I'll be at @COLM_conf next week to share a new decoding method R^3: Retrieve, Realize, Repeat designed to jointly improve both coverage of salient information & faithfulness for long-context summarization.

RT @elmo: Elmo just hopes both teams have fun!

RT @_arohan_: Llama’s 🦙 are going to be really fun this year!

RT @JoshEberley: Imagine being told in 2016 that the Mavericks would acquire Anthony Davis, Kyrie Irving, and Klay Thompson in the near fut…

RT @Han_Fang_: New paper from my team Think Smarter not Harder: Adaptive Reasoning with Inference Aware Optimization

RT @jaseweston: 💭🔎 Introducing EvalPlanner – a method to train a Thinking-LLM-as-a-Judge that learns to generate planning & reasoning CoTs…

@Francis_YAO_ Does it help to warm up the model with SFT data in order to first explicitly learn the proper format for exact match?

RT @sunjiao123sun_: I read the DeepSeek-R1 paper the day it came out, and I don’t think GRPO is the key to its success. Instead, here’s wha…

RT @lateinteraction: Excitement is justified, but the most reinforced takeaways from R1 seem to be that: - If you know how to measure a do…

RT @sriniiyer88: We're hiring PhD interns for Summer 2025 in Seattle to work with us on improving BLT even more! If this is something that…

RT @lateinteraction: A lot of o1/o3 progress is attributed to test-time compute, which makes sense (CoT and maybe discrete search). But a…

RT @badboygargar: mocktails are awesome because they ask the question ‘how much could juice cost?’

@_arohan_ @AIatMeta Welcome aboard !

@puppiesDoglover I totally feel that

🚨🚨I'm excited to share that I'll be joining @AIatMeta next week as a researcher working on Llama agents (planning, reasoning, tool use). It feels like a pretty exciting time to be part of open-source science on AI agents as they begin to see mainstream adoption!

RT @_philschmid: Want to replace BERT in 2025? The time has come. ModernBERT introduces a new, refreshed encoder model family that outperfo…

RT @sanjana_rampi: Do Automatic Factuality Metrics Measure Factuality?🤔 We examined metrics designed to measure consistency between genera…

RT @AIatMeta: New research from Meta FAIR: Large Concept Models (LCM) is a fundamentally different paradigm for language modeling that deco…

RT @raja_biswas: It's been an incredible week! - Proud to be part of the team behind ModernBERT! It's a much needed upgrade on BERT across…

@raja_biswas @kaggle Congrats Raja !!! Wow - so impressive

RT @jxmnop: vLLM is Good Software in an era where Software is no longer valued - easy install - no configuration - no needless warning m…