Same here. 8h hours later I was in.

Curious about people’s experience here. I am working on a small model, verify by jev, and escalate to big model. Some cases, the small model is not a model but some regex.

cheap-confirm-escalate

Using jev as the confirm step.

This probably throws a spanner in the wheels there:

Every piece of reasoning the frontier model does for free has to be rebuilt here as deterministic state.

EDIT: Not to shit on this though. I totally believe that some smart mixture of LLM-reasoning + Jev-style + determinism is going to be pretty amazing.

I had a long talk with chat gpt about this today as well. I think its duable and prolly not too hard either, also you could do lotsa funky stuff with stitched frames of a video in one 4x4 grid for example and send that as one image for analysis. that way temporal understanding can be had for fractions of a second by jev... also because vlm works in pixel space you can get around the whole state machine issue as well, so many possibilities...

I stopped reading at “The honest caveat”…

1/3 of articles warrantying such a comment means it is still important information. If it was 100% of the articles I would agree that it gives no information, but with 1/3 one cannot claim it does not contribute information.

I'd rather read it's slop, it actually adds something to the conversation.

I'd say "imagine if half the posts on HN aren't worth your time" but this is genuinely how it feels nowadays. I'm not sorry

The problem is that not everyone thinks like you. You're dictating what is and isn't a signal to others. OP left a comment to signal to people like me. It saved me a click.

But it takes so little effort to wade through an LLM readme and cut out or edit it down to something palatable.

Just the bare minimum effort to do that would be nice.

For many new coders, LLMs are so good at writing the code, asking it to write the ReadMe sounds like a good idea. Clearly it’s the first impression your project makes so handwriting it is important.

The point is that if they looked at the result they'd question themselves. And the fact that they don't is a violation of the social contract: I can hardly be expected to care about your work, if you don't.

I apply the same rule to the code, unless it is on the job and I have to follow along with the herd, or else....

You’re right that a poor quality readme doesn’t mean a poor-quality product, but it seems more likely than not to me.

Slop is an instant tab close for me. If something’s good, it’ll come around again. I’ll catch it when there's some evidence that it’s worth my time.

If the author is too lazy to even write their own commits, I doubt there would be any effort put into the codebase.

yeah, my point is just that we agents can have their own docs and we can also have a nice curated experience for humans.

Hmm, I wonder if we can automate that

Especially when the article _is plainly about the use of LLMs to do LLM things_.

If this was talking about buying the best apples are the supermarket, sure, complain away about the use of an LLM to talk about apples.

But it's really a specious argument about something that likely only exists because an LLM is good at creating that type of project.

That risks not communicating the reason for the downvotes / flags to the submitter. I think the value of complaining loudly is signalling to those who think LLM write-ups are acceptable (some folks are acting in good faith without understanding why the content is unwanted).

Clearly the people writing these annoying articles don't know.

I wish they'd just use Astra instead. It doesn't write this awful prose.

Everybody knows.

Awesome!

I think these repo owners should be required to film themselves reading out their Claudemade readmes with a straight face.

Every honest caveat. Every seam. Every "Frames lie, so the walk prunes hard".

I prefer when I see other comments to confirm or infirm my opinion.

That's 6 more commits than I'm used to seeing for these sorts of efforts, honestly. (But I will cede that people might be doing all the changes locally and then pushing a single commit, without any real understanding of how version control works.)

I don't quite agree with the Assembly => Compilers => LLMs as some sort of "stages of evolution" argument tbh.

LLMs are an additional way to generate code, but it's not like code generation or "low-code" tools are a new thing that made compilers and programming languages obsolete.

LLMs "just" allow to provide the code generation input directly in human language. But whether providing such a specification in "sloppy" human language instead of precise source code is actually a good thing for long term projects is still arguable (as convenient as it is).

Mathematicians and engineers invented their own precise "DSLs" for a reason, and high-level programming languages are nothing else than precise "specification languages" that don't leave room for the ambiguities of "natural languages".

Releasing the harness as part of LLM-generated projects still makes sense of course.

An llm response here would be more respectful than a downvote without a comment.

And it’s quite disrespectful to dismiss a project that may have taken a lot of time and thought, even with agents, to build, and just dismiss it because the readme was created by an agent.

The title was perplexing enough for me.

So it's kind of like AI hype growing up and rediscovering its roots because the future isn't futuring soon enough.

It's a quick proof of concept of computer use powered by Jev, a new general-purpose classifier model. Here it takes the description of the interface (it can't do images yet) and outputs commands. It's faster and many times cheaper than using frontier generative models like GPT to do the same.

Jev docs gives a good explanation https://docs.typesafe.ai/concepts/system-one

Also see recent discussion https://news.ycombinator.com/item?id=49717558#49718350

It seems like the more honest comparison would be to OCR the screen and send that as input to the LLM?

In movie theaters, out now.

This is only tangentially related but is Jev trained on the same kind of data as the rest of the LLM world? (i.e. unethical)

At a glance it addresses two out of three of my "load bearing points" against AI, which is the cost to run the things, and that they can be used to generate slop.

If it was trained ethically that would "close the gap".

A demo video would be useful, so one can see what it does without having to run oneself.