hazbot

If open source models stay consistently <12 months behind the frontier, that shrinks the moat considerably; not for all use cases, but maybe enough to scupper profitability

> It’s true that so-called “hallucinations” are an artifact of LLM technology that can never be totally eradicated. But if you think about it, humans make a lot of mistakes as well. A calculator that gives you the correct answer 90% of the time is still an impressive technological achievement!

I know this article is meant to be dissing on LLMs, am I too caught up in the mania that this sounds to me unironically like a positive argument for LLMs?

Benevolent AI and biological immortality sounds like a good problem to have; at that point surely you only have yourself to blame for being boring

> Europe has been forced to depend on the US.

It's been more carrots than the (non-zero) sticks.

I think it's under selling Europe's agency to characterise the relationship as 'forced'.

And now that the US is increasingly seen as unreliable, Europe is (slowly?) decoupling - which is what you'd expect as a mutually beneficial relationship sours.

AI;DR very much an over-detailed dump of events without a good filter on the meaningful errors

> However, the paper itself and the formalization that it claims for support are not in a form that is readily usable to anyone who wants to understand or check it. This is not just because of the needless length or complexity of the argument (which is indeed overwhelming as presented), or because the formal repository is not in normal form, or because much of the text is misleading or poorly organized, but mainly because there is so much irrelevant puff to push through.

Zing

Not proof but it does waggle it's eyebrows suggestively!

Like sure the marketing is undoubtedly overblown. But in 2025 there was no amount of money you could spend to make such impressive but weirdly unreadable progress.

> How do we even know the premises of the "verified" Lean proofs are correct?

We let the experts investigate. If the results are dodgy, then the next batch of results will have to do more upfront work to demonstrate their worth. If there is gold in them hills, then this is exciting though very disruptive for the math community.

I do sometimes worry it's just that the LLMs got a better disguise

To be fair it's a plug from 1997 and the domain has since changed hands

I think it's fair to say the the parent comment's usage of 'crime' was likely in the informal sense, not in the legal technical sense.

How could you sleep with one of these in your house? How could you trust it around small children?

If we see rapid advances in robotics will alignment become a visceral concern for the population at large?

Love the 8601 evangelism.

I read Simon Singh's book as a young teenager - loved it!

It's because git replays your commits one at a time on top of a new commit; so if you resolve a conflict but then your next commits also touches the same code, the intent is ambiguous to git and it asks for your help with the conflict.

Wsc891 https://news.ycombinator.com/user?id=wsc981 has a good resolution for this in the comment directly above: 1) rebase -i onto your own history and squash 2) rebase onto the new branch.

I always wonder what the politicians are really thinking here.

Is Albo like "this is optics for the plebs, just a bit of window dressing bureaucracy on top of the normal market mechanisms where the data centre buying electricity will stimulate supply"

Or is he like "without these rules, data scenters will shoot themselves and the market in the foot and there won't be enough supply"?

Can confirm the Baldur's Gate 2 manual was an endlessly fascinating tome.

The generous interpretation of the OPs comment it was gentle hyperbole but directionally correct relative to the not so distant past.

Ahhhh but cancer treatment has gotten significantly better over the last 50 years.

Cancer research has a half-decent feedback mechanism that means fads eventually have to prove themselves or die out.

I suspect it is the feedback from the stricter compiler, not differences in training data between python and rust

> I feel like 20 or 30 years ago, the average person seemed to have more disposable income.

Big if true but is it true?

Totally - the (useful) economic models they are bringing are being extrapolated into very unusual territory!

If AIs become smart enough to outperform a critical mass of humans across the board, that's the singularity and whilst the economic model says all humans out of jobs means no consumers, the model isn't really valid anymore because the underlying assumptions are just broken too, so you can't take the model too seriously. Obviously it would be very disruptive but the demand destruction economic model just isn't making useful predictions anymore.

It's satirical argument against claims that computers will never be as intelligent as humans.

I enjoy having a DVD player for the kids, it makes curating their options so much easier.

Thankyou, this is the comment I came here for - advice on alternatives!

Yes, it is much easier to train someone to use AI than to train them to have sufficiently baked-in math and language skills to be able to leverage the AI.

Integration of the air cooling and water heating. For example, I have a air conditioner pumping heat out of my house right next to a box that's putting heat into the water coming into my house.

We are a happy paying customer at $DAYJOB.

Also, I looked into omfiles for a project recently, I was very impressed with their speed and blown away by the compression ratios!

Yes this is much stronger and relevant evidence