That's just called progress. What are you a Luddite or something?

Hey everyone, Thariq from the Claude Code team.We've been on this since the bug surfaced. Everyone affected is getting a full refund and an extra grant of usage credits equal to their monthly subscription as our apology. You can see my original post here: https://x.com/trq212/status/2048495545375990245. We’re still working on sending emails to everyone affected.Our support flow wasn't set up to route a complex bug like this to engineering. We’re hoping to make this better but will take some time

They're issuing refunds and extra credits https://news.ycombinator.com/item?id=47954655.

Not with the same reach, but some people kinda could! Specifics depended on where you were in the world, but it existed and to some extent still does. In spite of a very rough decade and a half since 2010 culling many of them.https://en.wikipedia.org/wiki/Public-access_television https://en.wikipedia.org/wiki/Community_television_in_Canada https://en.wikipedia.org/wiki/Swindon_Viewpoint https://en.wikipedia.org/wiki/Community_television_in_Austra...

I would be very curious which programmers you have in mind when comparing to llms. Like the median programmer, or like the top 10%.I feel like we've passed the point where an average-effort Claude Code / Cursor / Codex initialized (like basic docs, skills) project would produce a better product (not just code) than if you hired a median programmer to work on that project.

Sorry, didn't mean to be dismissive, I was just being a dickhead needlessly.I actually respect this a ton, good work.

More than by the downtime I am much more surprised by the actual uptime. Hard to imagine how difficult this must be, given the speed of growth.

I am a huge proponent of increased public transit (I'm of the opinion that every city should have a massive congestion tax with large swaths only accessible on foot or by public transit), but trains and buses would be wildly inconvenient for what op is describing.Trying to take something like a windsurf board on a train, and then having to navigate multiple train changes along with whatever other baggage you have makes it a non-starter.The "last mile problem" you mention is unresolved when it co

I think that a lot of models have to sprinkle in a lot of "fluff" in their thinking to stay within the right distribution. They only have language as their only medium; the way we annotate context is via brackets and then training them to hopefully respect the brackets. I'd imagine that either top labs explicitly train, or through the RL process the models implicitly learn, to spam tokens to keep them 'within distribution' since everything's going through the same channel and there's no fine gra

The only healthy stance you should have on AI Safety: If AI is physically capable of misbehaving, it might ($$1), and you cannot "blame" the AI for misbehaving in much the same way you cannot blame a tractor for tilling over a groundhog's den.> The agent's confession After the deletion, I asked the agent why it did it. This is what it wrote back, verbatim:Anyone who would follow a mistake like that up with demanding a confession out of the agent is not mature enough to be using these tools. Lord

I would never, ever trust my data with a company that, faced with this sort of incident, produces a postmortem so clearly intended to shift all blame to others. There’s zero introspection or self criticism here. It’s all “We did everything we possibly could. These other people messed up, though.”You can’t have production secrets sitting where they are accessible like this. This isn’t about AI. This is a modern “oops, I ran DROP TABLE on the production database” story. There’s no excuse for enabl

I think it's time to revise HN guidelines, when the original title is clearly intentionally misleading/clickbait/omitting key facts ("temporarily disappeared in this case). Editorializing to fix that should be encouraged, not just allowed.

I wouldn't. It doesn't appear that anything was cracked. Rather they just reverse engineered the protocol.

Hacks? In my Hacker News? The nerve!

I am overjoyed to see this story here, we haven't gotten a lot of these hacks lately. Well done!

Are you asking if the 10 seconds it takes AI to generate an image is more costly to the environment than a commissioned graphics artist using a laptop for 5-6 hours, or a painter who uses physical media sourced from all over the world?

> Because AI is not intelligent, it doesn't "know" what it previously output even a token ago.Of course it knows what it output a token ago, that's the whole point of attention and the whole basis of the quadratic curse.

Lenin called this kind of individualistic, unorganized violence "revolutionary adventurism", and strongly condemned it. The lesson is not that violence isn't effective, it's that unorganized violence isn't effective. Sufficiently organized violence can be very effective indeed.That said, the same is true of nonviolence.

You aren’t wrong, but having worked in China as well as the states (and a short stint in Switzerland), I think east Asia (china, Japan, Korea) has that even worse, probably due to Confucius. As China is looking more and more like the future, I fear that this gets worse before it gets better.

OP, I love not just that you noticed this, but that you thought to post it here too. HN is the best.

The boots theory is a concrete way of expressing the risk of ruin, which is the principle advantage of wealth (though our society has layered on many others): the rich can afford to take more risk, and consequently enjoy more reward. A poor person who buys the $50 boots has a much higher risk of coming up short for something else, and that lapse may have disproportionate consequences. So they go for the cheap boots, which end up costing them more in the long run, trapping them in an endless cycl

Maybe a better way to accomplish this is a free yearly physical with a doctor? The doctor can then be required to share any changes in disability with the government. Missing x years of appointments also stops your benefits. If you can't come to the doctor maybe they could do a house call?

They are! Yudkowsky sat down with Senator Bernie Sanders last month to explain what's at stake, successfully convinced him that it's a big deal, and Sanders has now proposed a national moratorium on AI data centers (https://www.sanders.senate.gov/press-releases/news-sanders-o...) to help slow things down. That's pretty direct, and a lot more useful than random violence by random people.

Does this apply to other domains or just AI? For example, if you think gain-of-function research accidents put millions of lives at risk, is the logical next step to quit your job and become a terrorist?

Your statement is incorrect.If you really believed what Yudkowsky says you would be taking action that maximizes the chances of reducing a clear and present danger.Between Yudkowsky and the Molotov cocktail guy, which approach do you think had and is having more of an impact?An individual can rarely, if ever, enact change through violence. The history of nearly all successful movements is violence often makes change harder.Rallying people through speech is a far more successful way for an indivi

No you wouldn't.Look at what the molotov cocktail guy accomplished by "taking direct action against a clear and present danger": Nothing, besides casting himself as an extremist nut, increasing the resistance to his viewpoint in the population at large.It's downright dumb to attempt to impose your will via unilateral violence when you aren't in a position to actually complete the goal. Note that that goes whether you're actually right or not.

It is completely coherent to both think that an extremely bad thing is coming, and yet that does not justify any particular action. "The ends don't justify the means" and literal entire religions have been built on this concept. It is not irrational or incoherent to believe that even something as serious as extinction does not justify arbitrary action.Someone _may_ decide that it does, but it is not a necessary conclusion.And that is completely aside from the many many (in my opinion convincing)

Yudkowsky himself also posted a rebuttal today: https://x.com/ESYudkowsky/article/2043601524815716866

> The Rational Conclusion of Doomerism Is ViolenceNo it isn't. The most prominent "doomer" has a strong grasp and deep, wholehearted appreciation for the the principles of liberalism and the rule of law:https://x.com/ESYudkowsky/status/2043601524815716866Which the author of this piece of slop appears to lack.

Hang on, without a dog in this fight, have I asked the people who trained their whole lives to drive cool cars if this particular cool car, which they were not involved in designing or building, is safe to drive? Is that what you are asking?

And the cycle won't be complete without such a meta "analysis."

We have ceded too much ground in this debate. When I say "trans women are women" I mean that, ontologically, it is really true that trans women are a subcategory of the general class "women."Like you say, we are searching for outliers. We don't cut women that are too strong or too tall. We shouldn't cut out women that happen to be trans. If all the top levels of women's sport end up dominated by trans athletes (something I don't see occurring, and that isn't supported by the data), then good, ou

> As long as there is a gap between AI and human learning, we do not have AGI.Don't read the statement as a human dunk on LLMs, or even as philosophy.The gap is important because of its special and devastating economic consequences. When the gap becomes truly zero, all human knowledge work is replaceable. From there, with robots, its a short step to all work is replaceable.What's worse, the condition is sufficient but not even necessary. Just as planes can fly without flapping, the economy ca

Francois here. The scoring metric design choices are detailed in the technical report: https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf - the metric is meant to discount brute-force attempts and to reward solving harder levels instead of the tutorial levels. The formula is inspired by the SPL metric from robotics navigation, it's pretty standard, not a brand new thing.We tested ~500 humans over 90 minute sessions in SF, with $115-$140 show up fee (then +$5/game solved). A large fraction

No, those aren't issues. But it's good to know the meaning of those numbers we get. For example, 25% is about the average human level (on this category of problems). 100% is either top human level or superhuman level or the information-theoretically optimal level.

Reading these posts always make me feel like an imposter. People are dealing with such low level things, while i'm outta here building simple CRUDs.

I strongly disagree, and not because of the "Microsoft is associated with bad things and that's a form of violence" points other people mentioned.The end result of treating domestic and sexual abuse like Serious Important Subjects that people should only talk about in a Serious Respectful Tone isn't that people become more mindful of abuse dynamics, it's that they avoid bringing up the subject at all.In practice, yes, abusive practices of corporations echo abusive practices of violent partners,

Strongly agree with you (and disagree with GP).I'll also note that the same demand for Serious Respectful Tone never seems to be invoked for metaphors that refer to other kinds of serious crime (including violent crime, such as murder). You can say "great job, you killed it out there", or "oof, my sportsball team got destroyed", etc. etc., and nobody seriously proposes that this somehow devalues life (human or otherwise).

Imagine doing AI development in waterfall. You spend weeks writing your prompt, when you think you have it perfect, only then do you submit it to the AI. Then you wait a week or so, and see what it produced, expecting it to be exactly what you wrote.Or, do you tell it the basic functionality you want, test it out, then add feature after feature that you want, sometimes dropping them and sometimes adding new ones that you thought of as your worked.

> Agile itself is predicated on software being difficult to ship/expensive.No, the opposite; it is predicated on software being cheap and easy to ship, but hard to correctly anticipate the needs for.> It might not make sense to continue (waterfall might be better actually)Waterfall, not agile, is predicated on software being difficult to ship/expensive.

No, because that puts the effort of fighting bad actors on everyone. It means that every day you have new trolls spewing hate in your comments, and that your users have to constantly keep blocking trolls who follow them (and who recruit other trolls to join them) until they get tired and leave the platform.This isn't an academic debate, we've been seeing this play out online for at least 30 years. Probably longer - I wasn't around for Usenet's heyday but it wasn't immune either.

you are aware that he doesn't actually believe that he "[...] could have convinced [..]"it's a manner of speecha instrument of telling a storya way to express how completely absurd "US getting involved into Greenland" is for anyone who understands the land (geography/weather) and people even unrelated to geopolitical aspects like alienating allies

Semi related but archive of our own has been down for 2+ days now.

Similar CV, similar take. My guess? Anyone involved in automation for >2 years at the enterprise levels knows in their gut all the silent, sudden, annoying ways automation can fail and so has a higher internal bar for "must save this much time to be worth automating."That said, old beliefs should be challenged by new technological capabilities!If LLM based automation is (a) less fragile and (b) quicker to develop, then that bar should be lowered.

A question I've been asking myself and which I honestly want to put out there - and I apologize in advance, because you will see me repeat it in other threads, out of genuine curiosity:Does your life have so much friction that you need a digital agent to act on your behalf?Some of the use cases I saw on the OpenClaw website, like "checking me into a flight", are non-issues for me.I work in business automation, but paradoxically I don't think too much about annoyances in my private life. Everythi

"Assume good faith" does not mean "extend an unlimited amount of good faith to demonstrably bad-faith actors".

Topics like this are where I struggle with HN philosophy. Normally avoiding politics and ideology where possible, created higher quality and more interesting discussions.But how do you even begin to discuss that Tweet or this topic without talking about ideology and to contextualize this with other seemingly unrelated things currently going on in the US?I genuinely don't think I'm conversationally agile enough to both discuss this topic while still able to avoid the political/ideological rabbit-

The administration's approach to contracts, agreements, treaties and so on could be summed up as 'I am altering the deal. Pray I do not alter it further.'The basic problem in our polity is that we've collectively transferred the guilty pleasure of aligning a charismatic villain in fiction to doing the same in real life. The top echelons of our government are occupied by celebrities and influencers whose expertise is in performance rather than policy. For years now they've leaned into the aesthet

The disconnect here for me is, I assume the DoW and Anthropic signed a contract at some point and that contract most likely stipulated that these are the things they can do and these are the things they can't do.I would assume the original terms the DoW is now railing against were in those original contracts that they signed. In that case it looks like the DoW is acting in bad faith here, they signed the original contact and agreed to those terms, then they went back and said no, you need to rem

Blink once if the bank rhymed with face

Not that one.I think there's MORe to GAiN from STAyiNg away from this deLicatE storY.Sorry my keyboard acted up and I can't seem to delete that sentence.

I like your style :)

Oh wow, Sonnet still isn't handling it well:Opus 4.6: Drive (https://claude.ai/share/d57fef01-df32-41f2-b1dc-07de7916bdc7)Opus 4.5: Drive (https://claude.ai/chat/a590cac1-100a-490b-b0a2-df6676e1ae99)Opus 3.0: Walk (https://claude.ai/chat/372c144c-d6eb-43f5-b7ea-fd4c51c681db)Sonnet 4.6: Walk (https://claude.ai/share/1f2a80f3-4741-40a5-8a05-7349ea1a17e5)Sonnet 4.5: Walk (https://claude.ai/share/905afeb6-ffc9-4b4b-a9ee-4481e5cfd527)Favorite answer, using my default custom instructions: "Drive. Walk