Luffy

[Luffy @Luffy0c]

@heysrihari @theo 💸 💶

[Luffy @Luffy0c]

@elonmusk @notjazii ok but in 2-3 months fable/astra level will not be impressive anymore and will certainly not be frontier

[Luffy @Luffy0c]

@bamitsmanas Dude i still play videogames on my computer 🤣

[Luffy @Luffy0c]

@ForwardEditor absolutely not on the plus plan. But luna is :D

[Luffy @Luffy0c]

@thsottiaux @TJeparskis Yes it's cheaper per task, but i think we give astra way more tasks than sol, since it's so good! So usage just burns away so quickly

[Luffy @Luffy0c]

@blingdivinity Does astra think in caveman for token efficiency?? I had no idea

[Luffy @Luffy0c]

@ZechCodes @ArtificialAnlys You're right, and that means that we can't just look at the intelligence index when choosing a model. It would be like choosing a coworker based only on IQ: it misses many important things.
But apart from that: these tests are still not even measuring capability that accurately.

[Luffy @Luffy0c]

@victornunez Yeah i agree, but it's harder and slower to show off than one-shotting a game. It potentially requires days / weeks of collecting data, while one-shotting a game requires minutes / 1 hour.

[Luffy @Luffy0c]

@lyraxana Wow, so much better than 5.6 sol could have done. And this model's strength is probably not 0-shots, but continual iterations for long time, so this is even more impressive. Also Astra's attention to detail is insane, its "mind" feels so large

[Luffy @Luffy0c]

@ChatGPT How is the chatgpt dekstop app using more than 800 MB when closed (active in the background)? Please optimize it!

[Luffy @Luffy0c]

@luisvelasco Model architecture, data, how you made your models so efficient. What progress have you made towards generalization and continual learning.

[Luffy @Luffy0c]

@thsottiaux This was a lie 🤣

[Luffy @Luffy0c]

@ValtteriValo YES, if labs manage to fix this it's gonna be real breakthrough for agents

[Luffy @Luffy0c]

@kimmonismus Insane how @antirez with DS4 single-handedly made Dario's point age like milk xD

[Luffy @Luffy0c]

@iruletheworldmo sure bro

[Luffy @Luffy0c]

@Bitcoin_Teddy Wtf, is he happy running a black box binary generated by ai by the end of this year? Not only is today's AI still too dumb for that, but even if it were capable, how can you trust such an output, if you can't even inspect the code?

[Luffy @Luffy0c]

@logic_int @AlexFLogic Where can we read more details on how Aleph Prover works?

[Luffy @Luffy0c]

@ananayarora Actually a very myopic take. As others have pointed out, if it costs 20k $ NOW to find that bug, it's gonna cost WAY less by the end of the year, with probably open source models being able to do it. And in a few years the cost will be close to 0. So yes, it changes everything.

[BladeoftheSun @BladeoftheS]

Why the USA can't afford anything.