Raymond

@rayd

24 Following

43 Followers

July pfp

I think good taste is also acquired by building your own sense of aesthetics. What I mean by that is this idea that you need to have a build a highly sensitive radar for what you like and what you don’t like. Something that is much easier to say than it is to do, yeah.

14 replies

3 recasts

50 reactions

Raymond pfp

Making a website and telling all your friends you work at a startup

0 reply

0 recast

1 reaction

Raymond pfp

1 reply

0 recast

1 reaction

Raymond pfp

spot the odd one out

1 reply

0 recast

1 reaction

Raymond pfp

the great irony of AI coding is that it helps so much with using niche libraries but only once they've been around for a while :(

0 reply

0 recast

2 reactions

Raymond pfp

no, super simple! wallet got created on my phone automatically afaict, just had to set up the profile. even syncing up my laptop was a breeze. main friction was working out in practice how to engage with people, which you helped with a lot

1 reply

0 recast

1 reaction

Raymond pfp

every second was worth it :')

1 reply

0 recast

1 reaction

Raymond pfp

the social proof is real

0 reply

0 recast

1 reaction

Raymond pfp

if you want to really understand hallucination, you gotta read feynman's lecture on how to not fool yourself https://calteches.library.caltech.edu/51/2/CargoCult.htm

2 replies

1 recast

2 reactions

Raymond pfp

nothing like the rush of joining new platforms and just sticking the hose pipe in your mouth until you figure out the vibe

1 reply

1 recast

1 reaction

Raymond pfp

what's a big question about the next 10 years where you're 50/50 on the answer

1 reply

0 recast

1 reaction

Raymond pfp

Hey! Lovely to be here

1 reply

0 recast

1 reaction

Raymond pfp

Second is on taking tasks where larger models do worse (like prompt injection) and helping models do better (30% gain) by treating them as mixtures of distributions, and downweighing the bad distribution

1 reply

0 recast

1 reaction

Raymond pfp

Glad to be here! First paper is on why base models doing offline prediction are theoretically doomed to hallucinate etc until they get external feedback, and how the same underlying property makes it hard for them to generalise strategies better than their training data (1/-)

1 reply

0 recast

1 reaction