ViscountPenguin

Reddit is an absolutely hostile platform now, the amount of dogpiling is insane. Say the wrong thing and apparently you're simultaneously a bot, and a valid target for harassment (why people are so invested in harassing people they think are bots, I have no clue?)

Choice does a pretty good job, and is pretty in-depth.

Iirc, there's a big difference between the computational complexity of what a transformer can solve and what it can solve in practice; what it can solve in practice tends to come from things like singular learning theory (which describes the forms of generalizations that a given ML algorithm can make), but I think (iirc) that transformers can compute a pretty broad class of O(N^2) functions?

That definition also gives a nice introduction to e's taylor series!

Yes, most people have to use a toilet occasionally?

I can personally attest that having mild gi issues can make most of Europe a bit of a hellscape, with their expensive and simultaneously gross toilets. As an Australian I never appreciated clean public toilets before going there.

The power of bootstrapping is pretty insane really, I'm still impressed by the performance of random forests, and even bootstrapped confidence intervals are really cool.

That only matters if there's an incentive for LLM a to collaborate with a seperately trained LLM b in order to help the LLM b achieve its goals. So long as you don't let the evaluating LLM tinker with the weights of the evaluated LLM, and as long as you don't believe in very dubious ideas of acausal trade there's no reason to expect this to be an attack vector. For more certainty you can repeat the same analysis with different model classes or approaches aswell.

Huh, I wasn't aware; I haven't looked into the construction, I was just taking the blog posts claims for granted. Does it extend to all finite fields, that'd be super cool if so.

> Yeah I’m gonna guess 99.9% of people making that criticism against academic philosophy don’t know those names.

The majority of people who hold just about any viewpoint are going to be uninformed, so arguing against the average holder of any idea is going to be pretty trivial as long as you know a bit of the subject matter. The value of an idea is surely roughly proportional instead to the strongest argument for it.

Considering this line:

" In 1973, Jacques Morgenstern proved that you couldn’t get a Fourier transform faster than ½ n log n if you required that complex coefficients used in the computation had bounded magnitude"

I highly doubt this is going to somehow inspire people to make faster algorithms than the mathematical bound...

I have no problem with pure mathematics, my training was in it; but it seems very possible to me that most people funding theoretical computer science (as opposed to pure maths) probably assume the work has a more realistic computational model backing it up.

I think that argument made way more sense back when people were manually programming. Considering how small a delta staff cost will be compared to token cost, it makes way more sense in the new world to hire more expensive staff who better understand the nuance of your requests imo

Huh, apparently I "look like a bot". Must be my sick-as apm.

It really depends why wider society values people who are good at maths though. Unlike the arts, society (in my view) values people who are good at maths for purely utilitarian reasons; up until now strong mathematicians have been neccessary to meet various end goals. There's a very real risk that LLMs (or AI in general) will in the near term future surpass humans in all of those pursuits.

Fine-tuning language models for calibrated probability predictions has been a thing since the Bert era though, that's really nothing special.

If things we understand can't instantiate consciousness then I reckon we better shut down science as quick as we can, who knows how long it'll be before we understand something important!