Gaurav Ghosal
@gaurav_ghosal
Ph.D. Student @mldcmu | Former Undergraduate Student @berkeley_eecs and Researcher @berkeley_ai |
1/So much of privacy research is designing post-hoc methods to make models mem. free. It’s time we turn that around with architectural changes. Excited to add Memorization Sinks to the transformer architecture this #ICML2025 to isolate memorization during LLM training🧵
@abitha___ will be presenting our work on training language models to predict further into the future beyond the next token and the benefits this objective brings. x.com/gm8xx8/status/…
Looking beyond the next token TRELAWNEY inserts future tokens <T>...</T> during training to teach models to plan ahead—boosting reasoning, coherence, and control. Highlights: - NO ARCHITECTURE CHANGES. JUST SMARTER DATA. - works with standard decoding - enables controllable…
In "Mind Your Step (by Step): Chain‑of‑Thought can Reduce Performance on Tasks where Thinking Makes Humans Worse", we connect human "overthinking" insights to LLM reasoning, offering a new lens on when thinking‑out‑loud backfires. 📄 Read the full paper: arxiv.org/abs/2410.21333…
One of the better posters I saw today at #icml25 This gets at the root of the problems we were thinking about when we conceived and wrote the CoT paper.
RL with verifiable reward has shown impressive results in improving LLM reasoning, but what can we do when we do not have ground truth answers? Introducing Self-Rewarding Training (SRT): where language models provide their own reward for RL training! 🧵 1/n
Training with more data = better LLMs, right? 🚨 False! Scaling language models by adding more pre-training data can decrease your performance after post-training! Introducing "catastrophic overtraining." 🥁🧵+arXiv 👇 1/9
New work on training reasoning models to be more efficient! Would love to hear your thoughts on this.
Excited to share my first work at CMU, led by my fantastic student Daman @amuseddaman ! 🚀 We post-train reasoning models with reinforcement learning to reduce token usage while largely preserving accuracy.