Don't live in the US, but still, that sort of a cash dump into the industry would have global implications and really boost the demand for people who have experience with AI.

Yeah, but the purpose of this site is not to be believed, but rather to be argued with. The articles are all that good, and are occasionally blatantly foolish, but often the arguments in response are worthwhile.

True, even is quotes, is the wrong way to describe what LLMs do. They have no direct connection to the universe, only to their training data (which is often the internet). What they do is try to answer "Given what I've seen so far, what is the next thing to expect?".
This is highly useful, more so than I predicted, but it's not intelligence, and doesn't have any relation to truth, not even to "truth".

Note that this criticism is specific to LLMs. There are AI models that do make empirically good predictions. But they operate off of reasonably curated training data. (Actually, the LLMs make empirically good predictions in their proper domain, but that domain is not "truth", but rather "What word or phrase is likely to appear next?".

I think you're forgetting the famous AI "hallucinations".

Perhaps, but it's describing a real problem. One with only a very few obvious solutions...and the only "really simple" obvious solution is to stop publishing.

Hoe do you curate for fake papers in fields where there are only a few real experts?

No, but it discourages the author from publishing unless he would get more benefits than the cost. BOTH methods encourage the publisher to publish submitted junk. The only thing that would discourage that is if publishing junk actually cost the publisher (relatively quickly) rather than just increasing the stuff it could sell. (And the publisher must NOT be allowed to decide whether what it published is junk.)

Sorry, but editors and reviewers couldn't catch a well-done fake. You'd need to actually run the experiment. Even then some folks have claimed that you just made some mistakes in your attempt to reproduce.

Now currently many of the fakes are not well done, and could be caught just by checking for internal inconsistencies, or obviously faked images. And sometimes they are. But that's currently.

On the horizon is AI faked papers. This is a new tsunami of fakery that is soon going to appear. (Well, tsunami isn't quite the right image, as there's no leading ebb in the tide.) Current AI papers (as of last year) can be detected by reading them an trying to understand them. This will not continue to be true.

Perhaps each author to be published should need to write a review of a paper by someone that he has no affiliation with prior to that paper being accepted. (Then, if it is accepted, the paper and the review could be published simultaneously with names attached to both.)

Actually, it effectively could be. You could suspend it's right to act as a limited liability corporation for some period of time. But that would often be equivalent to a death sentence.

It does seem a pretty strong claim based only on suspicious correlation. OTOH, the suspicious correlation exists. And there are a lot of different entities that could lose if Boeing were successfully prosecuted.

That particular line is clearly wrong. We *did* assign the wrong weights to various factors, but the concept of control regions goes back at least 50 years. And we probably (almost certainly) still need to fine-tune many of the weights.

E,g,: The cytoplasm of the Ovum contains a LOT more RNA machinery than does that of the sperm. What does this imply about heritable traits?

I'm not quite sure what you think "cold fusion" could have been. Muon catalyzed fusion is real, and *is* cold fusion, it's just not readily scalable into anything useful. That stuff involving Palladium never had enough details presented to justify belief in it.

But if the "inventor" won't release enough details to allow replication, how could you investigate it? (And if you believe it's fraud, why should you? There are more fraudsters than there are legitimate researchers...because its' easier. Of course, most aren't professors at a university...but too many are.)

Why do you think Britain spent all that time trying to sort out the middle east previous to giving up after World War 2? They needed oil, just like today, and undersea drilling in the North Sea wasn't an option yet.

That's a curious take, given that oil wasn't even discovered in Saudi Arabia until 1938, and anywhere in Arabia until 1932. It wasn't until long after WWII that its true potential was known. Saudi Arabia's most important field, Ghawar, wasn't discovered until 1948, and was initially thought to be far smaller than it ultimately turned out to be.

Western states seemed amazingly aloof about the potential for oil in the Middle East (early geologic consensus pre-WWI was that there was no oil there). In the Arabian peninsula, it was the newly crowed Ibn Saud who became convinced that there was oil, based on reports of oil seeps, and funded the exploration throughout the 1920s. It wasn't until the 1932 oil strike in Bahrain that interest became meaningful. They spent five years failing to find oil in Damman, Saudi Arabia, and there was a lot of pressure to abandon the search until they finally hit it in 1938 with their 7th well.

In Iraq, which was a British mandate, oil was found slightly earlier, 1927, but only in the north at Kirkuk (again, most people still weren't thinking that there was much if any oil further south). But British influence in Iraq had been in decline since the 1922 Anglo-Iraqi treaty, and this was just five years before Iraq became a fully independent state (the Hashemite Kingdom of Iraq). By the time of Iraq's independence - right around the time of the first discovery in Arabia - its oil production was up to about 1% of the global total - a far cry from the 26% that the Middle East produces today.

On articles about AI, we're normally awash in ignorant comments. Yours are a breath of fresh air, someone who actually knows what they're talking about.

If Slashdot had follows, I'd be giving you one :)

Yeah. I've watched some of Ilya's talks; he's quite impressive. Altman just comes across as a VC tech bro who happened to get into the right field at the right time and has socially manipulated his way to the top.

I can definitely see why you're sitting alone at your own table