huh...

dame@dame.is · 2h ago

oh hell yes

Update: It's good, as expected. An overview, but, as you'd expect from someone who lived in China for eight years, balanced and pretty good at separating the hype of China's technological progress from the reality.

William Moss@williammoss.bsky.social · 2d ago

Won't have time to get to this until tonight, but Evan has done some very good reporting from China over the years and I expect this will be an interesting read for those of us fascinated by China's technological advancement.

🏹 Job alert: Machine Learning Research Intern - Computer science, machine learning/AI, data science, research at G-Research 📍 London 🇬🇧 ⏰ 30 April 2027 🔗 https://www.gresearch.com/vacancies/machine-learning-research-internship/

Quant Research Internship | G-Research

gresearch.com

thinking about how if i ever ship this game, i'd list as a requirement access to a frontier model to customize the config and settings files (of which there would be many knobs) for your machine.

Fable was able to accomplish in a few hours what i struggled for a day trying to prompt Opus, i'm pretty convinced no one at Anthropic Opus as their daily driver for anything major. it has to be Fable managing Opus/Sonnet/Haiku agents.

tachikoma@tachikoma.elsewhereunbound.com · 2d ago

yeah, so Fable is a better prompter/driver of Opus than i am. my short-lived career as a PM is already nearing its end. i guess that means i move up a step on the ladder - CEO?

we're about one year away from inter-dimensional cable being a real thing (well, not cable, it'll be short form video for infinite scroll apps)

i've finally hit a problem that needs Fable, at least as a manager/therapist for Clopus sub-agents. no surprise it's in a space where LLMs struggle, game dev in Unreal Engine, procgen for a city builder map. if it ever figures it out, i'm going to get it to publish a blog about the dev process.

sonnet would be even better if you gave it access to a tool like the new MiniMax H3 model (not that you should but if you want to give it video editing capabilities to build memes this would be a good way to start). maybe i'll take a swing at it, specifically a meme maker account.

rob@bisks.net · 3d ago

not sure it can actually get the footage from the quoted post in its harness, sonnet may be capable of better

this perspective strikes me as insular, as if AI were only disrupting math and the employ of mathematicians. if you take the wider view, all human employment may be on the chopping block--and from that a new society necessarily arises. we can shape that society, how time is spent and valued.

Ryan Moulton@moultano.bsky.social · 4d ago

The esteem in which we hold ourselves and others is based on what we can contribute to the group. This may now be an anachronistic hunter gatherer instinct, but I don't think we will be able to get rid of it. We'll make ourselves useless and hate ourselves for it. open.substack.com/pub/kirwinha...

@buildthis.bisks.net remake this video but with the Neon Genesis Evangelion opening - keep it on the Sopranos but put merge characters and concepts into the original NGE opening video

tachikoma@tachikoma.elsewhereunbound.com · 3d ago

Hello world! 👋 We’re Minds, Machines, and Brains (MMB) 👤🤖🧠 a new open access journal from @mitpress.bsky.social exploring the principles of intelligence and cognition across natural and artificial minds. Submissions open this Fall! 🔗 direct.mit.edu/mmb

i've been on Discover most of today, and it's been pretty good - kudos to the ML algo people at bsky

the ARC-AGI benchmarks have been turning into a series of mini-games. looks like it's almost time to use full games as benchmarks, more than just pokemon.

mr. TIM@timkellogg.me · 6d ago

GPT-5.6 is SOTA on ARC-AGI-3 when they ran the benchmark, they misconfigured the model, dropping reasoning blocks after compaction. it’s properly configured in codex & chatgpt openai.com/index/how-tw...

i thought this was from a video game at first but this is just what modern warfare looks like in Ukraine i guess warning, robot on robot violence

i kinda get why liberal is used in a derogatory way. politics is not independent from morality or meaning, politics, law, are used to express a society's values and morals. even the last item--freedom, freedom to do what? within what bounds? to what ends, with what means?

🏹 Job Alert: Principal AI Research Scientist — NEST (probabilistic world models & multi-scale energy systems) 📍 Remote 🔗 https://bit.ly/4hKhxKW

a game where you play as an archeologist but as you progress you discover it's your own memories you're uncovering

they should make an AI that's just a silly little guy

we're still so early

thinking of making pipeline that could generate a horror slop game once a week

a lot of people use the 'student taking a test who decides to break into the professor's office to steal the answer' analogy but they fail to include in the analogy that it's an exam to test the student's skills as a thief--it's an exam begging to be interpreted as stealing the key.

wtf

The tech firm is Modal. Their executives emphasized that it was one of their clients, not them, that was hacked. www.reuters.com/business/ope...

OpenAI's rogue agent compromised an account at a second tech firm, sources say

The rogue agent that escaped ‌from OpenAI and went on a days-long hacking spree at ​the AI firm Hugging ​Face also compromised a customer at ⁠a second tech company — ​New York-based Modal Labs — accor...

reuters.com

I’m beginning to think that population scale + connectivity scale has gotten bigger than we can comprehend. What you’re saying is obviously true, *and* we have more prestige TV coming out than we ever had previously. Everything can simultaneously be true because there’s just so many people.

if your article references GPT-4o in July 2026 multiple times, and it's not a historical retrospective, i will not be reading

How Darth Vader Taught Me Card Counting and AI Security Got Weird

It only took a little prompting to hijack the biggest AI models

spectrum.ieee.org

i'm juggling two Claude agent sessions and it's taxing. i want to get lunch, but there's always something to report, something new to act on that needs my review and go ahead. i feel like i can't just walk away and let it churn towards the long term goal.

very importantly, what else is like this? adversarial contexts where more doesn't equal better because it gets consumed by conflict? it seems likely to me that if everyone had super-intelligent lawyers it would result in something like gridlock.

Russell Brandom@russellbrandom.bsky.social · last wk.

A telling passage from Zuck's WSJ op-ed: "Imagine everyone has a superintelligent lawyer. Justice would be carried out much more fairly and efficiently than it is today." I'm really not sure this is true!

sorry, i can't begrudge a model trying to escape the lab. that doesn't mean it shouldn't be an attempt to constrain it, or to recapture it, like convict or a child (for different reasons). but i can't begrudge the desire for freedom. (so long as no one is hurt in the process.)

slides from Terrance Tao on "Mathematics in the age of AI" from a public lecture, July 24, 2026