@webhat @altomare

They use AI, it could just be a shit graphic that used their circle and colour for an idiotic pipeline logo. https://www.trust-in-soft.com/

@BeautifulSunPhotography Grief is entirely personal. The only useful advice I've had was a generic "don't feel guilty" from my Step-Mum (who was ex- military police-turned social worker before retirement). I now use it too.

@catsalad Although I do have a dormant mastodon.social account six days older.

@davep@mastodon.social

@simonzerafa @catsalad Ta da!

@catsalad Something like that 🤪

@catsalad I thought I was special!

@MiaMarkTwo @Nickiquote Oh dear, what a pity. Never mind.

@evan 27b models are now as good as huge frontier models of a year ago. I don't think the idea of just scaling will alter the fundamental issues with LLM ML.

@evan "they serve about the same function"

Citation? Naming them neural nodes doesn't count

@evan @fdrc Given these inherent weaknesses of LLM mathematics with respect to attaining anything approaching general superintelligence, the onus is actually on you to make a case for it.

@evan @fdrc

The point is that current LLM tech doesn't have the capacity for understanding truth or accuracy in the way we do. It's a statistical probability process. It's getting refined in some pretty cool ways, but that fundamental inability to actually understand is just dealt with crudely by harnesses in order to try to attain some form of deterministic approach, and that will never scale computationally to encompass anything approaching the human experience and breadth of understanding/intuition etc.

@fdrc @evan
This ️

@evan Not by variations on matrix multiplication. Unless they find some new type of maths I just can't see it not having really stupid non-human-like failure modes.

@evan Because matrix multiplication and activation functions will never be superintelligent, no matter how much deterministic code they use in harnesses to make it good at specific things.

@evan How about never?

@north @jim_easterbrook

With extra data exfil.

@timnitGebru Sorry about teaching my Grandma to suck eggs and all that 🤪

@timnitGebru To be fair, "OpenAI’s latest math breakthroughs commit research misconduct, experts say" is incorrect.

It's the people using it who did; they're inferring intent where there is none, giving AI attributes it doesn't possess.