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.
We are trying to build a free LeetCode for AI researchers. You can learn how to code core ideas from important papers across various areas (e.g., LLMs, JEPA, RL, diffusion), contribute to the problem bank, and importantly, post your thoughts to find like-minded people for collaboration or job opportunities.
Of course, the current version is still far from mature, so anyone who wants to help us with this project is welcome!
one elegant post-training method that caught my attention in 2026 is VPO who in a nutshell post-train the models to output diverse set of good answers.
in this 2h odyssey we cover:
- why diversity is important
- a magic trick
- reward shaping for exploration
- a top tier figure
The curriculum for learning the core RL+inference systems ideas is not long:
- trainer: hf ultra-scale playbook
- inference: Baseten inference engineering
- algorithms/recipes: Lambert’s RLHF book
- kernels: PMPP (foundation chapters suffice)
Blog posts fill the rest (below)
Dating App Profile Review #16
Nerdy/Athletic Indian
This guy put effort into his profile and it shows
But - like most guys - he didn’t put his effort into the things that actually WORK on dating apps
You need sex appeal & contrast
These are the root of all attraction
ANOTHER UNCENSORED LOCAL VIDEO MODEL WITH ZERO REFUSALS
sulphur-2 is based on ltx 2.3 and lets you generate videos locally
→ text to video
→ image to video
→ no refusals
→ fully local generation
full control over what you create
get it here: https://t.co/YMPLgRMkk3
They don't care about your rights, dignity or self respect. They don't love you or want to be your bf or husband. To them, you are an inferior set of holes worth lining up for to use. You have been reduced to a cheap whore...a fuck toy and cumdump for BBC.
really impressive that Prime Intellect's RL/distributed training roles come with a reading list suggestion. small thing, but it says a lot about the culture:
1. Designing Data-Intensive Applications https://t.co/5BuzoG2ZYI
2. Programming Massively Parallel Processors https://t.co/D0VcTuHDtu
3. Machine Learning Engineering Open Book https://t.co/BfSLgEcb30