Atlas will evolve to collect data for training. There's a bunch of context and content bots can't process or access, but a browser not only gives the mothership a closer look at all the walled-garden services and virals a user consumes but also a residential IP address.
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.
You can use the built in task agent. When you have a plan and ready for Claude to implement, just say something along the line of “begin implementation, split each step into their own subagent, run them sequentially”
subagents are where Claude code shines and codex still lags behind. Claude code can do some things in parallel within a single session with subagents and codex cannot.
Yeah, in parallel. They don't call it yolo mode for nothing! I have Claude configured to commit units of work to git, and after reviewing the commits by hand, they're cleanly separated be file. The todo's don't conflict in the first place though; eg changes to the admin api code won't conflict with changes to submission frontend code so that's the limited human mechanism I'm using for that.
I'll admit it's a bit insane to have it make changes in the same directory simultaneously. I'm sure I could ask it to use git worktrees and have it use separate directories, but I haven't (needed to) try that (yet), so I won't comment on how well it would actually do with that.
Fundamentally, there is no difference. Blocking syscalls in a Docker container is nothing new and one of the ways to achieve "sandboxing" and can already be done right now.
The only thing that caught people's attention was that it was applied to "AI Agents".
Thank you, I don't think I'm doing anything particularly special with my process. I like blacks, and keep photos slightly underexposed, something that most phone cameras will avoid like the plague, brightening everything up. Cropping and composition is also more important than it seems IMO.
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.
Okay. Let's assume your top priority is to test for genetic health issues, I will also assume that you either plan on reducing the risk of future complications or you are factoring this information into account when deciding to get children (otherwise, why would you want this information?).
In the first case (reducing risk factors), chances that you will at least check positive for one thing are almost 1 (certain) and you don't need a test to know that. In almost all cases, limiting the risk will involve a) eating well b) stopping alcohol / smoking c) regular physical activity d) taking some Aspirin for the remainder of your life. You don't need the test results to start doing this. Either you sincerely want to live longer and shouldn't wait for genetic screening to take your matters into your own hands, or you think that test results will suddenly turn you into a monk. News flash: won't happen.
In the second case (whether or not to procreate): any decision you make on the hypothesis that your children will inherit it and that science will not have solved it in the next 50 years will very likely be a bad reason to not have children (there are many good reasons to not have children, though, but having a rare genetic conditions is not a good reason).
The only people who should not have kids are bad parents and people who don't want kids. Guess what? These two groups that probably have the highest amount of kids on Earth.
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.
Yes, I can agree with you, did not think of that and it seems to be a perfectly valid motive.
Still, I wonder: who would fall in the "highly susceptible" category? Wouldn't these people need to be tested prior to actually knowing they should be tested? (what would push someone to be tested for susceptibility of birthing malformed children before being tested?)
This is perhaps an off topic question, but I mean it seriously. What do you hope to gain from knowing your potential genetic health issues?
If the report showed you had a chance of developing say Parkinsons, how would that change your life now? Could you implement those changes regardless?
I mean, pretty much all genetic markers are just "risk". And lifestyle choices (smoking, drinking, sugar, exercise et al) seem to be well known, and "good for everything". So, knowing that, it seems like there's lots you can do regardless of genetics.
I confess, for me personally, I'm not really interested in my genetic risk factors. Much less my Neanderthal content. Hence the genuine question- what will it tell you thst you care about?
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.
Sure, I get that things like Huntingtons would certainly be worth knowing about. But for genetics like that simply looking at your own previous generations would give you the answer there. (And of course, that assumes you have access to that, which clearly not everyone does.)
Having children is obviously a big responsibility, and every parent wants only the best for them. Doing genetic testing can at least feel like you've done your due diligence.
I would point out though that genetic diseases are a tiny tiny fraction of the things that can go wrong. The overwhelming majority of children turn out fine. But some don't. That's life I'm afraid. Genetic testing isn't a guarantee they'll be OK. I'd go so far as to suggest it's basically meaningless in that context (especially if you have access to parents, grandparents etc.)
I'm not sure which one of us you're referring to, but the primary "cost" as I see it is disclosing your DNA to some 3rd party, who are then free to use that DNA, or sell it, or whatever as they like.
So, in a sense, if you use a DNA service you are essentially putting your unique DNA marker "into the world", and it can be identified to you (especially if any of our relations have also been tested.)
While there may not be any immediate implications, that DNA might be used by law enforcement, insurance companies, employers etc (either now or in the future.) So there are potentially "high costs" - we just don't know what they are yet. But this bell cannot be unrung.
Balancing this possible harm are the possible benefits. Personally, I don't really see any compelling benefits, but my goal with this reply to the original question was to determine if I was missing something.
> 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.
Their findings on KV-cache invalidation are spot on for a single-context approach.
Strata's architecture is philosophically different. Instead of loading a large toolset and masking it, we guide the LLM through a multi-step dialogue. Each step (e.g., choosing an app, then a category) is a separate, very small, and cheap LLM call.
So, we trade one massive prompt for a few tiny ones. This avoids the KV-cache issue because the context for each decision is minimal, and it prevents model confusion because the agent only ever sees the tools relevant to its current step. It's a different path to the same goal: making the agent smarter by not overwhelming it. Thanks for the great link!
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.
How are startups supposed to afford this? How are talented H-1B workers supposed to start companies? And no, the answer is not always an O-1. I know plenty of foreign founders contributing meaningfully to the US economy, now slapped with a 100k fee.
Ever live somewhere that isn't a city, but has access to talent from a local university? No one is sticking around to be hired for $70k a year when they can make $120k a year in a city. Yet, there are plenty of hires due to a local migrant population, which commonly has generational support. This disrupts that. It hurts more than migrants. It hurts communities.
