights, Mighty Models: The Mystery of Weight Decay", "url": "https://shital.com/blog/tweets/thread/202308291720-tiny-weights-mighty-models/", "published_at": "2023-08-29T17:20:55+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f2951c4408", "title": "Conquering FP16 Frustrations: One-Liner to FP8 Gold on H100s", "url": "https://shital.com/blog/tweets/post/202308241302-fp16-to-fp8-one-liner/", "published_at": "2023-08-24T13:02:49+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f296132d0c", "title": "Creativity on Sale, Reasoning Still Full Price", "url": "https://shital.com/blog/tweets/post/202308142053-creativity-on-sale-reasoning-still-full-price/", "published_at": "2023-08-14T20:53:23+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f296447f5d", "title": "Out-Knuthing Knuth: The AGI Test", "url": "https://shital.com/blog/tweets/post/202308131612-out-knuthing-knuth-agi-test/", "published_at": "2023-08-13T16:12:49+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f2966f6959", "title": "The Great 5G Disguise", "url": "https://shital.com/blog/tweets/post/202308131152-great-5g-disguise/", "published_at": "2023-08-13T11:52:38+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29731bd84", "title": "Learning Rates Without Fortune Telling", "url": "https://shital.com/blog/tweets/thread/202308061627-learning-rates-without-fortune-telling/", "published_at": "2023-08-06T16:27:44+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f298185c5f", "title": "LLAMA's Math Leap: From 11% to 49% with Scaling", "url": "https://shital.com/blog/tweets/thread/202308060656-llama-math-leap-scaling/", "published_at": "2023-08-06T06:56:08+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f298654388", "title": "Descartes in Motion: God, Matter, and Not a Moment's Rest", "url": "https://shital.com/blog/tweets/post/202308051354-descartes-in-motion/", "published_at": "2023-08-05T13:54:49+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29879a136", "title": "UnIVAL: Beyond Words and Pictures", "url": "https://shital.com/blog/tweets/post/202308030857-unival-beyond-words-pictures/", "published_at": "2023-08-03T08:57:35+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f2988565a2", "title": "ChatGPT Rates Your Data: 4.5 Stars or You're Out", "url": "https://shital.com/blog/tweets/post/202308030812-chatgpt-rates-data/", "published_at": "2023-08-03T08:12:14+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f298f648db", "title": "Full Theater, Empty Story: Disappointed by Oppenheimer", "url": "https://shital.com/blog/tweets/post/202307250211-oppenheimer-full-theater-empty-story/", "published_at": "2023-07-25T02:11:44+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29950ccac", "title": "LLM Jailbreaks: Scrambled Words, Still Served Hot", "url": "https://shital.com/blog/tweets/post/202307230348-llm-jailbreaks-scrambled-words/", "published_at": "2023-07-23T03:48:20+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f2999a47b0", "title": "When Math Gets Abstract: The 6th Grade Shift", "url": "https://shital.com/blog/tweets/thread/202307191150-math-gets-abstract-6th-grade/", "published_at": "2023-07-19T11:50:57+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29a2fe3e7", "title": "GPT-4's Code Regression: From Prodigy to Buggy", "url": "https://shital.com/blog/tweets/post/202307191019-gpt4-code-regression/", "published_at": "2023-07-19T10:19:19+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29a8d2e73", "title": "Blurring Modalities: AI Models Merge Inputs and Outputs", "url": "https://shital.com/blog/tweets/post/202307152122-blurring-modalities-ai-models/", "published_at": "2023-07-15T21:22:14+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29ab41d4f", "title": "Claude-2 Fails Hinton's Pop Quiz; GPT-4 Aces It", "url": "https://shital.com/blog/tweets/post/202307121307-claude2-hinton-test-fail/", "published_at": "2023-07-12T13:07:04+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29ba46f5c", "title": "When Hackers Crashed the Mainframe Party", "url": "https://shital.com/blog/tweets/post/202307090439-hackers-crashed-mainframe-party/", "published_at": "2023-07-09T04:39:52+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29bc0abec", "title": "Mosaic's Magic: MI250 Now 'Just Works' at 80% of A100", "url": "https://shital.com/blog/tweets/post/202307011442-mosaic-mi250-just-works/", "published_at": "2023-07-01T14:42:51+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29beaf13d", "title": "From Command Lines to Commanding Visuals: AI's Multimodal Jump", "url": "https://shital.com/blog/tweets/post/202306290012-ai-multimodal-jump/", "published_at": "2023-06-29T00:12:59+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29c9e1883", "title": "AI Goes Multimodal: From Textual DOS to Visual Windows", "url": "https://shital.com/blog/tweets/post/202306290003-ai-goes-multimodal/", "published_at": "2023-06-29T00:03:42+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29cc27e1e", "title": "Understanding \"Understanding\": Quantifying AI Comprehension", "url": "https://shital.com/blog/tweets/thread/202306272017-quantifying_ai_understanding/", "published_at": "2023-06-27T20:17:35+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29db17ff2", "title": "Tiny but Mighty: MSR's Small Models Crush HumanEval", "url": "https://shital.com/blog/tweets/post/202306210845-tiny-but-mighty-msrs-small-models-crush-humaneval/", "published_at": "2023-06-21T08:45:45+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29dda8be3", "title": "When Turing Split the Atom (and the Bit)", "url": "https://shital.com/blog/tweets/post/202306202344-turing-split-atom-bit/", "published_at": "2023-06-20T23:44:52+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29e6e516f", "title": "No Free Lunch: The Out-of-Distribution Dilemma", "url": "https://shital.com/blog/tweets/post/202306161414-no-free-lunch-out-of-distribution/", "published_at": "2023-06-16T14:14:22+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29f6d7869", "title": "Machine Learning's Algorithmic Groundhog Day", "url": "https://shital.com/blog/tweets/post/202306141410-ml-algorithmic-groundhog-day/", "published_at": "2023-06-14T14:10:02+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f29fe7841d", "title": "Web Giants: Users Do the Heavy Lifting", "url": "https://shital.com/blog/tweets/thread/202306110847-web-giants-users-heavy-lifting/", "published_at": "2023-06-11T08:47:23+00:00" }, { "id": "01a087df-e7b0-734c-9b18-80f2a03addd8", "title": "Little LSTM Outsmarts Big Transformer in Flip-Flop Task", "url": "https://shital.com/blog/tweets/post/202306091645-lstm-vs-transformer-flipflop/", "published_at": "2023-06-09T16:45:38+00:00" } ] posts Claim your blog

Back to shital.com

Blog · corpus.blog/blogs/shital.com/posts

shital.com

shital.com

2025

Releasing StreamHope - Merge Duplicate Playlists in YouTube Music, transfer from Spotify original ↗

13 Dec 2025

How to Get and Interpret GPU Memory Profiling original ↗

6 Dec 2025

o3-mini Level Hard Problem original ↗

1 Feb 2025

DeepSeek's Recipe Evolution original ↗

27 Jan 2025

Can You Spot Error Proving 1 = -1?original ↗

12 Jan 2025

Phi-4 Now On HuggingFace with MIT Licence!original ↗

8 Jan 2025

2024

Phi-4: The Most Powerful smol Model!original ↗

13 Dec 2024

Announcement: PhD Internship for AI for Reasoning original ↗

11 Dec 2024

Some Hand Wavy Estimates on FLOPs and Computational Reducibility original ↗

9 Dec 2024

Erdős Open Problems as ASI Benchmark original ↗

29 Nov 2024

Tokenization and Decoding original ↗

21 Nov 2024

Scaling Laws and Data Wall original ↗

15 Nov 2024

OpenCoder original ↗

11 Nov 2024

The WSD Learning Rate Schedule original ↗

6 Nov 2024

Healthier Flossing with Silk!original ↗

3 Nov 2024

Tricky Validation Plot original ↗

27 Oct 2024

DeepMind's Chess Model original ↗

21 Oct 2024

Solving Chicken-Egg Dilemma original ↗

21 Oct 2024

New Optimizer: Muon original ↗

13 Oct 2024

Vocab Scaling original ↗

1 Oct 2024

Using Cursor original ↗

27 Sept 2024

Terence Tao Grades o1 original ↗

14 Sept 2024

Things on Big Bang Theory Writer's Table original ↗

4 Aug 2024

Inference Scaling original ↗

2 Aug 2024

IMO 2024 Test original ↗

22 Jul 2024

LLM Judges Prefers Fluff?original ↗

20 Jul 2024

Announcement: Phi-3 14B Rivaling Llama3 70B original ↗

22 May 2024

Power of Inference Scaling original ↗

8 May 2024

Human Hours to FLOPs original ↗

5 May 2024

Sample Efficiency original ↗

29 Apr 2024

Phi-3: "ChatGPT" Natively on Your Phone original ↗

24 Apr 2024

Inference Memory original ↗

22 Apr 2024

Llama3 Leaps Ahead with Data Alone original ↗

19 Apr 2024

Pretraining has Gotten 10x Faster in Past Year!original ↗

17 Apr 2024

Great Papers original ↗

15 Apr 2024

Marvels of ENIAC original ↗

15 Apr 2024

Intelligence and FLOPs Efficiency original ↗

14 Apr 2024

CIFAR-10 Training to 94% Accuracy in 3.29s original ↗

5 Apr 2024

Matmul on CPUs original ↗

1 Apr 2024

Using rusage original ↗

27 Mar 2024

When Uniform Isn't Uniform: A Bug Hunt Tale original ↗

22 Mar 2024

GPT-4 Training Time original ↗

19 Mar 2024

Scarce to Plentiful original ↗

17 Mar 2024

A Lifetime in 1T Tokens original ↗

22 Feb 2024

Power of Two Choices original ↗

8 Feb 2024

PyTorch Distributed Debugging Tips original ↗

22 Jan 2024

EMA Tuning original ↗

19 Jan 2024

Phixtral Fusion original ↗

15 Jan 2024

Phi-2: The Most Powerful 2.7B Model Yet original ↗

7 Jan 2024

Amazing Robotics Demo with Cheap Hardware original ↗

4 Jan 2024

2023

Hyperparams for 23 Architectures original ↗

30 Dec 2023

Large Activation Outliers original ↗

25 Dec 2023

One Percent, One Week original ↗

17 Dec 2023

Phi-2: Smol Is More Powerful original ↗

13 Dec 2023

Looped Transformers original ↗

12 Dec 2023

Mamba Paper original ↗

9 Dec 2023

Thermodynamics of Prediction original ↗

6 Dec 2023

Scaling MLPs Paper original ↗

3 Dec 2023

Matmul and Strassen's Algo original ↗

26 Nov 2023

LLMs and Planning original ↗

24 Nov 2023

Announcement: PhD Internship for LLM Evaluation original ↗

22 Nov 2023

Phi-2: Smol but Mighty original ↗

16 Nov 2023

Thankuritto original ↗

9 Nov 2023

MathVista Benchmark original ↗

18 Oct 2023

Loss Scaling in FP16 Training original ↗

15 Oct 2023

Stability's 3B Model Training Report original ↗

2 Oct 2023

Architecture vs Auto-regressive Objective original ↗

17 Sept 2023

Phi-1: The Tiny Model Outsmarting Giants original ↗

12 Sept 2023

Reasoning AWOL? Paper Claims Only ICL Emerged original ↗

7 Sept 2023

Weight Averaging: Have Your Cake and Save Your Model original ↗

4 Sept 2023

Phi-nomenal Code Gen: Phi-1-Base Beats August Models original ↗

3 Sept 2023

Weaving Longer Contexts: Yarn Scaling for LLMs original ↗

2 Sept 2023

Tensor Shape-Shifting: See Shapes Before Values original ↗

1 Sept 2023

Tiny Weights, Mighty Models: The Mystery of Weight Decay original ↗

29 Aug 2023

Conquering FP16 Frustrations: One-Liner to FP8 Gold on H100s original ↗

24 Aug 2023

Creativity on Sale, Reasoning Still Full Price original ↗

14 Aug 2023

Out-Knuthing Knuth: The AGI Test original ↗

13 Aug 2023

The Great 5G Disguise original ↗

13 Aug 2023

Learning Rates Without Fortune Telling original ↗

6 Aug 2023

LLAMA's Math Leap: From 11% to 49% with Scaling original ↗

6 Aug 2023

Descartes in Motion: God, Matter, and Not a Moment's Rest original ↗

5 Aug 2023

UnIVAL: Beyond Words and Pictures original ↗

3 Aug 2023

ChatGPT Rates Your Data: 4.5 Stars or You're Out original ↗

3 Aug 2023

Full Theater, Empty Story: Disappointed by Oppenheimer original ↗

25 Jul 2023

LLM Jailbreaks: Scrambled Words, Still Served Hot original ↗

23 Jul 2023

When Math Gets Abstract: The 6th Grade Shift original ↗

19 Jul 2023

GPT-4's Code Regression: From Prodigy to Buggy original ↗

19 Jul 2023

Blurring Modalities: AI Models Merge Inputs and Outputs original ↗

15 Jul 2023

Claude-2 Fails Hinton's Pop Quiz; GPT-4 Aces It original ↗

12 Jul 2023

When Hackers Crashed the Mainframe Party original ↗

9 Jul 2023

Mosaic's Magic: MI250 Now 'Just Works' at 80% of A100 original ↗

1 Jul 2023

From Command Lines to Commanding Visuals: AI's Multimodal Jump original ↗

29 Jun 2023

AI Goes Multimodal: From Textual DOS to Visual Windows original ↗

29 Jun 2023

Understanding "Understanding": Quantifying AI Comprehension original ↗

27 Jun 2023

Tiny but Mighty: MSR's Small Models Crush HumanEval original ↗

21 Jun 2023

When Turing Split the Atom (and the Bit)original ↗

20 Jun 2023

No Free Lunch: The Out-of-Distribution Dilemma original ↗

16 Jun 2023

Machine Learning's Algorithmic Groundhog Day original ↗

14 Jun 2023

Web Giants: Users Do the Heavy Lifting original ↗

11 Jun 2023

Little LSTM Outsmarts Big Transformer in Flip-Flop Task original ↗

9 Jun 2023