Reposted by @ehsnn.bsky.socialAnja Surina @anjasurina.bsky.social · 26/04/2025In conclusion, EvoTune is a step toward self‑improving LLMs for algorithm design. Full paper: arxiv.org/pdf/2504.05108 Website and code: claire-labo.github.io/EvoTune/ 11/12 161
Reposted by @ehsnn.bsky.socialAnja Surina @anjasurina.bsky.social · 26/04/2025Excited to share our latest work on EvoTune, a novel method integrating LLM-guided evolutionary search and reinforcement learning to accelerate the discovery of algorithms! 1/12🧵 12010
Reposted by @ehsnn.bsky.socialAlexander Doria @dorialexander.bsky.social · 20/04/2025Really interesting experiment scaling generalist LLM search agents to 150k websites with synthetic tasks (like "find a font suitable for a children book") and synthetic evaluation of search patterns. Likely how OpenAI DeepResearch was made. openreview.net/attachment?i... 1194
Reposted by @ehsnn.bsky.socialSebastian Raschka (rasbt) @rasbt.bsky.social · 19/04/2025Just shared a new article on "The State of Reinforcement Learning for LLM Reasoning"! If you are new to reinforcement learning, this article has a generous intro section (PPO, GRPO, etc) Also, I cover 15 recent articles focused on RL & Reasoning. 🔗 magazine.sebastianraschka.com/p/the-state-... 16110
Reposted by @ehsnn.bsky.socialEugene Yan @eugeneyan.com · 18/04/2025Interesting paper from Google that challenges a core assumption in translation evaluation—a single metric can measure both accuracy & naturalness. They found that the best systems had neural metrics that did not correlate with human preferences. arxiv.org/abs/2503.24013arxiv.orgYou Cannot Feed Two Birds with One Score: the Accuracy-Naturalness Tradeoff in TranslationThe goal of translation, be it by human or by machine, is, given some text in a source language, to produce text in a target language that simultaneously 1) preserves the meaning of the source text an... 1132
Reposted by @ehsnn.bsky.socialSung Kim @sungkim.bsky.social · 17/04/2025ByteDance's ReTool, a Tool-augmented Reinforcement learning framework explicitly designed to guide LLMs towards optimal strategies for leveraging external computational tools during reasoning. ReTool-32B achieves 72.5% accuracy in extended settings, surpassing OpenAI's o1-preview by 27.9% 1232
Reposted by @ehsnn.bsky.socialBarack Obama @barackobama.bsky.social · 15/04/2025Harvard has set an example for other higher-ed institutions - rejecting an unlawful and ham-handed attempt to stifle academic freedom, while taking steps to make sure students can benefit from an environment of intellectual inquiry, rigorous debate and mutual respect. Let’s hope others follow suit. 15568914518160
Reposted by @ehsnn.bsky.socialAi2 @ai2.bsky.social · 26/03/2025Meet Ai2 Paper Finder, an LLM-powered literature search system. Searching for relevant work is a multi-step process that requires iteration. Paper Finder mimics this workflow — and helps researchers find more papers than ever 🔍 611723
Reposted by @ehsnn.bsky.socialSebastian Raschka (rasbt) @rasbt.bsky.social · 23/03/2025My next tutorial on pretraining an LLM from scratch is now out. It starts with a step-by-step walkthrough of understanding, calculating, and optimizing the loss. After training, we update the text generation function with temperature scaling and top-k sampling: www.youtube.com/watch?v=Zar2... 06112
Reposted by @ehsnn.bsky.socialBarack Obama @barackobama.bsky.social · 23/03/2025Figured I’d hop on here today for the 15th anniversary of the Affordable Care Act. With everything going on right now, it’s easy to feel like regular folks can’t make a difference – but the ACA is a reminder that change is possible when we fight for progress. 48567271312612
Reposted by @ehsnn.bsky.socialVicki @vickiboykis.com · 17/03/2025If you need a break from SEO, LLMs, and the general emptiness of the modern internet, go here and browse for a bit marginalia-search.com/exploremarginalia-search.comMarginalia Search Engine - Marginalia Search - Explore 712920
Reposted by @ehsnn.bsky.socialCharlie Marsh @crmarsh.com · 16/02/2025uv is one year old today. Hard to believe its only been a year. The growth, adoption, and impact surpassed my wildest expectations. Happy birthday, uv! 🥳 1741136
Reposted by @ehsnn.bsky.socialJeff Dean @jeffdean.bsky.social · 15/02/2025I can't believe they've just cancelled the Epidemic Intelligence Service program at CDC. This program trains the best & brightest epidemiologists, who then go on to have distinguished careers in public health, serving at CDC, in state health departments, overseas, ... 26430107
Reposted by @ehsnn.bsky.socialGus @gusthema.bsky.social · 01/01/2025If you, like new, use Matplotlib to plot some data but don't go too much into making it beautiful, this post will look like magic to you: towardsdatascience.com/from-default... Great post!towardsdatascience.comFrom Default Python Line Chart to Journal-Quality InfographicsTransform boring default Matplotlib line charts into stunning, customized visualizations 3564
Reposted by @ehsnn.bsky.socialVicki @vickiboykis.com · 06/02/2025probably one of the most useful products Google could release at this point is an LLM trained exclusively to help you navigate GCP docs 81186
Reposted by @ehsnn.bsky.socialPaige Bailey (webpaige.dev) @dynamicwebpaige.bsky.social · 02/02/2025🤯🤖 Oh, wow! Someone hooked up the @GoogleAIStudio Gemini APIs to a LEGO Mindstorms robotic head - so Gemini can see, talk, and operate it. The model can even play Connect 4 and chess: 🔗 www.youtube.com/watch?v=TTja... 💻 github.com/CreativeMind...youtube.comCan I beat my AI LEGO Robot at Chess?YouTube video by Creative Mindstorms 0153
Reposted by @ehsnn.bsky.socialantirez @antirez.bsky.social · 28/01/2025Think at this: if it goes in the best way (not obvious, but possible), AI may provide us a jump in medicine so that, many that deeply suffer, will have better lives. It may provide us with tools to help the environment and with assisted choices that will lower the gap between poor and rich people. 2132
Reposted by @ehsnn.bsky.socialJeff Dean @jeffdean.bsky.social · 22/01/2025A great talk on the history and design decisions in Google's TPUs by my longtime colleague Norm Jouppi, winner of the 2024 Seymour Cray Computer Engineering award. Talk: www.youtube.com/watch?v=a-1x... Award announcement: www.computer.org/publications...youtube.comSC24 IEEE-CS Seymour Cray Computer Engineering AwardYouTube video by SC Conference Series 28613
Reposted by @ehsnn.bsky.socialKrassensteins @krassenstein.bsky.social · 21/01/2025In other words donate to Wikipedia. Streisand-effect it! Please share. 24626421095
Reposted by @ehsnn.bsky.socialEugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 21/01/2025Given the r1 furor, folks should really read this paper on policy-gradient based search: arxiv.org/abs/1904.03646arxiv.orgPolicy Gradient Search: Online Planning and Expert Iteration without Search TreesMonte Carlo Tree Search (MCTS) algorithms perform simulation-based search to improve policies online. During search, the simulation policy is adapted to explore the most promising lines of play. MCTS ... 2294
Reposted by @ehsnn.bsky.socialJeff Dean @jeffdean.bsky.social · 17/01/2025Nice-but-depressing multi-time series look at how our climate is being affected over recent decades. 05010
Reposted by @ehsnn.bsky.socialJeff Dean @jeffdean.bsky.social · 07/01/2025Did you know you can use '@' <tab> or '<at>Gemini' to access Gemini in the desktop Chrome address bar? Super convenient! You can also use: <at>Tabs to search your tabs <at> History to search your history <at>Bookmarks to find bookmarks 9to5google.com/2024/05/02/g...9to5google.comChrome's Gemini address bar shortcut rolls outFollowing Tuesday's announcement, the Gemini shortcut in desktop Chrome's address bar has rolled out. In the Omnibox... 27714
Reposted by @ehsnn.bsky.socialSasha Rush @srushnlp.bsky.social · 07/01/202510 short videos about LLM infrastructure to help you appreciate Pages 12-18 of the DeepSeek-v3 paper (arxiv.org/abs/2412.19437) www.youtube.com/watch?v=76gu...youtube.comFlash LLMs: Pipeline ParallelYouTube video by Sasha Rush 🤗 3285
Reposted by @ehsnn.bsky.socialVincent D. Warmerdam @koaning.bsky.social · 23/12/2024Today I learned about 'I dont know' tokens. These are special tokens that you can try and teach an LLM by assigning some probability mass to it during training of wrong predictions. By doing so, maybe it can learn to doubt some more. Wrote a TIL about it here: koaning.io/til/idk 1322
Reposted by @ehsnn.bsky.socialPatrick Morgan @itspatmorgan.bsky.social · 21/12/2024Don't dismiss software engineering just yet. The OpenAI o3 demo packs in A LOT of knowledge that only a person with SWE training would have. Let’s dissect each prompt, shall we? 1114
Reposted by @ehsnn.bsky.socialBeth Sutton-Ramspeck @bsutton-ramspeck.bsky.social · 08/12/2024Imagine there's no countries It isn't hard to do; Nothing to kill or die for And no religion, too. Imagine all the people Livin' life in peace. You may say I'm a dreamer But I'm not the only one. I hope someday you'll join us And the world will be as one. John Lennon, 10/9/40-12/8/80 829299345435
Reposted by @ehsnn.bsky.socialPaige Bailey (webpaige.dev) @dynamicwebpaige.bsky.social · 06/12/2024✍️ A favorite use case for the new Gemini Exp 1206 in aistudio.google.com is to help with citizen science activities on @zooniverse.bsky.social. Gemini's *really, really* solid at handwriting detection - even the most spidery of historical texts! Excited to see what insights this might unlock: 1322
Reposted by @ehsnn.bsky.socialFrançois Fleuret @francois.fleuret.org · 26/11/2024My deep learning course at the University of Geneva is available on-line. 1000+ slides, ~20h of screen-casts. Full of examples in PyTorch. fleuret.org/dlc/ And my "Little Book of Deep Learning" is available as a phone-formatted pdf (nearing 700k downloads!) fleuret.org/lbdl/ 461252247
Reposted by @ehsnn.bsky.socialBrandon Rohrer @brandonrohrer.com · 23/11/2024Adding my love letter to arxiv.org/pdf/2304.01315 Empirical Design in Reinforcement Learning by Andrew Patterson, Samuel Neumann, Martha White, Adam White JMLR 25 (2024) 1-63 #ReinforcementLearning These aren’t the heroes we deserve, but they are the heroes we need.arxiv.org 721749
Reposted by @ehsnn.bsky.socialXiming Lu @gximing.bsky.social · 22/11/2024Are LLMs 🤖 as creative as humans 👩🎓? Not quite! Introducing CREATIVITY INDEX: a metric that quantifies the linguistic creativity of a text by reconstructing it from existing text snippets on the web. Spoiler: professional human writers like Hemingway are still far more creative than LLMs! 😲 3436
Reposted by @ehsnn.bsky.socialKatherine Liu @katherineliu.bsky.social · 19/11/2024A starter pack of starter packs: Robotics and AI go.bsky.app/DfAoaJ1 Computer Vision go.bsky.app/PkAKJu5 Computer Graphics Research go.bsky.app/ckQ1u9 Grumpy Machine Learners go.bsky.app/6ddpivr Reinforcement Learning go.bsky.app/3WPHcHg 79628
ehsnn.bsky.social @ehsnn.bsky.social · 19/11/2024Great read to get good understanding of transformers. 010
ehsnn.bsky.social @ehsnn.bsky.social · 17/11/2024“it’s easier to hold to your principles 100% of the time than it is to hold to them 98% of the time.” hbr.org/2010/07/how-...hbr.orgHow Will You Measure Your Life?Harvard Business School’s Christensen teaches aspiring MBAs how to apply management and innovation theories to build stronger companies. But he also believes that these models can help people lead bet... 010
Reposted by @ehsnn.bsky.socialPaige Bailey (webpaige.dev) @dynamicwebpaige.bsky.social · 14/11/2024🙌 Our latest Gemini model is at the top of the LMSys leaderboard! Try it out today at AI Studio: aistudio.google.com 1424
Reposted by @ehsnn.bsky.socialThe Verge @theverge.com · 13/11/2024Bluesky crosses the 15 million user markbuff.lyBluesky crosses the 15 million user markBluesky’s latest signup surge continues. 933694509
Reposted by @ehsnn.bsky.socialChris @chris.blue · 26/10/2024Bsky (not atproto) has a pretty interesting architecture. The’re using SQLite (and I think Litestream?) for each account’s data. Good write up here: newsletter.pragmaticengineer.com/p/blueskynewsletter.pragmaticengineer.comBuilding Bluesky: a Distributed Social Network (Real-World Engineering Challenges)Bluesky is built by around 10 engineers, and has amassed 5 million users since publicly launching in February this year. A deep dive into novel design decisions, moving off AWS, and more. 47314
