@IsaacKing314 @Raemon777 @jessi_cata the guy who killed Shinzo Abe

@mattyglesias @eerac Thomas Massie?

@dwarkesh_sp isn't it too soon to say whether Deep Research can't make new connections? what's your standard for what counts as new, and have people seriously tried to use it to produce new insights?

@jerrrrrrryyyyy the idea that there was a "vibe shift" is going to age poorly I think

@lukefrymire If they use quadratic attention!

@tmkadamcz Agreed!

In other words, you can be concerned that AI will use too much energy, or you can be concerned that AI won't be useful. But you shouldn't worry about both at the same time! Eventually, people won't train and use bigger models if they aren't better.

Despite the ubiquity of discourse around this topic, there's been surprisingly little interest in revisiting this >year-old estimate. Our post is also very much not the final word, even if the AI field stops evolving as quickly as it has been these last few years.

RT @EpochAIResearch: How much energy does a ChatGPT query consume? One common estimate is 3 watt-hours per query. However, in this week’s…

@AndyMasley all the new models coming out made this kind of hard to write but it's up now!

@AlexGodofsky Trump 1 had the opposite effect on culture

@coldhealing this is what a residential neighborhood in an average American city looks like, not an average American suburb

@sam_d_1995 YoY comparisons are tricky. There's a gradual post-covid transit recovery happening in the background.

@tszzl this is kind of a strong claim in that operator is, what, 10-100x slower than humans now? And I'm pretty sure human observers can follow computer actions at up to 3x normal speed. so if operator speed can >triple in a month, will it reach human parity a few months from now?

@Simeon_Cps Or join a sham "stealth startup"

@chercher_ai @___frye 🌚

@asymmetricinfo This is also a misunderstanding. DeepSeek never claimed to develop their model with just $6m in total, or that the 2000 chips they used to train the model were all the chips they had. It was other people who heavily misinterpreted and distorted what DeepSeek said.