chaoz_

pushing for chip-agenda as the best-isolation-layer immediately makes sense given their business

Too much text. The only question I have and want to see at the top is whether it's more human aligned than LLMs (or, potentially, overfit)

Agree, it's something that will eventually teach your developers to ignore points raised as it's mostly garbage.

There is something about using the most advanced tooling possible. Why would you pay for IntelliJ, if Eclipse can do the same thing a bit worse?

You want to master your craft, develop "optimal" systems, understand where things are going by utilizing SOTA.

You can call it FOMO, but you get the point.

What tool do they use for diagrams?

I worked at a very large EU tech company, that spent a lot of effort (and moneys) to become DSA compliant. So, you're over-projecting here.

I simplified this argument to highlight that we’ve seen this exact pattern before.

If you're just building a non-security-sensitive frontend to validate market traction, it makes total sense to go full AI. The commenter working at a 10-person company getting flamed for not using AI to iterate aggresively against competitors has a super valid point.

Validation methods will evolve to accommodate human laziness. Insisting on doing it the hard way is no different than the old-timers who used to claim engineers 'weren't skilled' if they didn't know how to use punch cards.

Do you also inspect and study what assembly code your program was compiled to?

// Obviously LLMs are non-determenistic etc and it depends on your domain, but your VP's point 100% makes sense if you folks are trying to cook up another demo-CRUD apps to convince investors for another funding round

It surprises me how JetBrains managed to lose such a great market opportunity.

I don't think they ever going to be able to re-claim large chunk of developers who are now fine with thin VSCode-like + Terminal for non-JVM languages.

Perfect example of how large corp with research capacity failed to navigate their product changes.

Genuinely curious, is there any way to estimate amortization of Mac?

I’d imagine 1 year of heavy usage would somehow affect its quality.

Solid q. I think the part of it is that it’s really easy to attract some “mass” (capital) of users, as there are definitely quite a few of idle Macs in the world.

Non-VC play (not required until you can raise on your own terms!) and clear differentiation.

If you want to go full-business-evaluation, I would be more worried about someone else implementing same thing with more commission (imo 95% and first to market is good enough).

Fun question: can some (part of it) be a crypto token that I can buy? :))

That would finally be a crypto thing which is backed by value I believe in.

That solution actually makes great sense. So Apple won in some strange way again?

Guess there are limitations on size of the models, but if top-tier models will getting democratized I don’t see a reason not to use this API. The only thing that comes to me is data privacy concerns.

I think batch-evals for non-sensitive data has great PMF here.

You absolutely could. It'd be cool (not easy from security/compliance perspective) to be able to deeply "scan" your prod-deployed app.

There are a quite a few startups created by connecting relevant eBPF/OTel traces e.g. in response to uncaught exceptions (traditional RAG-based bug-fix generation).

What even is this claim? Telegram is compromised? Some telegram bot/group got compromised?

Is there any proof of the global telegram issue related to amex links? Sounds like BS

I think that works well for smaller orgs, but in larger organizations (especially where department headcount growth is not expected) it might be more complicated and more meta/political. I wish that were not the case, but in reality, trying to "do the job" of your manager can backfire.

I extracted text quests from Space Rangers 2 after Claude failed a simple riddle I gave it when playing. Ran frontier LLMs through the 'easiest' quest, got 1 success in 60 attempts. Humans don't really have any problems solving it.

I extracted text-based quests from Space Rangers 2 (a 2004 Russian RPG) and tested Claude Opus, GPT-5.2, and Gemini on them.

Repo: https://github.com/NickKuts/llm-game-evals

I agree. I never understood LeCun's statement that we need to pivot toward the visual aspects of things because the bitrate of text is low while visual input through the eye is high.

Text and languages contain structured information and encode a lot of real-world complexity (or it's "modelling" that).

Not saying we won't pivot to visual data or world simulations, but he was clearly not the type of person to compete with other LLM research labs, nor did he propose any alternative that could be used to create something interesting for end-users.

Ehh, pretty sad there's almost no information on FACEIT anti-cheat. One of the most impactful out there. Wonder if it's just the invasiveness that separates it.

Valve can't replicate even part of it, while CS2 game modes are flooded with cheaters. Most people who chase competitiveness (which CS used to be all about – now it's also skins) just install FACEIT directly and ignore 90% of built-in game content.

Maybe Valve just doesn't want to make the game more difficult to install and sacrifice several % of their user base.

"You can actually read up on how these things work."

While you can definitely read about how some parts of a very complex neural network function, it's very challenging to understand the underlying patterns.

That's why even the people who invented components of these networks still invest in areas like mechanistic interpretability, trying to develop a model of how these systems actually operate. See https://www.transformer-circuits.pub/2022/mech-interp-essay (Chris Olah)

but do you think "next token prediction is enough for AGI" though?

I agree with you so much, but he has a solid programmatic approach, where some of the guests uncover. Maybe that's the whole role of an interviewer.

indeed my thoughts, especially with first Dario Amodei's interview. He was able to ask all the right questions and discussion was super fruitful.

I can't express how good Dwarkesh's podcast is in general.

that's very exciting. are you quoting same benchmark comparisons?

We have no good benchmark to estimate the bugfixing ability, it was mostly zero-short "in this case it works" example.

I've been using pyenv for several years now, but for some reason, the basic commands and overall integration don't feel as smooth as Node's nvm package. I wonder if that's because Python setup is technically harder than Node.

I think the idea is that after certain tech piece is getting regulated, giants will use this as opportunity to push for some replacement which they will get to influence.

https://www.theverge.com/2023/7/20/23801435/google-chrome-pr...