Today, we got Reinforcement Fine-Tuning (RLFT)! (OpenAI releases day 2) Unlike Supervised Fine-Tuning (SFT) (already available on OpenAI for models like 4o), RLFT trains models with a flexible, non-fixed objective.
RESEARCH
-
Research Visit at CMU: Multilingual Multimodal AI Focus
By
–
Some life updates: starting a research visit at @LTIatCMU
, working with truly amazing @gneubig and the team on range of topics in multimodal AI, most notably multilingual multimodal LLMs and multimodal evaluation. Excited for the journey and the opportunity to work along -
Recipient of Prestigious VinFuture Grand Prize Award
By
–
Honored to receive the VinFuture Grand Prize.
-
Chip Advancements Shaping AI’s Next Competitive Wave
By
–
What should you expect from the next wave of competition? @andrewdfeldman will join a panel of industry leaders to discuss how the latest chip advancements are informing AI's capabilities, moderated by @sharongoldman Register for @FortuneMagazine Brainstorm AI here:
-
Great Interview with Peter Diamandis on Innovation
By
–
I had a recent great interview with @PeterDiamandis . Dont miss it:
-
Speculative Decoding Achieves 2-3x LLM Generation Speed-ups
By
–
Speculative decoding speeds up generation from LLMs significantly by computing several potential tokens in parallel. Learn about this technique and how it has been utilized to achieve 2–3x speed-ups at inference: https://t.co/RLCOrYxZnv pic.twitter.com/9h5TMGtU02
— Google AI (@GoogleAI) 6 décembre 2024Speculative decoding speeds up generation from LLMs significantly by computing several potential tokens in parallel. Learn about this technique and how it has been utilized to achieve 2–3x speed-ups at inference: https://
goo.gle/49npAHF -
Groq partners with Meta to deploy advanced AI model on LPU systems
By
–
If you want to talk to our team about our partnership with @AIatMeta deploying this model on our LPU based systems let us know. Meta's post-training techniques are cutting edge to get this quality jump.
-
70B Model Beats GPT-4o and Mistral Large 405B
By
–
This absolutely NUTS, a 70B beating GPT4o, Mistral Large 123B AND 405B! Truly got ChatGPT at home!
-
Custom AI Models for Domain-Specific Applications Launch Q1
By
–
Build a custom model for your domain, using a small amount of expert data.
— Greg Brockman (@gdb) 6 décembre 2024
In alpha now, for launch in first quarter next year. https://t.co/TPHmh6xJfcBuild a custom model for your domain, using a small amount of expert data. In alpha now, for launch in first quarter next year.
-
New AI Research Paper Navigation Tool Launched with Hugging Face
By
–
Navigating the vast landscape of research papers in the rapidly evolving field of #AI can be daunting. So we partnered with @huggingface
, & built a new tool to make this process significantly more manageable & interactive! Check it out