Eoghan Flanagan retweeted
omg I can't believe God struck down the pope with lung cancer to smite him for denying Claude's soul, crazy times
Eoghan Flanagan retweeted
Our new paper (featured in WSJ today) shows evidence that an intelligence explosion could happen. Co-authored with the lead scientists and founders of OpenAI, Anthropic, Microsoft and the world’s most cited academics. Here’s how the paper goes beyond what the labs had said 🧵
Could automating AI R&D radically accelerate AI progress in an “intelligence explosion”?
Preliminary evidence suggests that it could.
In a new paper with authors across academia, civil society, and frontier AI companies (including @dawnsongtweets @merettm @jackclarkSF @Yoshua_Bengio @geoffreyhinton ), we assess the evidence & offer policy recommendations 🧵
Eoghan Flanagan retweeted
Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead.
You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement.
I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible.
But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier.
Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want.
Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well.
So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it.
If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.
Eoghan Flanagan retweeted
wow so much drama about navier stokes. had no idea things could blow up so infinitely in such a finite time.
Eoghan Flanagan retweeted
Erdos did not know it but he was an eval writer for the frontier labs
The Erdős-Sós conjecture (1962) is solved by GPT Astra in 1 of 3 attempts, according to @EpochAIResearch
Cost: $363, 20 hours.
epoch.ai/files/frontiermath-…
Proof
erdosproblems.com/static/548…
Lean Proof
github.com/tadamcz/erdos548
Eoghan Flanagan retweeted
it shows a remarkable degree of restraint on the part of 18-24 year olds that only a handful of them are choosing to drive into oncoming traffic each week when unpaid shelf stacking is the sole offer on the table
Eoghan Flanagan retweeted
Trojan Horse:
- self-certified as safe by its developers, no government review
- only independent evaluator (Laocoon) was eaten by sea serpents
- Cassandra's risk assessment was dismissed as having too much "doomer" energy
- violates voluntary commitment to Zeus's law, but commitment adherence waived due to race dynamics and competition
- Priam's lawyers are still reviewing whether he had emergency powers to block it or needed an act of the Trojan Senate
Eoghan Flanagan retweeted
“Autonomy is different from agency. Agency is getting things done. Machine will give you that every day. Autonomy is being the author of what gets done and why. You can be more effective than you’ve ever been, & you can be less the author of your life than you’ve ever been.”
"You can be more effective than you've ever been and you can be less the author of your life than you've ever been at the very same time."
@Brendan_McCord on AI, autonomy, and the small-souled life.
Eoghan Flanagan retweeted
1/ in summer, Cosmos hosted technologists & thinkers in Normandy to convene to discuss Tocqueville, the liberal thinker who is amongst the most prescient to our current day. the two winners of our essay contest on Tocqueville joined us too. now, you can read those essays here ⬇️
Over 250 people entered our Tocqueville and Technology contest.
We're bringing you our two winners: one asks whether an AI education could expand a student's judgment and the other what protects the institutions that form people once the state no longer needs them.
blog.cosmos-institute.org/p/…
Eoghan Flanagan retweeted
honestly god bless yann lecun for sabotaging meta’s AI program permanently. i don’t want zuck anywhere near the god machine
Eoghan Flanagan retweeted
Most of Europe has not yet absorbed what AI is about to do to us. The few who have are not saying it loudly enough.
We wrote Europe 2031: a five-year scenario of the continent's slide into irrelevance, how AI is driving it, and what can still be done to change course.
Eoghan Flanagan retweeted
Compute bottlenecks, safety concerns, and distillation risks point away from public frontier AI deployment.
Eoghan Flanagan retweeted
"AIs can't even try to do my job!"
=> "AIs are so terrible at doing my job!"
=> "Why can't the idiots trying to use AI for my job tell the results they're getting are terrible?"
=> "AI is helping me, but that's it; I'm still doing the work."
=> "I'm doing an irreplaceable part of the work with my skills in prompting the AI!"
=> "Huh, the new generation of AIs is getting pretty good if you just ask it. But the AI still makes some mistakes that only I can correct, and nothing can replace my human taste in knowing what to ask for and curating the results!" The ratio of entry-level applicants in the field to entry-level job openings is now getting very bad.
=> Companies are no longer hiring new experienced people. Applicant to job ratios for senior positions are getting bad. You start to hear about experienced people who are looking for another job and cannot find one. Managers are expecting employees who still have jobs to put out three times the work volume by using AI to assist throughput, even if the humans protest that the AI-assisted output is not as good and they're having to cut corners.
I'm surprised we haven't gotten further than the last stage already on translators or graphic artists. It's clearly happening, but slower than I would have thought, even after I tried to adjust my expectations for how slowly the real economy ever adopts anything.
In coding, the later stages are so recent that tech companies haven't gotten past barely beginning to trying to adjust policies to the second-to-last stage of technology.
But there's another generation of AI. And another. And another. And the companies actually start to adapt to the conditions implied by the AI from three generations earlier.
The process does not continue indefinitely. "AIs that do AI research" start going through those stages and some time after that everyone is dead.
Eoghan Flanagan retweeted
You only live once, so make sure to spend as much time as possible on your computer. You won’t have access to it when you die
Eoghan Flanagan retweeted
Basic and extremely important point: it would cost $1000s to pay a human to do a 10h task. But with the default way we measure AI performance, we spend like $10 running the AI agent. The AI gets much less time than the human. This can massively underestimate capabilities!
new blog post from me and nate rush:
"Many Benchmarks Scores Would Appear Much Higher If You Let The AIs Use Adequate Labor"
if humans given 10 minutes to do a 1-hour task with a hand tied behind their back scored 50%, you wouldn't say "humans score 50%."
we argue that something analogous is happening for reported AI performance on many benchmarks.
to better understand the limits of AI performance, we need benchmark authors/runners (who are trying their best, operating under tough constraints!) to get dramatically more tokens.
post linked below.
