RT @togethercompute: Most AI benchmarks test the past. But real intelligence is about predicting the future. Introducing FutureBench — a….

Work done during my internship at Together AI is being presented at #icml25. Come and check it out! . We propose a new model alignment pipeline that harness collective intelligence from open-source llms!

Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.

RT @NehzUx: LLMs are getting more powerful, but they still struggle with super long documents. A common trick is "Divide and Conquer" - cho….

Curious why backtracking works so well? Dive into our latest study that unpacks the mechanisms behind its success.

RT @james_y_zou: Our new #icml2025 paper w/@togethercompute shows how to use synthetic data from Mixture-of-Agents to boost LM fine-tuning….

RT @togethercompute: 🚀 Introducing Mixture-of-Agents Alignment (MoAA), a new method to "distill" the collective intelligence of open-source….

Linguistic markers like “hmm” and “wait are a good predictor of response correctness.

Correctness gap widens with increasing problem difficulty.

For reasoning models, shorter responses are more likely to be correct. Non-reasoning models show no trend.

Non-reasoning models augmented by inference-time methods cannot match reasoning models. Find out more in our preprint 👇.📄 Awesome collaboration with @ShangZhu18 @JonSaadFalcon @ben_athi @QingyangWu1 @JueWANG26088228 @Leon75421958 @ce_zhang @bhuwandhingra

Excited to share work from my @togethercompute internship—a deep dive into inference‑time scaling methods 🧠. We rigorously evaluated verifier‑free inference-time scaling methods across both reasoning and non‑reasoning LLMs. Some key findings:. 🔑 Even with huge rollout budgets,

RT @iScienceLuvr: Think Deep, Think Fast: Investigating Efficiency of Verifier-free Inference-time-scaling Methods. "This work conducts a c….

RT @LindaHe49140661: Excited to share our work on scaling LLMs to handle million-token contexts! Training models for ultra-long sequences i….

RT @AnnieFeng6: #ICLR2025 Oral. LLMs often struggle with reliable and consistent decisions under uncertainty 😵‍💫 — largely because they can….

RT @togethercompute: Introducing Open Deep Research!. A fully open-source Deep Research tool that:.• writes comprehensive reports.• does mu….

RT @james_y_zou: Happy to share that Mixture of Agents (MoA) is selected as a #ICLR2025 spotlight!⚡️ @togethercompute. Our updated paper ha….

RT @bhuwandhingra: **New paper from @RoyXie_ **. Do LLMs know when they have read enough to answer a question?. We show how language models….

RT @ClementDelangue: Our science team has started working on fully reproducing and open-sourcing R1 including training data, training scrip….

To be presented @ICLR2025 !!! Super excited for my first iclr conference😆.

f1 is very different from above in that it may scale inference-time compute not through search or self-correct, but through compound AI techniques like Mixture of Agents (. I view methods like this as another pathway of inference-time scaling and can be