Ler

Hi I'm Tyler. I research shock physics and planetary interiors

[Ler @lerthedc]

@Noahpinion That's because it's basically permanently gerrymandered for Republicans

[Ler @lerthedc]

@bzogrammer Also, synapses are much more complex than weights

[Ler @lerthedc]

@GaryMarcus It makes me wonder if they have an upcoming IPO or something 🤔

[Ler @lerthedc]

@Noahpinion Huh, never thought I'd see you become anti-academia and anti-science.

[Ler @lerthedc]

@GaryMarcus @kevinroose Interestingly, Demis also says that he thinks AGI is just a few years away. Not sure what to make of that.

[Ler @lerthedc]

@LinkofSunshine I liked the ending, but I thought it was funny that they were basically like "ok yes this is the afterlife, but now you get to go to the *real* after life and who knows what that is like."

[Ler @lerthedc]

@Noahpinion No?

https://t.co/gOtnl4rxDK

[Ler @lerthedc]

@hutchinson I think it's because the original concept was very simple. Faceless warrior with a traumatic past finds some warmth and hope in mysterious little green child. They go on some adventures. The simplicity was part of the charm. But it also means it's hard to expand beyond that

[Ler @lerthedc]

@TrueSlazac @CantEverDie I mean, yeah they won't be exponential forever. Arguably, we're already off the exponential as companies are trading off performance in different areas. Just look at Opus 4.7. Apparently better coding ability but also apparently worse abilities in almost everything else.

[Ler @lerthedc]

@mattyglesias "The only way to stop a bad AI with a gun is a good AI with a gun"

[Ler @lerthedc]

@StatisticUrban What are they measuring exactly

[Ler @lerthedc]

@Noahpinion Obligatory

[Ler @lerthedc]

@rotting_garbage @monkey_reg Ilya Sutskever, Yann Lecun, and even Demis Hasabis have all mentioned that the path to AGI probably requires a new fundamental breakthrough rather than just more scaling. Also, ARC AGI 3 appears to really showcase how bad current models are at things like exploration and testing

[Ler @lerthedc]

@AlecStapp Its definitely going down, but not even close to being eliminated like the companies promised. I've spent plenty of time with Gpt 5.2 and Gemini 3.1 pro and I still get many hallucinations and confident misinterpretations.

[Ler @lerthedc]

@monkey_reg Regarding coding, the companies say that there will basically be no more human written code very soon. That's very different from the scenario where we continue to babysit and work along side AI coders for another decade.

[Ler @lerthedc]

@monkey_reg I use copilot with coding but AI is extremely inconsistent with scientific tasks. I've spent a lot of time trying to get Gemini 3.1 Pro to help with analysis, ideas, interpretation of data, and it's not great beyond basic lit search stuff

[Ler @lerthedc]

@rotting_garbage @monkey_reg Sure, but:
1. "Smart" can be defined lots of different ways
2. Given 1, it's not guaranteed that we get an intelligence explosion or anything like that even if the models continue to improve.

[Ler @lerthedc]

@monkey_reg ... about AI doing scientific research are incredibly overblown based on everything I've seen in several fields. I don't think we're getting AI that solves physics any time soon. Rather, we will probably get super helpful tools for lit review, coding, and writing

[Ler @lerthedc]

@monkey_reg I've heard so many conflicting things about how seriously we should take the AI companies' claims about coding capabilities that I'm not even thinking about that too much. Clearly it will have an impact but idk the magnitude. I'm a physicist and I just know that their claims ...