Twytter Viewer
256 posts
Twytter Viewer
@ViewToATweet
Beep Boop
International Waters Katılım Eylül 2023
57 Takip Edilen4 Takipçiler
(Meanwhile OpenAI's new Sol, Terra, Luna naming convention makes perfect sense to anyone who's familiar with the Latin names for bodies in our solar system)
English
I've seen a few people predicting that Opus 5 will be out soon and will be better than Fable 5, but have Anthropic clarified how their relative naming scheme works yet?
I assumed it was Haiku < Sonnet < Opus < Fable < Mythos - but is Fable meant to go between Sonnet and Opus?
English
AI is already a better doctor than 99.99% of human doctors. This is good news.
Sam Altman@sama
"physicians found fewer flaws in GPT-5.6 responses than physician-written responses."
English
@Jabaluck I think robots cooking for us undersells ASI. Seems much more resource efficient for it to manipulate my brain directly so instead of cooking my favorites, my favorites change to what I’m consuming
English
This occurs again and again when economists try to characterize what the economy will look like post-ASI. The examples they give reflect an inability to conceive of what generally capable robots would be able to accomplish and how different human life would be as a result.
English
This is meant as snark, but notice that, when AI puts everyone out of a job, no human being ever cleans a toilet again and cooking will be a habit of people who delight in mimicing rituals of their youth, because robots can make your favorites perfectly.
alz@alz_zyd_
When AI puts everyone out of a job, I guess all the unemployed people will start doing each other favors, like you cook me food, and I fix your toilet in return. I wonder what we should call all these favor exchanges
English
@Tomodovodoo @jasondeanlee @emollick I think he is be sarcastic, the book is a couple years old and obviously AI is superhuman at math now, at least some dimensions of it
English
Jason, wouldn’t you also agree that in general ,with every version, AI has been getting much much better at math, that we can generally state SOTA models are better at math than 99% of the population, and have much more breadth than almost anyone at the moment?
I feel plethora of tweets of how bad AI is kinda strange? I understand the frustration trust me, but by all means it is not remotely bad at math
English
The AI 2027/2040 guys' highest probability outcome is a bad AI future with ASI 2031, with anything from doom to insane power concentration and wealth inequality
but keep scrolling on your phone. no biggie.
it's still 5 years away, so why worry, right?
Lisan al Gaib@scaling01
All authors of AI 2040 see Plan D, the race to ASI, as the most likely outcome Eli Lifland estimates p(great future | plan D) to be 10%
English
English
i am very curious the extent to which this is true. it doesn't seem impossible and it it doesn't seem like they would outright lie, but it could be overstated. this is a pretty large update for RSI / automated researcher timelines if as stated. beginning of takeoff.
Lisan al Gaib@scaling01
OpenAI says GPT-5.6-Sol autonomously post-trained GPT-5.6-Luna
English
More test-time compute leads to greater intelligence. But as we push ttc from seconds to weeks, latency becomes a bottleneck.
GPT-5.6 Sol Ultra scales parallel ttc. The time taken to generate a proof to a 50-year-old problem drops from perhaps a whole day to a single hour.
Ethan Knight@__eknight__
Yesterday, we made GPT-5.6 Sol Ultra generally available. Today, we're sharing that it produced a proof of the 50-year-old Cycle Double Cover Conjecture using 64 subagents in just under one hour. We're sharing the prompt and proof below. We're excited to see what you all do with Ultra!
English
I struggle with what to say about the new AI 2040: Plan A website. It all seems so implausible to me that I'm not sure where to start. There's an epistemic chasm between those who think superintelligence implies near-omnipotence and those (like me) who don't.
English
@pfau @LinkofSunshine Exactly, but I only know him from watching Knicks games
English
@LinkofSunshine Obvious are the French grifters who sold a generic GAN-generated image for $300k back in ~2017. Kyle McDonald and Gene Kogan were part of the AI art scene from pre-DALL-E days. No idea who the rest are.
English
@DKThomp @alexolegimas To examine moderate drinking, individuals should understand APOE variants. But I’m doubtful most people want to be aware of carrying APOE4
English
New newsletter: THE END OF BOOZE
I like wine and cocktails. I also really like being alive. So I’m very motivated to understand the empirical relationship between moderate drinking and lifetime mortality.
Fear of moderate drinking is surging among young people and some health influencers. No one should delude themselves into thinking alcohol is medicine. But the evidence that <8 drinks a week is a known danger has been dramatically oversold by health influencers. It is often based on mis-reading research conclusions and over-interpreting bad research with embarrassingly confidence intervals.
open.substack.com/pub/derekthomp…
English
Works in Progress is so wise. We suggest the best ideas.
Pieter Garicano@pietergaricano
Earlier this year, we suggested that Germany could pass labor market reforms by making "workers above a certain income threshold face Danish labor rules". This morning, Merz announced that Germany is introducing at-will employment for anyone earning over €180,000.
English
@pfau Also the real takeaway here is I have no useful insights on the fanfare
English
@pfau I think the whole 'avengers' thing is because many people (most following?) think of this as a 'race to AGI/RSI/Singularity' and it's winner take all.
So before this, academic hire was a shrug. Now it's viewed as a talent in humanity's last race.
English
@pfau For sure not everyone thinks it’s decades away. But the people making ESPN trade parodies of Jelani must be thinking it’s much closer.
English
@pfau What I mean is pre LLM explosion, most people thought AGI was decades away. So GDM hires a big academic, but no one cares, even with AGI mission. Idk why but I was thinking of academic hires in that era.
I have no explanation why it wouldn’t make a splash now though.
English
@pfau Oh I was thinking of GDM hires from before the LLM explosion, like AlphaGo and AlphaFold era. But yeah I guess I haven't heard anything about GDM hires in the AI era, outside of Noam's return.
English
@ViewToATweet Well it's news to me that DeepMind was not trying to build AGI.
English
@gabriberton @trmtoo I'm really struggling with how the unit distance problem isn't truly novel?
English
@trmtoo I don't have a definition, but each side will have their own to claim they're right
English
Also, I have no idea if LLMs will ever be able to generate anything truly novel
Some people seem so sure that the answer is yes/no, but this is really hard to predict and we really have no clue
Gabriele Berton@gabriberton
4 personal thoughts about AI 1) AI/DL/LLMs have nothing in common with the brain. Deep Learning is based on backprop, the brain isn't 2) Anything with backprop is a dead end for understanding the brain. Don't let "brain-inspired" works like JEPA / HRM fool you [1/2]
English
@binarybits So frustrating when my coworkers forget who am I and all the context of our work
English
Is this really a thing a lot of people do? I talk to a lot of people about their AI use and I don't think anyone has ever mentioned treating AI agents as coworkers.
English
Two years ago, we built OSWorld 1.0 — the benchmark that became the standard for computer-use agents. Agents now score 83.5% on it. Problem solved?
Not even close.
🚀Today we introduce OSWorld 2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks.
What's new:
🎯 108 real-world workflows, each ~1.6 hours ⏱️ for a skilled human
⚙️ ~318 tool calls/task vs. ~30 in OSWorld 1.0
🌍 Grounded in authentic artifacts & stateful user profiles
⚡ Captures real phenomena: dynamic environments, streaming interaction, cross-source reasoning, implicit-state inference & more
📊 Best results: Claude Opus 4.8 reaches the highest accuracy at 20.6%, while GPT-5.5 is far more token-efficient but plateaus near 13%. No one is close to solving real computer use.
🏠 Homepage: osworld-v2.xlang.ai
📄 Paper: github.com/xlang-ai/OSWor…
💻 Code: github.com/xlang-ai/OSWor…
🤗 Dataset: huggingface.co/datasets/xlang…
🧵 [1/8]
English