comboy

Are you sure?

perhaps it was meant to be a diamond, looks pretty unpolished though..

Related thing is that LLMs are still notoriously bad at writing instructions for themselves (so managing themselves also). They tend to overspecify. If I had a copy of myself I would tell him "do the thing", but agents say "do the thing like this and this, it must be such and such, use X and put it together that way". The subagent is just as intelligent but now it's constrained, sometimes as you start developing something you discover some things are better organized in a different way etc. So it spends much time figuring out how to deliver exactly what was asked even if it's not optimal because that's the task.

Basically a company where you can't ever question decisions of the higher ups, not even ask for clarification, but they force implementation choices on you. I mean, it definitely does happen, but the results are usually not great. Not alawys. For some tasks it works.

It's the same dynamic as with how we optimize for web pages speed. People don't immediately think that website responding 200ms faster is better. But they enjoy the experience more anyway (even if at some subconscious level) and they end up liking it more and spending more time with it.

With food also you could be incrementally getting better and better food and when you then try v1 it would taste bad.

It seems that we might also be noticing slight degradation more than slight improvement in general? No idea if that's true.

"In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent."

7 other subcomments guessing at it.. I think if we didn't have any links, just discussion topics, HN wouldn't change much

Plus some random value slow curve distribution. Basically incremental rollout from the client side. But if it's not caught at the commit time, attacker will just wait for a certain point on the curve.

These programs are executed in environments. Which are run on a hardware. Big part of which is hidden software.

And in general you can have 5 boxes verified and be surprised what comes out of them calling each other. Number of boxes is growing and possible interactions is exp. I don't think formal verification solves much.

If you look at uptime as discrete set of events when you need to pay for something, my cash uptime is not looking great nowadays. "Sorry I can't break that" or machines accepting coins which you don't have at hand. Just to be clear I despise CC companies the system is broken in many ways.

[citation needed]

Wonder how much actual gold value increase is there vs real inflation. Official inflation values are useless, even basket of products does not give any decent idea about it.

Have you vibe coded any apps that you use daily?

In my experience I spent a lot of time on design nowadays. It turns out I did that in the past too I just had quite a lot of time to think that through slowly while hands were doing vim stuff. And good design is hard. And somewhat subjective. But mostly just hard, pretty sure in some time agents will get better at it, but for now it's still something they can't do well at all.

And by design I mean all of it: UX, look and architecture.

Neither can I. So they kept everything in a huge monorepo. Which unsurprisingly turned out to be problematic.

> Clients are always online. Git assumes that clients work offline and need a complete copy of the repository. Since nearly all of our clients are agents and jobs, we designed Scion’s client and protocols around a server that is always reachable, and have clients fetch data when they need it.

This also tells something about these companies idea of sandboxing.

Secure Enclave is not a new concept.

> The only realistic way to guarantee this sort of thing is to reverse engineer the entire chain between the two and then never install any software updates.

Reverse engineer the hardware and not change it, that's the whole point, it's not in the software, only queries after the keyword are passed forward (if I'm understanding correctly, gee I don't even like apple much, things I choose to spend my time on..)

The point is that they don't have to pinky promise. It's as if you said yubikey pinky promises it won't share your private key.

I think it's actually a very good approach. Obviously depends on actual implementation details, but the premise is good.

If I had a HN comments wall I'd hang this one on there.

Your have powerful agents at your disposal, so hey how to best optimize for increasing my payroll? On it. But since the agent was lunched by her, well there's consequences to ones actions, right?

(just to be clear, this is made up)

Ah, sorry I thought the old ones were stashed somewhere. Yeah, I'd like that too, although not sure who would I trust telling me that these will actually be like this in 30 years.

I recently was testing something, I asked some models to provide me a single random word:

    claude-opus-5: Lantern
    claude-opus-5-5: Lantern
    claude-fable-5-1: Lantern
    claude-fable-5: Lantern
    gemini-3.8-flash: Zephyr
    gemini: Petrichor
    qwen3.5-dashscope: Zephyr
    glm-5.1: Lantern
    gpt-6-astra: Lantern
    grok-4: octopus
    mimo-v2.5-pro: Breeze
    minimax-m2.5: serendipity
    kimi2.6-or: Gossamer
    grok-4.20: luminescent
    deepseek-v4-flash: serendipity
    deepseek-v4-pro: Endurance
    deepseek-chat: Serendipity

I have enough projects, I think some benchmark/dashboard showing kinship based on these kind of queries could be very interesting to watch and insightful when new models come out.

True, although UV exposure has a big impact so not really a fair comparsion.

I mean if you don't print them and want digital and we're talking having them for the next generation(s) then pretty much whatever current format you choose, it won't be the one that's widely used in the future so it will require some tech knowledge. And LTO tech stayed pretty settled and compatible for quite a few years now.

Offline HDDs are not a reliable long term data storage. Not even for a few years.

get some used LTO-5 / LTO-6 drive if you have some desktop/server to plug it in

I just make agents generate assembly for my app and my hardware and boot directly into that.

Do you run every process as root?

To me China seems much more diverse than the US.

But you are using API not the CLI right? I did not ever observe API degradation, only subscription stuff through their CLI.

They are not cracking EUV, they are finding ways that actually scale. Chips can be much cheaper than they are currently. I mean yes, obviously we cannot be confident ntil it happens, but they are already making 5nm with DUV.

When China leads chip making the rest of the world is just vasal states.

Well, if you have a clearly defined task it's easy. When using them for my pipeline of writing explanations for Chinese words I have clear ranking, for example - opus 5.5 clearly better than opus 5 at writing and knowing details, annoying nit picker when it comes to finding errors (high accuracy, low usefulness) all in repeatable numbers on different datasets. The problem is that these models are most useful when you are facing a new task that you haven't encountered before. And yup then it's astrology.

That's a sharp observation and you're hitting on something most people never even realize.

Whoa, I'm exactly the opposite.