"An LLM is just as deterministic as any other computer program" is not refuted by pointing out hardware limitations that would affect any other computer program implemented at similar scale (weather forecasts, or even just computing the average of a large stream of sensor readings).

I'm not really trying to 'refute' the original statement. Certainly, an LLM is just doing some calculations that can be done deterministically in principle. However, I think it's worth pointing out that there are practical barriers to doing those particular calculations both deterministically and efficiently. People who worry about LLM output not being reproducible aren't necessarily misunderstanding what an LLM is doing; they are responding to a real feature of most practical LLM implementations.

Agreed.

If you wanted to and had enough engineering effort to spare, you could run an LLM deterministically at relatively small impacts to performance.

One approach is to make sure you run things in the same order. Another is to change your operations so that more of them become associative or even commutative.

See eg the paper 'A Lattice-Based Approach to Deterministic Parallelism' for some interesting ideas on the latter.

That past certainly did exist, within living memory (mine, for one).

Well, your heart contains both mesoderm derived cells (enervation) and ectoderm derived cells (musculature).

The truth about tobacco was known for years but ignored because 1) industry lobbying and 2) the Nazis were the ones who figured it out, and pushed anti-smoking hard, so smoking became something "our side" did, 3) tobacco was still a huge US export.

It was easy to communicate, but we were motivated to not listen.

No the deaths were buried deep in the statistical data. It’s not apparent and we did not know either way.

Yes there were people who said it’s bad but their word was as good as bullshit because until we had the statistics there is literally no way to truly know.

No, the Germans had good statistical data linking smoking (as well as asbestos and other carcinogens) to cancer by the 1930s; this was known and ignored (or actively derided) by the allies. It's one of the classic and well known examples of ad hominem.

Congratulations, you are now quoted on the page!

> More out-of-distribution ways to think of things.

Openly admitting that it only caught fire once the Plagiarism-as-a-Service took hold is a good first step. It's the part that everyone knows and few will openly acknowledge.

"Outside the box" is out.

> If worries about AI safety and the end of the world entered your life recently, it’s probably because of a resignation letter.

No, it's because of an extensive and well coordinated FUD campaign. The resignation letter was just one salvo, and, like the rest, orchestrated.

I'm not a huge fan of calling LLMs "AI" and acting like they're our savior, but I'm even less of a fan of the brain-dead fearmongering.

That (the ZIRP period) was the anomaly; this is (historically) normal.

That's my point. ZIRP never should have happened.

Now we have a bunch of people who think that ZIRP was the norm. The industry's full of people who would not, and could not, have made their fortunes if they had to have gone to a bank to get another loan at 5% to make companies like Uber work. At some point, the answer would have been no, and they'd be just like every other failed business person instead of a member of the Cult of the Founder.

Bot farms may make up the bulk of "normal user behavior" at this point. It would explain at lot, actually.

I think you're probably right, and even a lot of the corporate "help" is bots.

Depends on your time frame for "relative to"; rates have been high compared to ZIRP-era, but still moderate to low when compared to historical norms.

I'm referring to ZIRP, since the OP was referencing cheap cash during ZIRP