🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

👋Remember waving your hand, desperate for the teacher's attention?👋 Teacher attention is important for student learning!!! Our new study uses AI to reveal who gets attention and who doesn't across **1M+** tutoring utterances. 🧵👇

“the phd is like a marathon” experiment: run first marathon conclusion: phd is way worse than a marathon… also still haven’t finished phd 😭 jokes aside - thank you Aishwarya Mandyam for being a rockstar coach & friend ❤️

love this -- AI that doesn't just preserve human agency, but can also better enable it. it's exciting to think about future human-AI paradigms that empower people seamlessly in the real world

🎙️ Speech recognition is great - if you speak the right language. Our new Stanford NLP Group paper introduces CTC-DRO, a training method that reduces worst-language errors by up to 47.1%. Work w/ Ananjan, Moussa, Dan Jurafsky, Tatsunori Hashimoto and Karen Livescu. Here’s how it works! 🧵

really neat taxonomy on how students are using AI in education!

one of the biggest qualms I have w/ google scholar is that it feels like its search incentivizes cherry-picking; results i get regurgitate my query and don’t rlly get the depth I need but can’t yet express.. chatgpt search/ai tools for research have been fun to play around with

one month left to submit! at Waymo, we have a unique perspective on what the next-gen of ai techniques requires, and built our open dataset around it ...

!!! I'm at #ICLR2025 to present 🧄Aioli🧄 a unified framework for data mixing on Thursday afternoon! 🔗 arxiv.org/abs/2411.05735 Message me to chat about pre/post training data (mixing, curriculum, understanding); test-time compute/verification; or to try new food 🇸🇬

Big congratulations to Rose on receiving her Stanford University PhD today!

Understanding and preventing misalignment generalization Recent work has shown that a language model trained to produce insecure computer code can become broadly “misaligned.” This surprising effect is called “emergent misalignment.” We studied why this happens. Through this

this is such a creative, fun, and informative experiment!!! 👇 please more work like this

Ideas are cheap—it's execution that matters and reveals the value of good ideas. Love that this paper empirically shows this, and SO happy it's finally out! Congrats, CLS!!!

really like these frameworks! echoes design principles that went behind Tutor CoPilot arxiv.org/pdf/2410.03017 - build system around real world problems - deploy to users - post train/improve with evals based on deployment