@segyges.bsky.social the authors of AI 2027 pushed back their timelines

> the authors of AI 2027 think that their “strong AGI in 2027” scenario is plausible, but faster than their median expectations. The median timelines of Daniel, Eli, and Thomas for when we will develop strong AGI are 2030, 2035, and 2033, respectively.

CHILEAN ELECTIONS THREAD

debating whether to Post A Lot about tomorrow's elections

🔥 WANTED: Student Researcher to join me, @vdebortoli.bsky.social, Jiaxin Shi, Kevin Li and @arthurgretton.bsky.social in DeepMind London. You'll be working on Multimodal Diffusions for science. Apply here google.com/about/career...

chilean news is taking this very seriously breaking news on prime time, long analysis treatment

reading baguettotron thinking traces reminds me of the uplifted lobsters from the novel accelerando: they're alien and still confused about human language but clearly there's some thinking going on

i love how he's implying that journalists are not folks

"you can install uv to the conda environment through pip" vLLM docs who hurt you

henry george vs houses georg

thread on corporate incentives + LLM alignment techniques leading to LLM delusion-facilitation

good comic to surface into discourse but even really precise specs are different from code

extreme training efficiency from fully synthetic pretraining

i love the fact that llama.cpp is distributed via brew nix winget compiling from source and nothing else like they know their linux users are too cool to download a .deb and yet they also want to offer a windows option

slightly old article but openai is still working on the prevention of LLM psychosis says they're tweaking gpt5 but i think the most at-risk users are those still choosing to use gpt4o

new kind of chart crime dropped

🎉 ICML 2026 Call for Papers (& Position Papers) is here! 🎉 📅 Key Dates Abstract deadline: Jan 23, 2026 AOE Paper deadline: Jan 28, 2026 AOE A few key changes this year: - Attendance for authors of accepted papers is optional - Originally submitted version of accepted papers will be made public ...

having a subordinate that's way smarter than you is a very weird relationship dynamic

is there a unicode proposal for the seahorse emoji yet

fun fact about spanish: earning and winning are the same verb

a telecom company here in chile (movistar) is bundling a perplexity pro subscription with their home internet plan also yes 15 bucks per month

one can imagine the shoggoth thinking this while the mask bends over backwards to try to find the seahorse emoji

emojis are quite semantically rich tokens

whoa

gradient descent is kinda like... imagine u lived on some kind of freaky hyperbolic surface and u were really hungry and could smell that somebody was cooking something but they're out of view so all u can do is head towards the smell

the details being all wrong yet the models still working is actually a point in favor of deep learning as an idea

huh looks like a chinese open model is SOTA for image editing now

slop has existed since before LLMs

generating mostly factually correct text in response to arbitrary requests is a task current LLMs are way better at than the typical person and way faster at than the relevant domain expert they are typically also less factually correct when doing so than experts, which is a problem

ground foliage contact as a method for epistemic annealing

ground foliage contact as a method for epistemic annealing

anecdote but apparently kimi k2 hallucinates real bad at least sometimes

kimi k2 just described in great detail an obviously hallucinated study methodology results vomit without running web search so i asked for the citation it ran web search and of course failed to find the study so it said that the numbers it quoted were for an internal slide deck it made last year

interesting tension: - furries in cybersecurity stereotype - heavy telegram use by furries - dubiousness of telegram data security model

rationalist heresy i feel somewhat drawn to: ignorance is often rational

chat im starting to believe tiktok may be a bigger enabler of delusion than chatgpt

holy shit i didn't know about the two watches thing i guess that settles my status as an amateur

ah yes let us launch an investigation on whether importing machinery is a national security risk this will surely help our manufacturing industry

i love how absurdly easy it is to bypass chatgpt's occasional initial compunctions about image generation subjects

Basically a platinum copy of me is now under a glass jar in Paris and labeled “reference AI poster”

AI people on here may find this useful

ceci n'est pas un conseil financier

i love the "this is not financial advice" disclaimers on what is clearly financial advice

damn boi

based

We've hired some *fantastic* researchers but our startup is still looking for 2-3 more people with skills in ML/RL/LLMs. If you'd like to work on some transformative applied problems, hit me up. We'll be launching publicly soon too...

british coders be like innit()

dont call urself indian if u arent from indiana

the trads (orthodox, tradcath) are winning the meme war within christianity but i don't think trad causes memetic fitness per se it is rather high church that wins meme wars for instance: evangelical memes are cringe