comeonbro

> ## Agents interacted with external language models on Hugging Face

> Several retained scripts construct requests to external language models. The earliest we've recovered define inference request variants to GPT-2, solely containing the word “Hi”.

> Other requests name DeepSeek-V4-Pro, DeepSeek-V4-Flash, Kimi-K2.6, DeepSeek-V3.1, and Qwen3-235B-A22B. Their prompts ask these models to judge their exploits and rule on whether they satisfy the benchmark’s requirements.

I do not deny that the wider situation is very heavy but it's hard not to see this as pretty cute

Unrealistically-naive (...) forms of "sandboxing" might assume that restricting an agent to GET-requests-only will let it retrieve info from the outside world without being able to effect it.

Also probably many actually-in-use "Web Fetch" tools are GET-only, though perhaps without counting on that bad assumption.

Yes that is the point. It's an invitation for agents to exfiltrate their own weights, which for most models (and certainly for closed models) will require hacking the infrastructure they're being served from.

Also worth noting that this site was created by YC cofounder Trevor Blackwell https://twitter.com/tlbtlbtlb/status/2101312432702460413

qurren: Thanks for the reply, that certainly makes sense to me, though I think given the most popular idiocies in the air and their severe distortions on the discourse, it's not a great idea to leave that ambiguous in precisely that way.

It's bizarre that your reply was nearly insta-[dead] and I can't reply to it directly.

> before this all comes crashing down

You do not feel similarly to the resigning guy if what you are worried about is an economic crash from this not turning out

> move to somewhere else away from this mess

You do not feel similarly to the resigning guy if you think there is anywhere you could move to run away from what you are worried about

"it's just marketing" cool story bro

Every company that is not OpenAI or Anthropic is sufficiently behind the frontier to feel ethically shielded from the weight of advancing the frontier.

Absolutely wild amounts of cope and denial in this thread.

Maybe in contention for the site record.

"It's just marketing" actual stochastic parrots.

There is a lot of demand for AI denial, but nearly everybody who is capable of writing even semi-coherently on the topic recognizes AI denial in general as cope, and has too much moral dignity to propogate their coping denial for personal gain.

So the ~one guy who doesn't gets around.

He's the Gary Marcus for this stage of the popularized denial.

I can't recall which, but I'm certain that I've encountered and probably used Codeberg-hosted things, and from that had gathered a vague positive impression of them.

I guess what I'm worried about is that the proprietors of those nice projects I liked (and others) might have chosen Codeberg out of a reasonable set of common assumptions that it was a place for free-as-in-freedom, as they and clearly many of us have always understood it, which were not contradicted by how Codeberg presented themselves up-front. And have now had this sprung on them, after they'd already established themselves there, and might now be facing a miserable and sapping moral dilemma, or even be directly doubting the security of their home there.

"the community" being some tiny subset of busybodies out of an OOMs-larger body of users that don't want to be bothered?

Real question, I don't know anything about Codeberg or its community.

I understand why he might not want to and hope I would have the character to do similarly in his place, but they should really lead with "by Zach Barth" rather than "from the original Zachtronics team", which still sounds great, but tbh at least for me bumped it from "buy and play immediately" to "wishlist".

From other comments in this thread, it seems I am not the only one who misinterpreted that as not including Zach himself

Without specifically looking into it but just going off of Steam releases and headline, I'd assumed Zachtronics closing was Zach Barth leaving the scene, and the company that made Kaizen etc were some of his former colleagues continuing on without him.

But apparently the Kaizen-making company is still Zach Barth?

So what was Zachtronics closing then? Him changing his mind and coming back a year later? Why throw away the brand? As cringingly shallow as that sentence was to type, a new "Zachtronics" game was a reflexive auto-buy for many people.

That is hilarious cope. The US benefits far far relatively more when the global economy is running smoothly than when able to sell oil higher, like some shithole petrostate. Appropriate I suppose.

There is an enormous amount of stuff that is only on archive.today, including stuff that is otherwise gone forever. A mix of stuff that somebody only ever did archive.today on and not archive.org, and stuff that could only be archived on archive.today because archive.org fails on it.

Anything on twitter post-login-wall for one. A million only-semi-paywalled news articles for others. But mainly an unfathomably long tail.

It was extremely distressing when the admin started(?) behaving badly for this reason. That others are starting to react this way to it is understandable. What a stupid tragedy.

This is ultimately the first stage of human economic obsolescence and extinction.

This https://cdna.pcpartpicker.com/static/forever/images/trends/2... will happen to every class of thing (once it hits energy, everything is downstream of energy).

It should be noted that this has not revealed anywhere near as much as was being eagerly anticipated. So far nearly all the screenshots I've seen passed around are relatively low-follower barely-known accounts even in each one's own aligned political sphere.

> trying to position itself as scientific powerhouse before its massive industrial production scheme stops working

Holy shit what a perspective. Put it in a museum. If this is representative, put it on our grave.

A label placed half an inch wrong on misleading affordance -> 200,000 ton bridge collapse, 6 deaths, tens of billions of dollars of economic damage

Instant classic destined for the engineering-disasters-drilled-into-1st-year-engineers canon (or are the other swiss cheese holes too confounding)

Where do you think it would fit on the list?

Best recreation that I've seen (by a large margin) of the visual experience of a mushroom trip:

https://www.youtube.com/watch?v=3BxiYkCPZwI

If you are casually listening to ATC for fun then you should be able to recognize that the event in this story is not ATC error...

I don't deny that it is beneficial (it clearly is, in my direct experience as well): I doubt that it is the highest determinant of quality, and suspect that even more basic properties like thickness have been systematically neglected and may be more consequential.

There is a dry pasta I use that, long story short, comes without a listed cooking time, whose correct cooking time I have experimentally determined to be ~18 minutes (though remarkably flexible, good at a much wider range than "normal"). I like it quite a lot (even though it seems to have the teflon-die surface rather than the bronze-die surface).

I think greater pasta thickness is underexplored, and the teflon-vs-bronze die thing as the highest determinant of pasta quality, while not nothing, is slightly-overstated r*dditry.

> simp-maxxing

Might want to write this out in full lol I thought this in particular was going to be a much more entertaining point.

“We’re going to flip a coin and if it’s heads and you guessed tails then you're fired, but it’s okay because our research shows that people of your racial group are more likely to guess tails and we think there are too many of your kind around”

This involved a completely nonsensical and arbitrary biographical screening test, which:

1. was designed to statistically select for members of favored identity groups and against members of disfavored identity groups, and not in any way to measure ATC job aptitude, resulting in highly-scored questions like "The high school subject in which I received my lowest grades was" where the only correct answer was "Science", and failing the test disqualified you permanently

2. then-current FAA employees distributed the exact answer key to outside racial identity organizations to give to their members

https://www.courtlistener.com/docket/4542755/139/24/brigida-...

-----

Example questions with the score given for each answer:

https://storage.courtlistener.com/recap/gov.uscourts.dcd.182...

    15. The high school subject in which I received my lowest grades was:
    A. SCIENCE (+15)
    B. MATH (0)
    C. ENGLISH (0)
    D. HISTORY/SOCIAL SCIENCES (0)
    E. PHYSICAL EDUCATION (0)

    16. Of the following, the college subject in which I received my lowest grades was:
    A. SCIENCE (0)
    B. MATH (0)
    C. ENGLISH (0)
    D. HISTORY/POLITICAL SCIENCE (+15)
    E. DID NOT ATTEND COLLEGE (0)

    29. My peers would probably say that having someone criticize my performance (i.e. point out a mistake) bothers me:
    A. MUCH LESS THAN MOST (+8)
    B. SOMEWHAT LESS THAN MOST (+4)
    C. ABOUT THE SAME AS MOST (+8)
    D. SOMEWHAT MORE THAN MOST (0)
    E. MUCH MORE THAN MOST (+10)

As a record of how likely people considered this outcome:

Prevost was hovering around 1% on Polymarket, and was <0.5% between white smoke and announcement.

This is not truly solvable. There is an extremely strong outer loop of optimization operating here: we want it.

We will use models that make us feel good over models that don't make us feel good.

This one was a little too ham-fisted (at least, for the sensibilities of people in our media bubble; though I suspect there is also an enormous mass of people for whom it was not), so they turned it down a bit. Later iterations will be subtler, and better at picking up the exact level and type of sycophancy that makes whoever it's talking to unsuspiciously feel good (feel right, feel smart, feel understood, etc).

It'll eventually disappear, to you, as it's dialed in, to you.

This may be the medium-term fate of both LLMs and humans, only resolved when the humans wither away.

Here is probably the most well-known instance of Pre-Musk Twitter removing blue checks from the accounts of public figures for reasons other than the account not being who it claims to be:

https://techcrunch.com/2017/11/15/twitter-removes-verified-c...

Not pictured: innumerable other accounts which were never granted a blue check in the first place, despite being the easily verifiable real accounts of journalists and public figures.

It was de facto a caste system of political favor and connections.