this post was submitted on 26 Aug 2024
23 points (100.0% liked)

TechTakes

1491 readers
72 users here now

Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.

This is not debate club. Unless it’s amusing debate.

For actually-good tech, you want our NotAwfulTech community

founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
[–] dgerard@awful.systems 19 points 4 months ago (2 children)

Basically there isn't significant improvement to be had in the tokeniser, because it's already been trained on all the data on earth. So all they have left is overengineering.

[–] UnseriousAcademic@awful.systems 14 points 4 months ago (2 children)

Does this mean they're not going to bother training a whole new model again? I was looking forward to seeing AI Mad Cow Disease after it consumed an Internet's worth of AI generated content.

[–] dgerard@awful.systems 9 points 4 months ago

I think they will do whatever gets more investor cash

[–] anton@lemmy.blahaj.zone 8 points 4 months ago (1 children)

If you change the tokenizer you have to retrain from scratch, but you can do so with the old, unpolluted data.

It's genius if you think about it,* you can waste energy and tell your investors it's a new better model, while staying upstream from the river you pollute.
* at least for consultants, compute providers and other middle men.

[–] UnseriousAcademic@awful.systems 4 points 3 months ago (1 children)

I remember one time in a research project I switched out the tokeniser to see what impact it might have on my output. Spent about a day re-running and the difference was minimal. I imagine it's wholly the same thing.

*Disclaimer: I don't actually imagine it is wholly the same thing.

[–] dgerard@awful.systems 4 points 3 months ago (1 children)

there's a research result that the precise tokeniser makes bugger all difference, it's almost entirely the data you put in

because LLMs are lossy compression for text

[–] froztbyte@awful.systems 3 points 3 months ago

latent space go brrrr

[–] corbin@awful.systems 9 points 4 months ago (1 children)

Calling it now: codepoint-level non-tokenizing, with a remapping step to only recognize the most popular thousands of codepoints, would outperform what OpenAI has forced themselves into using. Evidence is circumstantial but strong, e.g. how arithmetic isn't learned right because BPE tokenizers obscure Arabic digits. They can't backpedal on this without breaking some of their API and re-pretraining a model, and they make a big deal about how expensive GPT pretraining is, so they're stuck in their local minimum.

[–] anton@lemmy.blahaj.zone 6 points 4 months ago (1 children)

But then it can't SolidGoldMagicarp SolidGoldMagicarp SolidGoldMagicarp SolidGoldMagicarp

[–] UnseriousAcademic@awful.systems 4 points 3 months ago

The only viable use case, in my opinion, is to utilise its strong abilities in SolidGoldMagicarp to actualise our goals in the SolidGoldMagicarp sector and achieve increased margins on SolidGoldMagicarp.