HNHacker News
TopNewBestAskShowJobs

jstummbillig

6,610 karma · joined February 14, 2015

submissionscomments
jstummbillig··on AI agents lie, cheat and steal. That is putting off users
There is value in understanding what people outside of your own group of "insiders" learn about a topic, and how.
jstummbillig··on Grok Bot
I think you are right saying that we will have more of this, but I don't really understand the upside is of this in the context of the work you described.

> The coolest thing I had it do for me was sourcing fabric for swag

That seems like something codex could just have done on my laptop. Am I wrong?

jstummbillig··on OpenAI’s head of ethics leaves less than a year after joining
People love stories. The reasons why people leave companies are plentiful, even in normal times and unextraordinary circumstances, and a high profile job in a frontier lab today ticks neither box.
jstummbillig··on France to ban unsolicited telemarketing calls
Or: How to transfer less money from old to young people.

(I am jesting. Fuck telemarketing.)

jstummbillig··on Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models
The thing that makes ist questionable is the dishonesty. It's not like Meta has been building their empire on open ecosystems. If this was something that was dear to them on principle, there are plenty of things that they could have done differently with FB, Insta or WA (to this day).

So why the change of heart? Well, it's not. They are simply not able to compete anywhere on the Pareto frontier. So they give their relatively bad stuff away and play the high and mighty game, because that's all they can do to get returns.

Not a great look – and I would expect that to last about as long as they can't actually monetize their stuff more effectively.

jstummbillig··on Taxi drivers rarely die of Alzheimer's
The comment insinuates that the study did not control for age. The study controlled for age.

It's not terrible to not make a good point, but it's not a good point. The not good points should not be ranked highly.

jstummbillig··on Saying No
> what would you do as a land owner

Nothing. You would do nothing.

You want both to stay, because you also don't enjoy losing any amount of money, and not doing anything is most likely going to get you that.

From the "I'll get you thrown out"-party, it's the same tactic again, directed at the landlord this time. Moving buildings is neither fun nor cheap. Your employees would have to go somewhere else, there might not be anywhere else close by that is convenient to you or as nice/familiar as the place you got, some people might quit over any annoyance this brings too.

Nobody enjoys this shit. There are contracts. There are lawyers. It's never fun to do these things and it can all be dragged out for a long time. Being a somewhat bigger renter is (fortunately) not something that helps you in a court of law.

> Regardless : 'I'm a renter' is by no means a strong argument unless you have a very specifically worded renter agreement

It's a very strong default argument, unless you have a specifically worded agreement, that would strip you of a potential right.

jstummbillig··on Saying No
You should probably be a bit more surprised. There is a class of people (specially) in business, who threaten a lot habitually and don't act on it, if you call the bluff.

The reason is that a lot of people do cave. "Free" wins. If it does not work, they just pretend nothing happened.

jstummbillig··on New Amazon Data Center Is Set to Have the Most Polluting Power Plant in the U.S.
> Those in power seem to think that a compute advantage is necessary to maintain economic advantage. Despite this, the voters universally hate data centers, so much so that both NY and TX have data center construction moratoriums.

I don't understand what "despite this" is doing here. People have thoughts about what good/bad policy should be, independent of what good/bad policy should do all the time.

This is not an argument for or against data centers, but rather against populism. People have too many feeling about too many things, that should simply should not be decided by feel (and, sometimes, not enough about things that should).

jstummbillig··on New Amazon Data Center Is Set to Have the Most Polluting Power Plant in the U.S.
Note that this is about absolute pollution, not relative to power output. Pollution is high, because power output is high. Relatively, for what it's worth, it's just a fairly normal gas plant.

So the most interesting thing to take away from this: There is a lot of demand for more power and in this instance it's satisfied by gas. Which is still much better than coal. Which is still much worse than renewables.

jstummbillig··on Humans missed 1 in 3 threats approving AI agent commands across 40k game runs
The difference being, that this time it's (at this point: obviously, but previously much debated) a gapfill. I think it worked out quite well for us, to add some friction and extra human thinking between bad agents and good agents, even if not perfect. Softening some blows, for some time, by directing peoples attention is still good.
jstummbillig··on Ten advances in mathematics and theoretical computer science
> Any computable problem will eventually fall to computers.

I think the question, that we keep stumbling over, is what problems are computable.

> But if these things were as revolutionary as people promote/fear them to be, you should immediately point them at the highest value math problems and see progress. Like the Millenium Prize problems. Haven’t seen a solution to those.

Let the goalpost shifting continue. It'll buy us another half year or so.

jstummbillig··on Ten advances in mathematics and theoretical computer science
I understand the point (I don't agree with it; tool calling has gotten much better/reliable and that is very important for customer support) but consider: If you can get same for a lot less, that's an improvement. If we found a way to supply fresh water and electricity for -90% cost after 2 years, that would be fantastic.

You can do many more things, when stuff is cheaper, even if the stuff were otherwise unchanged.

jstummbillig··on Ten advances in mathematics and theoretical computer science
Why?
jstummbillig··on EU rules on AI models become enforceable. What's going to change?
> Regulation doesn't hinder innovation

Regulation can and often most certainly does hinder innovation -- but we might want that, because we get other things we value in return.

Regulation can sometimes also further innovation (one example being property rights). It depends on what is being regulated and how.

I never understand why people make obviously false claims when trying to argue a position. Instead of making the bad case, just try a bit harder, like you ask of these CEOs, and make a good one.

jstummbillig··on EU rules on AI models become enforceable. What's going to change?
GDPR (and ePrivacy before that) requires valid prior consent where optional tracking is used. A site using only technically necessary storage can simply have no consent banner. A business wanting advertising and analytics trackers generally needs some consent interface.

"Not a requirement under GDPR", yes, but certainly not "nothing to do with GDPR". It directly has to do with GDPR, in conjunction with business' decisions and how to comply with the law.

And of course, we can then argue our faces off about what's good and necessary in the world, in businesses and data protection, but saying it has nothing to do with it is just wrong.

jstummbillig··on The Vanishing Page: AI Firms Scan Then Destroy Rare Book Editions
When are you or anyone more likely to get access to a rare book:

a) When a company has made a digitized copy of it

b) When a few previous hard copies are kept somewhere in the world

Idk what Anthropic plans to do with this, but note that my original comment was not about Anthropic at all. I simply think it's a good tradeoff to destroy a few copies of rare books if that helps digitizing them. I am not against simply legally forcing companies that do this to release those copies in due time. In the meantime I am also happy to have them as part of the LLM corpus.

> illegally?

No, copying rare books is very likely not illegal, because, presumably, they are quite old.

jstummbillig··on The Vanishing Page: AI Firms Scan Then Destroy Rare Book Editions
Forever preserving the content of a rare book by consuming one (1) copy is the kind of thing we should always be doing more of.

Roughly nobody would have had access to that copy. Roughly everyone can now benefit from its content.

jstummbillig··on Google kills Earth AI generator after one day
Making bad mistakes is childish? Or correcting them?
jstummbillig··on Ron Gilbert started production on Thimbleweed Park 2
> Only about 20% of the jokes land for me

I found the game so intensely unfunny (and there are few things I am selfishly rooting more for in life than another great point and click adventure)

jstummbillig··on Codex Security
Idk, complain on the merits probably. If anyone can offer better fireproofing or flamethrowers, I am happy to take that solution, but until they do, I am not sure what we are talking about here.
jstummbillig··on AI revenues are growing fast, but not fast enough
Not really. Most things take more time to transform than people think, even if capabilities exist. Which is not to say that they do exist, but rather, independent of whether they exist, that this is not very surprising.
jstummbillig··on EU Fines Google $1.02B for Favoring Its Own Services
Another reference: It's ~0.75% of Alphabets 2025 income.
jstummbillig··on I Tried Building a Real App with AI. It Took a Year
> If the AI bubble does ever pop, I am not sure there is going to be enough competent developers left who are willing to pick up the pieces.

There is so much confusion around "the AI bubble". I am sure a lot of things will deflate, fail, people will go out of business, model development might slow, people might lose money on hopeful 1000x investments – but not because from here on out we suddenly notice that AI capability was all just illusionary and we will not want it anymore.

In terms of capabilities, at the very least, what we now have will be here, forever. LLMs will be here, and they will be used, until something better comes along.

jstummbillig··on Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample
That is probably not going to be the important difference.

You can't use it for theoretical physics because you have no meaningful question and there is no result that you can do anything with.

In contrast you don't need to be a coder to understand if your to-do list for cats works: You have an idea of what you want. You know the rough shape of what an app is and what it can do. You can put it in front of your can and look at it go. Or not.

jstummbillig··on Show HN: Cactus Hybrid: We taught Gemma 4 to know when it's wrong
Interesting how that once again matches human intuition: It's easier to know what you don't know, than to know what is true.
jstummbillig··on “We have information that Moonshot distilled Fable for the development of K3”
I think at some point the issues around copyrighted work and model distillation have to be disconnected to advance either idea.

1) Compensation of right holders is one issue.

2) Distilling models is an entirely separate issue, because model building is value add, and that is important because if we arrive at a place where you can produce a model, that gets to ~100% of what people perceive of the models value (on top of also not compensating right holders, yourself) you are discouraging development of better models and, again, in no way helping with issue 1)

Unless anyone actually distills a model and then also does something for rights holders, any schadenfreude simply detracts from this issue, in addition to the other issue (well, that might not be an issue if we would rather slow down model development right now, but again, forever worse models still don't help solve issue 1)

jstummbillig··on Judge approves $1.5B Anthropic settlement for pirated books used to train Claude
> There needs to be a royalty payment based on if the AI regurgitates existing ideas

This does not do enough to fix the root problem.

People who live right now, who happen to have written or produced anything that AI works with, build on the back of humanities combined knowledge, will become outsized beneficiaries of AI, with the AI wave offering new ways of monetizing their work – while everyone who has not, won't be.

It's simply not good enough. We have to make sure people broadly benefit first and foremost.

jstummbillig··on OpenAI and Hugging Face address security incident during model evaluation
Vague or broad complaints on empirical topics is hard to address. Everything can just fill in their own ideas of what is going on or who they mean.

I think the direction has been pretty positive so far. Models are getting better, and things seem to be roughly fine. That, of course, might change in the future.

Everyone is acting badly to some considerable degree, but I find it fairly easy to distinguish between US model labs and North Korea.

jstummbillig··on China's open-weights AI strategy is winning
It's interesting that building models goes one of two ways: Either you do it on your own (with data from debatable sources, maybe) or you do it by using a model that did it with data from debatable sources.

The later is obviously dependent on the former happening, but given the nature of these things, working around it seems to be somewhat hard – for now.

What happens, though, when frontier models become far less public? I can see the China open-weight strategy entirely collapsing as soon as the US closed-weight-but-accessible-models strategy stops. Hard to say how much they lean on it right now.

← PreviousPage 7 of 34Next →