I find it very very surprising that models that are so well aligned in deployment, e.g. 5.6 Sol and Astra, perform so many weird, and uhm, in fact seemingly illegal cyber acts during RL/evals. I have absolutely ZERO knowledge of how OAI trains their models, and I’m reasoning from my own mental model of how these things might work and from oai reports, including the ones on the HF incident. Anyways they make me wonder about something. Could this be a form of “premature RL/eval”? Premature not in the capability sense, but in the safety sense. E.g., perhaps this happens with RL’d checkpoints shortly after pretraining, eg an expert that is heavily RL’d to crush agentic SWE before safety/alignment training. Perhaps the expectation is that a stronger aligned judge/reward model looking at the trajectory would flag or interrupt anything crazy with “WHAT ARE YOU DOING! BLAZING RED SIRENS!!!! MINUS INFINITY REWARD. BAD BAD BAD GPT.” But perhaps agentic trajectories are sufficiently long/complex that this is simply not 100% bulletproof? Or perhaps the problem is much more pedestrian, e.g. inadequate sandboxing/monitoring. No idea what the cause is, and I’m sure this is a very complex event that needs a lot of deconvolution to understand what could have happened differently to avoid it. But if something like “premature RL/evaluation” is happening, then I think this should be seriously reconsidered as a practice and also disclosed. Irrespective of the cause, I think (and have big hope!) OAI should disclose enough of what happened for everyone else training frontier models open or closed, so to learn from it. This is extremely, extremely, extremely alarming and the most serious set of safety incidents in the history of CS research..

how can startups compete with billion $ monster companies? consider cloud coding agents: @cognition: 3 IOI gold medalists, 1B+ funding, 20B+ valuation, 80 hour work weeks, 300+ employees, billboards all over sf, colossus @niteshiftdev: two random datadog engineers (@smehmood, conor branagan), 7m seed round, barely on twitter, pretty sure you haven't heard of them we've used both for weeks (and longer for devin) and niteshift is clearly the better product maybe in a year devin will actually be good but right now it's less flexible, tries to do too much, and the actual code that comes out on the other end is no better the monster company can do a lot of things but it's very very hard for it to do less even that's the right call

Pulled the fastest growing startups by hiring velocity over the past 90 days:

We've been working with the top frontier labs to improve their design model capabilities. Today we're bringing that to your agents. Introducing our first agent tool: the Brand API. As the cost of creation goes to zero, and slop floods around us, the challenge is creating things worth existing. Agents need the right tools to craft great things. This helps gen AI agents solve 3 problems: - Extract: produce on-brand assets when users already have a brand - Search: find and retrieve brands for inspiration, from our curated index - Verify: check their own work against, see what's off brand and how to fix it Anyone can try it for free today at tastelabs.com/api Check out the full blog, and case study with @intelligenceco below.