5.4 Mini's OSWorld score is a pleasant surprise. When SOTA scores were still ~30-40 models were too slow and inaccurate for realtime computer use agents (rip Operator/Agent). Curious if anyone's been using these in production.
In my experience, especially with Opus 4.6, using subagents greatly mitigates the startup context hit. 4.6 has very obviously been RL'ed on subagent usage and it almost always spins up an Explore agent to get a feel of the codebase and get a token-efficient summary. The 1M context version of 4.6 further alleviates this.
My original question was more along the lines of implementing things like PR review yourself. I was tinkering with an internal service that spins up ephemeral CC instances to analyze PRs, but realized this can easily generalize across arbitrary tasks. Was curious what sort of things folks could use that for.
Curious about this too – does it use the standard context management tools that ship with Claude Code? At 200K context size (or 1M for the beta version), I'm really interested in the techniques used to run it for 30 hours.
I mostly agree with your first point, but is there not a chance for more serious disease to be detected? In which case specialized preventative care is the best path forward? There is definitely something to be said about staying ignorant of these, however.
> The only people who should not have kids are bad parents and people who don't want kids.
I don't think people who are highly susceptible to birthing malformed children should have children either. Genetic testing helps figure out if you might be in that group.
Thanks for spurring the thought. The main reason I'd like to find out is to avoid bringing a child into the world that might suffer from incurable disease.
Largely agree with your point on preventative actions being the same with or without testing, but there is a tail end that might warrant specialized action.
I only care about testing for genetic health issues. I don't want my DNA to be inserted in a database that can be cross-referenced to check for relatives, ancestry, etc.
> A natural reaction is to design a dynamic action space—perhaps loading tools on demand using something RAG-like. We tried that in Manus too. But our experiments suggest a clear rule: unless absolutely necessary, avoid dynamically adding or removing tools mid-iteration. There are two main reasons for this:
> 1. In most LLMs, tool definitions live near the front of the context after serialization, typically before or after the system prompt. So any change will invalidate the KV-cache for all subsequent actions and observations.
> 2. When previous actions and observations still refer to tools that are no longer defined in the current context, the model gets confused. Without constrained decoding, this often leads to schema violations or hallucinated actions.
> To solve this while still improving action selection, Manus uses a context-aware state machine to manage tool availability. Rather than removing tools, it masks the token logits during decoding to prevent (or enforce) the selection of certain actions based on the current context.
A 100k fee is well within the territory of killing job prospects for skilled foreign students graduating from US universities.
What percentage of the AI labs are staffed by either foreign workers or second/third generation immigrants? Look at the composition of high achieving high school students- almost certainly of Asian or Indian descent, certainly many belonging to families of recent immigrants. The pipeline this EO disrupts is immense.