scotty

Interested in general. All views are my own.

[Dmitry Lyalin @LyalinDotCom]

God damn it we do live in a simulation.

[scotty @oscottman]

@sergeykarayev This makes sense. Lots of good incentives here.

[scotty @oscottman]

@Tibo Show me the commands pls

[Chester @chesterzelaya]

Imagine matrix multiplication getting regulated

[scotty @oscottman]

@TrainGuyJeff @Jason Laws targeting development vs inference of models are very different though

[scotty @oscottman]

@olek_bio @Jason It would probably be a good idea to monitor models during development, and audit them before they’re released. Open source models could just be audited before cloud providers serve them

[scotty @oscottman]

@SicongJiang25 Internal research and experiments pose greater danger than model releases

[scotty @oscottman]

@jorgevayron @cryptopunk7213 Yes the labs can still develop in private, but they have more time to work on models before they are released. This is a good thing. Having increased transparency through this whole process is also a good thing. I’m not a Dario fan but this was a good move by him.

[scotty @oscottman]

@VitaliArbuzov @cryptopunk7213 You either didn’t read Dario’s essay or you’re rage bating

[scotty @oscottman]

@cryptopunk7213 I agree, what he is saying is very sensible and I commend him for it

[scotty @oscottman]

@kimmonismus RSI right now is just doing research work from codex/claude code, it’s not some super advanced new thing (yet).

[scotty @oscottman]

@LucienHeurtier 600 postdocs can’t crack Navier-Stokes in 88 hours

[scotty @oscottman]

@jenzhuscott Imagine it starts seeding its weights across P2P networks and potentially even running inference across P2P networks

[Andrew Wilkinson @awilkinson]

A fucked thought:

We are still in the punch cards phase of AI.

Year 3.

[Matthew Berman @MatthewBerman]

Remember when we were all laughing at AI because it couldn't count the number of Rs in the word strawberry? Good times.

[scotty @oscottman]

@ben_sage Have been thinking about this, and it seems likely that some people will spend large amounts of time studying ai outputs

[scotty @oscottman]

@kimmonismus Predicted this

[scotty @oscottman]

Prediction: frontier models will solve at least 1 more millennium problem by the end of next week

[scotty @oscottman]

@DaveShapi Have been thinking about this… and… ai will solve it

[scotty @oscottman]

Prediction: frontier models will solve at least 1 more millennium problem by the end of next week