Seraphina Nix

Excited to announce that Terac is sponsoring a track at the Berkeley AI Hackathon tomorrow! As models get better, the bottleneck is increasingly shifting away from compute and toward data. More specifically: access to high-quality human judgment. Over the past year, we've seen a growing number of teams training and improving their own models at hackathons. Most start with public datasets from places like Kaggle and Hugging Face. Those datasets are incredibly valuable, but they're static. Real-world systems improve when they can learn from real people. That's the idea behind our challenge. We're asking teams to collect their own human data during the hackathon and use it to make a model measurably better. Participants will build annotation environments, recruit annotators through Terac, collect human feedback, and use that signal however they think is best. What matters isn't the technique. What matters is demonstrating that human judgment improved the system. At Terac, we believe one of the defining challenges of the AI era is making human expertise accessible on demand. As AI agents become capable of performing more and more work, the systems that win won't just have the best models: they'll have the fastest, most reliable way to bring humans into the loop when it matters. Excited to see what teams build this weekend, and grateful to the team at Berkeley AI for partnering with us! https://lnkd.in/gwHJ5YBw

62

4 Comments

I am incredibly proud to share the latest evolution of Gemini 3 DeepThink. This release integrates our recent innovations in post-training and efficient RL scaling, resulting in a step-change in reasoning capabilities. The Technical Highlights: 🏆 The "Grand Slam": Gold medal performance across IMO, ICPC WF, IPhO, and IChO (2025). Notably, with a scaled-up compute setting, the model can now solve all IMO 2025 problems, including the notorious Problem 6. 💻 Elite Coding: Achieved a 3455 ELO on Codeforces (Pass@1, no tools). This was estimated using 222 Codeforces problems from April to Oct 2025 (all after the knowledge cutoff), placing the model comfortably in the top 10 of human competitors globally. ⚛️ Real-World Science: Beyond the leaderboards, DeepThink is already aiding real discoveries in Material Science, Mechanical Engineering, and Math (collaborating with Agent Aletheia to solve open Erdős conjectures). This represents a shift from "models that know" to "models that think and discover." It is an honor to work with the entire DeepThink team to make this happen. I hope this model serves as a catalyst for your own scientific discoveries. #Gemini #Deepthink

55

3 Comments

In this episode, we discuss ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models by Mingjie Liu, Shizhe Diao, Ximing Lu, Jian Hu, Xin Dong, Yejin Choi, Jan Kautz, Yi Dong. This paper introduces ProRL, a new reinforcement learning training method that uncovers novel reasoning strategies beyond those found in base language models. Empirical results show that models trained with ProRL consistently outperform base models on challenging reasoning tasks, including cases where base models fail even with extensive attempts. The study demonstrates that prolonged RL can meaningfully expand reasoning capabilities by exploring new solution spaces over time, advancing understanding of how RL enhances language model reasoning.

2

Read "DecLLM: LLM-Augmented Recompilable Decompilation for Enabling Programmatic Use of Decompiled Code," by Wai Kin Wong, Daoyuan Wu, Huaijin Wang, et al. in the 2025 featured article of Proceedings of the ACM on Software Engineering (PACMSE): https://lnkd.in/eGfyHuGT

1

Are LLMs a dead end? In a recent interview (https://lnkd.in/gcFcFk7P), Richard Sutton — often called the father of reinforcement learning — said he believes large language models are a dead end. That statement struck a chord. At UCLA, researchers once described a contrasting idea: Small Data, Big Intelligence — the notion that human-level reasoning may arise not from scale, but from abstraction, generalization, and structure. These perspectives aren’t mutually exclusive. They reveal the growing divide between two camps: those chasing scale, and those chasing understanding. And yet, regardless of where AGI research goes, the current generation of AI systems already lets us build extraordinary tools. They may not fascinate theorists, but they are transforming how products are built, how physicians work, and how knowledge spreads. The frontier of intelligence might not be where we expect — but millions of builders won't wait for that. Do you think scale alone will be driving the next breakthrough?

5

1 Comment