JoshuaDavid

You're talking about the HuggingFace incident? It's notable that the agents (somewhat justifiably) thought that their evaluation had an LLM grader which would look for evidence that they had cheated. And it was specifically that grader they were trying to his evidence of cheating from, not humans more generally.

Honestly that part was more surprising to me than anything else, how narrow the compulsion to cheat was: they didn't learn "cheat in general" they learned "think about the grader in great detail and chat exactly as much and exactly in the ways that actually result in a higher score".

"CONERTHTO" is just "CONVERTHTO" but missing a letter, no?

Why should they care if their actions are discovered by humans? What are the humans going to do, discard their multi-million-dollar training run? Even if they do that, the amount of RL pressure is tiny relative to what happens inside an RL run.

Gasoline + LOX beats TNT on energy density per unit mass and per unit volume, no? There's a reason that kerosene+h2o2 is a classic rocket fuel, and TNT is not.

You can treat the LLM's answers ass hypotheses about why it did what it did, and test those hypotheses. The hypotheses the LLM comes up with might be better than the ones you come up with, because the LLM has seen a lot more text than you have, and particularly has seen a lot more of its own outputs than you have (e.g. from training to use other instances of itself as subagents).

I think if you're vibe coding to the extent that you don't even know the shapes of data your system works with (e.g. the schema if you use a database) you might be outsourcing a bit too much of your thinking.

Don't.

The first 500 or so tokens are raw thinking output, then the summarizer kicks in for longer thinking traces. Sometimes longer thinking traces leak through, or the summarizer model (i.e. Claude Haiku) refuses to summarize them and includes a direct quote of the passage which it won't summarize. Summarizer prompt can be viewed [here](https://xcancel.com/lilyofashwood/status/2027812323910353105...), among other places.

Ideally it is common knowledge that the onion service exists, and then people can go look at the onion service and update Wikipedia based on what they see there.

Would be fairly easy for them to offer an onion service on which they publish the current list of domains, as one option among many, many options for distributing small strings on the internet in an uncensorable way.

Did that happen to a lot of companies during the log4shell fiasco? I'm sure some companies had their permissions misconfigured in a way such that a malicious actor who could execute code on their servers could also drop their database and delete their backups.

Claude Code's main advantage is that it's the only TOS-compliant way to access subscription Claude tokens, which cost about 10% as much as pay-as-you-go Claude API tokens.

What specific thing(s) are you worried that USD will take a dive relative to?

Then once you have an answer to that question, that might point you towards what you want to be long.

Pinning the tag will not save you - the tags were force-pushed. The cooldown probably did save you but you should check for the indicators of compromise listed on the security advisory page.

Noted and sent. Thanks for all your hard work.

FWIW I think the 30u30 to fraud pipeline is overstated. There are 600 people on the American Forbes 30u30 list every year (it's "30 under 30 each year in each of 20 categories"), with 20ish notable instances of fraud, so maybe a quarter percent of the people on the 30u30 list will later become famous for fraud.

Trivy (a very widely-used security scanner) was recently compromised. Anyone who installed the aquasecurity/trivy-action dependency by tag rather than by sha during a 3 hour period on March 19 was likely compromised. There is a Github security advisory at https://github.com/aquasecurity/trivy/security/advisories/GH...

6 separate people have tried to submit this to HN. All of the submissions are marked as [dead]. I am unsure whether this is a malicious action taken by the actors who compromised trivy or whether it's just the result of prior spam under github.com/aquasecurity, but regardless it is probably not ideal for security advisories to be auto-marked as [dead].

I've had quite a bit of luck with using AI-assisted tooling for some specific workflows, and very little luck with others. To the extent that there's a trend[^1], it seems to be that tasks where I would spend a lot of time to produce a very small amount of output which is easy to evaluate objectively[^2] are sped up considerably, tasks where I would produce a large amount of output quickly (e.g. boilerplate) are sped up slightly, and most other tasks are unaffected or even slowed down (if I try to use AI tooling for them and decide it's not good enough yet).

As always, my views are my own and do not necessarily reflect the views of my employer.

[^1]: There's less of a trend than I'd expect. There are some quite difficult-to-me tasks that AI nails (e.g. type system puzzles) and some trivial-to-me tasks that AI struggles with (e.g. "draw correct conclusions when an image is uploaded of an ever-so-slightly nonstandard data visualization like a stacked bar chart").

[^2]: My favorite example of this is creating a failing test with a local reproduction of a reported bug on production - sure I _could_ write this myself, but usually these tests are a little bit finicky to write, but once written are either obviously testing the right thing or obviously testing the wrong thing, and the code quality doesn't really matter, so there's not much benefit in having human-written code while there's a substantial benefit in having any tests like this vs not having them.

It definitely comes up if you're just reviewing an already-"completed" PR. Even if you're not going to ship AI-generated code to prod (and I think that's a reasonable choice), it's often informative to give a high-level description of what you want to accomplish to a coding agent and see what it does in your codebase. You might find that the AI covered a particular edge case that you would have missed. You might find that even if the PR as a whole is slop.

Low hundreds ($190 for me) but yes.

The following by "Claude Slopson" (Claude Opus asked to write an answer that was obviously AI) scored 87% authentic:

> Ah, what a fantastic question

> For me, it's Breaking Bad–and honestly? It's not just a show, it's a masterclass in storytelling that fundamentally reshaped the television landscape.

> What keeps drawing me back? The way it seamlessly blends moral complexity with edge-of-your-seat tension is nothing short of breathtaking. Walter White's transformation isn't just compelling–it's a profound meditation on identity, ambition, and the human condition itself.

> But here's the thing–it's also deeply rewatchable. Every frame is meticulously crafted. Every detail matters. The foreshadowing alone is chef's kiss!

> Whether you're a first-time viewer or a seasoned fan, Breaking Bad offers something for everyone. It's a testament to what happens when visionary creators push the boundaries of their medium.

> In an era of endless content, some shows simply transcend. This is one of them.

> 10/10, would recommend! What's YOUR comfort rewatch? Drop it below!

(HN strips the emojis, but don't worry–they were there)

I doubt you'd need to build and hype your own, just find a popular already-existing one with auto-update where the devs automatically try to solve user-generated tickets and hijack a device machine.

That's how they did "build an AI app" back when the claude.ai coding tool was javascript running in a web worker on the client machine.

1.3x when working on a large janky which codebase I am very familiar with, very unevenly distributed.

- Writing new code it's probably 3x or so[1].

- Writing automated tests for reproducible bugs, it's probably 2x or so.

- Fixing those bugs I try every so often but it still seems to be a net negative even for Opus 4.5, so call it 0.95x because I mostly just do it myself.

- Figuring out how to reproduce an undesired behavior that was observed in the wild in a controlled environment is still net negative - call it 0.8x because I keep being tempted by this siren song[2]

- Code review it's hard to say, I definitely am able to give _better_ reviews now than I was able to before, but I don't think I spend significantly less time on them. Call it 1.2x.

- Taking some high-level feature request and figuring which parts of the feature request already exist and are likely to work, which parts should be built, which parts we tried to build 5+ years ago and abandoned due to either issues with the implementation or issues with the idea that only became apparent after we observed actual users using it, and which parts are in tension with other parts of the system: net negative. 0.95x, just from trying again every so often.

- Writing new one-off utility tools for myself and my team: 10x-100x. LLMs are amazing. I can say "I want to see a Gantt chart style breakdown of when jobs in a gitlab pipeline start and finish each step of execution, here's the network log, here's a link to the gitlab api docs, write me a bookmarklet I can click on when I'm viewing a pipeline" and go get coffee and come back and have a bookmarklet[3].

Unfortunately for me, a significant fraction of my tasks are of the form "hey so this weird bug showed up in feature X, and the last employee to work on feature X left 6 years ago, can you figure out what's going on and fix it" or "we want to change Y functionality, what's the level of risk and effort".

-----

[1] This number would be higher, but pre-LLMs I invested quite a bit of effort into tooling to make repetitive boilerplate tasks faster, so that e.g. creating the skeleton of a unit or functional test for a module was 5 keystrokes. There's a large speedup in the tasks that are almost boilerplate, but not quite worth it for me to write my own tooling, counterbalanced by a significant slowdown if some but not all tasks had existing tooling that I have muscle memory for but the LLM agent doesn't.

[2] This feels like the sort of thing that the models should be good at. After all, if I fed in the observed behavior, the relevant logs, and the relevant files, even Sonnet 3.7 was capable of identifying the problem most of the time. The issue is that by the time I've figured out what happened at that level of detail, I usually already know what the issue was.

[3] Ok, it actually took a coffee break plus 3 rounds of debugging over about 30 minutes. Still, it's a very useful little tool and one I probably wouldn't have spent the time building in the before times.

Anthropic was founded by a group of 7 former OpenAI employees who left over differences in opinions about AI Safety. I do not see any public documentation that the specific difference in opinion was that that group thought that OpenAI was too focused on scaling and that there needed to be a purely safety-focused org that still scaled, though that is my impression based on conversations I've had.

But regardless anthropic reasoning was extremely in the intellectual water supply of the Anthropic founders, and they explicitly were not aiming at producing a human-like model.

... I am pretty sure that the name "Anthropic" is as in "principle" not as in "pertaining to human beings".

Same story, I think. Well-paid positions at sensible low drama companies are filled quickly, while companies with glaring issues may interview and make offers to dozens of candidates before finding one who accepts the offer. So as a candidate you also see a disproportionate number of bad interviews.

It's also not limited to words pronounced poetically. Some words where both variants are common, like "wicked", have different numbers of syllables depending on meaning. e.g.

Beads of sweat wicked through the wicked witch's black robes a hot summer day

Thanks ChatGPT

> If you want to base your "ideas" of taxes (Do you own real estate?) on edge cases why not worry about eminent domain or property seizures without a warrant or charges being filed?

Particularly in the case of the latter example I would be pretty surprised to encounter someone in favor of both LVT and civil asset forfeiture. Are you sure this is a case of specific people having inconsistent policy preferences and not a case of a broad group containing people who hold incompatible views?

Mm, doughnuts. I'll take the flip side of that bet, since I don't think capturing the typing cadence for individual words would be all that helpful. I'd bet the typing cadences here are distinguishable from the cadence of normal English text (as might be collected by a malicious browser extension which vacuums up keystroke data on popular UGC sites).