@preferencemodel
Pinned post
We looked at TerminalBench 4.0 and found a lot of issues. After fixing the issues Sonnet 5.5 scores 89.2%. So, TerminalBench is already mostly saturated.
Along the way we found a lot of crazy stuff, including reward hacks in the published transcripts.View quoted post on X
preferencemodel.com Preference Model Joins Fireworks' Frontier Ecosystem Teams that post-train their own models can now start from rigorous, verified RL environments built to resist reward hacking. The future of AI is specialized intelligence with models tuned to your data, your domain and your customers.
Today, we're proud to launch live from Forge the Fireworks Frontier Ecosystem with 60+ partners across the AI stack, helping you build and ship faster.
Check it out: fireworks.ai/partnersView quoted post on X
Today we’re announcing the Baseten Environment Network, a group of the best RL environment companies that will help our customers build custom environments for post-training.
The Network extends our applied research capacity, helping more customers build environments around their own tasks, data, and feedback.
Partners work directly with customers to create the environments, while we continue to handle model training and production deployment.
Learn more: baseten.co/blog/introduci…View quoted post on X
Wrote on how we're thinking about building RL environments for superintelligenceView quoted post on X
