Jack Galler

“Exercise more. Take in more information. There is something inside of you that no one else can create.”

[Jack Galler @gallerdude]

Until you take what you find online and use it offline, it will direct your life, and you will call it fate.

[Jack Galler @gallerdude]

@Itsmezar0 @CylindrCZ FWIW at first I hated the rings game and then I locked in and LOVED it.

[Jack Galler @gallerdude]

@tracewoodgrains CATHARSIS probably my favorite both in terms of looks and personality

[Jack Galler @gallerdude]

@dwarkesh_sp The medieval historian lady you had one was really interesting! Maybe when she releases her heresy book?

AI wise, maybe some people on the policy side?

[Jack Galler @gallerdude]

I think about this a lot. In a post-singularity world, I would like to attend academic courses just to learn for the sake of the beauty of learning.

I’d like to spend an extended amount of time in Tokyo, I have been there three times and the design of the city is so gorgeous and has so much depth.

And of course, lots of time with family and friends.

[Jack Galler @gallerdude]

@Lewchube Great work, this really matters. Sugimori art always had some magic to it, and this REALLY makes it more clear. Love that subtle turquoise in Mewtwo’s design.

[Jack Galler @gallerdude]

@aidan_mclau this is a longshot but I’m gonna be in SF from June 3rd - 7th for LessOnline, would you wanna get coffee?

[Jack Galler @gallerdude]

@nickcammarata @davidad That was the day I bought $10K of NVIDIA stock. In retrospect, later than it could have been, but honestly still proud I was pretty early. I don’t think we’re even close to people outside our bubble having a Mythos moment yet.

[Jack Galler @gallerdude]

@tomieinlove For some reason the OpenAI camp has consolidated around “AI as a tool, not as a persona”, which I don’t think is what people will want at all.

Talking with someone is one of the most natural things we do.

[Jack Galler @gallerdude]

@thinkingshivers I just needed them to adapt the relativity revelation better in the movie, it’s like the most fun realization in the book. Instead we got the goodbye scene that took 15 minutes.

[Jack Galler @gallerdude]

I’m not really sure

1. “Free” still takes a lot of hassle to run. I have a 4070Ti, a great card! But it can barely eke out gpt-oss-20b tokens.
2. So far when customers have been presented with the tradeoff of intelligence vs literally anything else (price, “censorship”, speed, offline access, privacy, transparency), they have ALWAYS picked intelligence. Everything else is a distant second.

[Jack Galler @gallerdude]

@MarioBrothBlog I always loved that DKC happens on the Kong’s home turf, then DKC2 on the K Rool’s. It feels like it adds to that darker atmosphere

[Jack Galler @gallerdude]

@synthwavedd 5.5 mini, 5.5 nano?

[Jack Galler @gallerdude]

@tszzl Loved messing around with the playground. Hard to explain but if you asked me back then it felt important for AI… but hard to explain why.

[Jack Galler @gallerdude]

@sama more parameters

[Jack Galler @gallerdude]

@aniruddhadak @aidan_mclau Already there

[Jack Galler @gallerdude]

@aidan_mclau I have not thought about this nearly as much as labs have. But I think people will want persons, not tools. Human connection is wired deep into us, like Jobs connected our fingers with multitouch.

[Jack Galler @gallerdude]

@Kveldred Hey thanks! This article should cover it well for you:

https://t.co/LF3DGmKrxB

Let me know if you have any other questions!

[Jack Galler @gallerdude]

@isthisnessicary @flowersslop In that case, you might be right! Still CRAZY to me the idea they would do a new pretrain and not even mention it on the release page.

[Jack Galler @gallerdude]

@scaling01 Many people are saying this!

https://t.co/qdPz6S0Rtv

[Jack Galler @gallerdude]

@deedydas Great benchmark idea! Seems like a more generalized version of Epoch/METR’s “mirrorcode”.

The 0% scores are good for shock value, but it’s probably be better to see median with box and whisker charts for % of tests passed per program.

[Jack Galler @gallerdude]

@isthisnessicary @flowersslop The study also showed like a linear increase across all like 5.x models, and I don't think they did a new pretrain for every single one of them.