lovasoa

If I understand well, they mean that the thing they proved in lean is not Navier Stokes. And they don't make any statement about whether the natural language proof is correct or not.

We could force Google to operate its app review service independently. Users could use it and pay for it, or alternatives. Currently Google forces everyone to use their own mediocre service and pay for it without knowing exactly where and how much you pay.

steam engines solved a problem people had

In 429 CE, Devadatta was born on the Deccan Plateau, near present-day Hyderabad. His hamlet was home to about 100 people. Among six children, he was the second. Devadatta died of neonatal tetanus or sepsis within days of birth.

Hardcore

I don't understand why changing effort levels busts the cache. Couldn't effort levels be a decoding-only thing where they just change the probability of the <end of thought> token? Are they literally adding a hidden system prompt that says "effort level: $level" ?

the writing has been on the wall for fifteen years, and yet the number of professional drivers has only increased since then...

Of course we'll rent them! I don't even want to keep a robot in my flat all the time.

But that means speed does matter after all. The renting company will need to clean as many flats as possible with a single robot in a single day.

I was skeptical after last week's announcements, and I have to say I'm impressed now. Over the last two days, I reproduced the entire chain of components involved, and the exploits at play, and even though none of the exploits are crazy smart, the long series of pivots demonstrates a level of agency I didn't think today's LLMs had.

https://github.com/lovasoa/hf-ctf

The timeline is indeed a little bit fuzzy, I haven't found a precise chronology, neither from HuggingFace nor from OpenAI. HF says the hack happened "over a weekend", so probably July 11th-12th. Do you think it took OpenAI a week to realize what happened ?

Maybe when HF published their blog post on the 15th, OpenAI already knew something had happened, had started to investigate, and already reported the issue to JFrog ? But looking at your other comment in the thread, I agree that CVE-2026-65925 and CVE-2026-66014 are better candidates.

Taking a step back, so many basic vulnerabilities in a security-oriented product just makes the headline "agent autonomously escaped containment" sound a little less spectacular.

There is one thing that is still unclear to me after reading a lot about the hack:

* were the ExploitGym solutions actually available somewhere inside huggingface's private datasets ?

* was the model really trying to extract the solutions ? or had some sub-agent drifted enough from the original context that it was not even trying to solve the initial challenge ? that would look much worse for OpenAI, PR-wise.

What they conveniently omit in the blog post is what the vulnerability was: it seems like they renewed JWTs without checking the signature at all ! You could write arbitrary info in an old token, and get it signed without any verification.

https://www.youtube.com/watch?v=q2KCrmQz9WE

The thing I would have liked to know is why they don't use an existing fast SQL parser. Was being slightly incompatible with all existing SQL dialects a product requirement?

3) The engineers working on the chip used ChatGPT from time to time.

I tried a few pptx from consulting firms available online. The rendering does not seem to actually be pixel-perfect, but all were quite readable with a good layout, which is already an impressive feat.

The details are interesting: https://en.wikipedia.org/wiki/Eskom#Corruption

How much faster/slower are you with that process compared to writing code yourself?

I use the latest codex with gpt5.4 and Claude opus every day. they hallucinate every day. If you think they don't, you are probably being gaslighted by the models.

I was interested in the human results, so I had an llm build a visualization for them: https://codepen.io/lovasoaaa/pen/QwKWGBd

You can see that 17% of answers come from India alone and that software developers got below average results, for instance.

The author of this project is also the author of redis. He knows what he is doing.

Running inference for a model, even when you have all the weights, is not trivial.

I use linux at home (with a HiDPI screen) and MacOS for work. The screen works well with both computers. I mostly just use a text editor, a browser, and a terminal though.

Linux has bugs, bug MacOS does too. I feel like for a dev like me, the linux setup is more comfortable.

How do you call the opposite of green washing? When you want to show that you are burning as much energy on training models as the others.

Because we can

I use a fork of sqlx in SQLPage [1]. I think my main complaint about it is runtime errors (or worse, values decoded as garbage) when decoding SQL values to the wrong rust type.

* [1] https://sql-page.com/

A consortium of investigative journalists investigated and attributed the attacks the Ukraine's secret service: https://en.wikipedia.org/wiki/Nord_Stream_pipelines_sabotage

No. The LLM's answer is correct.

No it is not: https://en.wikipedia.org/wiki/FIFO_%28computing_and_electron...

In my opinion, the clean way to implement this is with methods instead of attributes for name, desired_speed, etc...