Andi Peng
@TheAndiPenguin
Researcher @AnthropicAI | PhD @MIT_CSAIL | formerly @MSFTResearch @Yale @WHOSTP | cats are dope.
Tal vez te guste
More to come in the model card, but thrilled to be releasing our safest and most aligned model yet.
proud of you, and excited for what comes next!
This message is bittersweet. When I joined xAI, its impossibly ambitious mission drew me in. I also joined because of trust in Tony, a close mentor and friend. I knew it was where I could do and grow most. In retrospect, this was right: every year at xAI was incomparable to a…
One thing I'm especially excited about with these new models is how far we've driven down reward hacking - ensuring the best coding models in the world continue to execute meaningfully - WITHOUT cheating
We created a number of new evals to assess reward hacking propensity in our models (see details later in 🧵) . On average, across these evals, Claude Opus 4 demonstrates a 67% decrease in reward hacking and Claude Sonnet 4 a 69% decrease compared to Claude Sonnet 3.7.
Proud of what we cooked but even prouder of incremented integer naming 🥹
Introducing the next generation: Claude Opus 4 and Claude Sonnet 4. Claude Opus 4 is our most powerful model yet, and the world’s best coding model. Claude Sonnet 4 is a significant upgrade from its predecessor, delivering superior coding and reasoning.
Come to our workshop!! agentic safety is important!!
📢 Announcing the first @ieee_ras_icra workshop on Safely Leveraging VLMs in Robotics! #ICRA2025 🎯 How can we safely leverage vision-language foundation models to expand robot deployment? 📅 Short papers & failure demos due 04/11/23 🌐 tinyurl.com/safe-vlm 🧵(1/5)
no math, just pika pika
A few researchers at Anthropic have, over the past year, had a part-time obsession with a peculiar problem. Can Claude play Pokémon? A thread:
Introducing Claude 3.7 Sonnet: our most intelligent model to date. It's a hybrid reasoning model, producing near-instant responses or extended, step-by-step thinking. One model, two ways to think. We’re also releasing an agentic coding tool: Claude Code.
Do you work in AI? Do you find things uniquely stressful right now, like never before? Haver you ever suffered from a mental illness? Read my personal experience of those challenges here: docs.google.com/document/d/1aE…
Interested in making computer use agents safer (and more interpretable)? Consider applying to work with me or one of our other amazing mentors!
We’re starting a Fellows program to help engineers and researchers transition into doing frontier AI safety research full-time. Beginning in March 2025, we'll provide funding, compute, and research mentorship to 10–15 Fellows with strong coding and technical backgrounds.
ew
With 330 submissions and 21 acceptances (6.4% acceptance rate), I the NeurIPS high school project track may be the new most selective ML venue!
Awesome work from @esindurmusnlp and team!
New Anthropic research: Evaluating feature steering. In May, we released Golden Gate Claude: an AI fixated on the Golden Gate Bridge due to our use of “feature steering”. We've now done a deeper study on the effects of feature steering. Read the post: anthropic.com/research/evalu…
This morning the White House issued a National Security Memorandum declaring that 'AI is likely to affect almost all domains with national security significance'. Attracting technical talent and building computational power are now official national security priorities.
Announcing Transluce, a nonprofit research lab building open source, scalable technology for understanding AI systems and steering them in the public interest. Read a letter from the co-founders Jacob Steinhardt and Sarah Schwettmann: transluce.org/introducing-tr…
For me, the overarching goal for AGI has always been to create machines that can execute actions in the world to help humans. Beyond proud to contribute to our release of the first computer use agent today!
Introducing an upgraded Claude 3.5 Sonnet, and a new model, Claude 3.5 Haiku. We’re also introducing a new capability in beta: computer use. Developers can now direct Claude to use computers the way people do—by looking at a screen, moving a cursor, clicking, and typing text.
In more physics news today: we present a method to adaptively allocate more compute to "harder" problems, resulting in a reduction of up to 50% in compute at no cost to performance on math and coding tasks!
Inference-time compute can boost LM performance, but it's costly! How can we optimally allocate it across prompts? In our latest work, we introduce a simple method to adaptively allocate more compute to harder problems. 🔥 Paper: arxiv.org/abs/2410.04707 Learn more! 1/N
Lovely waking up and discovering that I did a physics PhD all along
BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”
United States Tendencias
- 1. Auburn 30.2K posts
- 2. Brewers 43.5K posts
- 3. Kirby Smart 3,002 posts
- 4. Michigan 57.4K posts
- 5. Kyle Tucker 2,459 posts
- 6. #ThisIsMyCrew 2,622 posts
- 7. Nuss 5,330 posts
- 8. Hugh Freeze 1,676 posts
- 9. Penn State 27K posts
- 10. Sherrone Moore 1,517 posts
- 11. #UFCRio 54.1K posts
- 12. Billy Napier 2,347 posts
- 13. #EnnisLima 2,517 posts
- 14. Indiana 48.4K posts
- 15. #MagicBrew 8,230 posts
- 16. Andrew Vaughn 1,620 posts
- 17. #FightOn 1,042 posts
- 18. James Franklin 14.3K posts
- 19. Chad Patrick N/A
- 20. Wisconsin 19.6K posts
Tal vez te guste
-
Andreea Bobu
@andreea7b -
Abhishek Gupta
@abhishekunique7 -
Deepak Pathak
@pathak2206 -
Dhruv Shah
@shahdhruv_ -
Pulkit Agrawal
@pulkitology -
Hancheng Cao
@CaoHancheng -
Karl Pertsch
@KarlPertsch -
Sarah Cen
@cen_sarah -
Anca Dragan
@ancadianadragan -
Jonathan Bragg
@turingmusician -
#CVPR2026
@CVPR -
Dylan HadfieldMenell
@dhadfieldmenell -
Brian Christian
@brianchristian -
Mitchell Gordon
@mitchellgordon -
Maithra Raghu
@maithra_raghu
Something went wrong.
Something went wrong.