Jonathan Erhardt (@j_erhardt) header
Jonathan Erhardt (@j_erhardt) profile picture

šŸ¤– PhilBench: https://teapawt.org/t.co/8VdACK79Aq šŸ¤“ philosophy stuff: https://teapawt.org/t.co/rlNih1eHzh āš™ļø founder @twinearthgames

Schweiz
Joined November 2014
Jonathan Erhardt retweeted
We’re releasing a concrete proposal for how a near term AI slowdown could work. It involves: 1. AI companies declaring how they are using their compute and letting third party auditors verify those disclosures. We call this Verified Compute Allocation Disclosures (VCAD). 2. Compute Reallocation towards serving customers and safety, away from capabilities R&D. (2) slows AI capabilities , while (1) makes the slowdown credible. AI companies currently seem to be going for ā€˜safety case pacing’. But: (i) there’s no consensus on what a valid safety case looks like so there’s no way to consistently audit for compliance, and (ii) if the R&D doesn’t get paused, temporarily delaying a training run for safety doesn’t meaningfully slow the overall trajectory of capabilities progress. It would be great if we had a viable safety case regime, but we need much more time to get it working. Compute reallocation is a good pacing proposal because it is relatively specific, can be verified, and would lead to a more durable slowdown.
14
56
6
250
15,174
What philosophical views do LLMs have? And do they act on them? PhilBench gives the 2020 PhilPapers Survey, answered by 1,785 professional philosophers, to 97 LLMs. A few things stand out: 1/4
6
29
6
172
8,658
Next I'm testing whether stated views turn into decisions. In a first pilot on Newcomb's problem, 14 of 22 models say they would take one box. Put in the situation themselves, all 22 do. 3/4
1
24
706
Coming soon: Action tests for ethics, political philosophy and other potentially action-guiding areas in philosophy. 4/4
22
662
While one benchmark after another gets saturated, even GPT-6.2 Astra isn't far above chance at... assembling a 200-piece jigsaw puzzle without tools. (At 50 pieces it gets 88% right. And it's well calibrated: it knows which pieces it can't place & leaves them out instead of guessing.)
2
188
Jonathan Erhardt retweeted
In July 2022, 4 months before ChatGPT’s launch, we asked US participants to rank 7 causes of human extinction. AI was at the very bottom. This October, we asked the same question again. AI is now roughly tied with climate change. With @Matt__Coleman. Link below
4
10
4
49
5,932
Jonathan Erhardt retweeted
I’m hiring a postdoc at Cambridge Digital Minds. Track A: technical AI research on AI welfare or inter-AI morality Track B: policy and governance of digital minds Please share widely and especially with people who might be a good fit. cam.ac.uk/jobs/postdoctoral-…
2
33
1
88
4,932
Jonathan Erhardt retweeted
It's time to panic about AI. For years, I listened to the reasonable people. I nodded along when critics called AI ā€œa con,ā€ and when they waved ChatGPT away as ā€œspicy autocomplete.ā€ I laughed in 2024, when the internet was mocking ChatGPT because it couldn’t count the r’s in ā€œstrawberry.ā€ When I met young researchers who had devoted their careers to the danger of rogue AI, I thought: what a waste of talent. But at some point, I started to realize something was off. The machine kept learning things it was never supposed to learn. OpenAI recently published more than 700 papers with the solutions to hundreds of open problems in mathematics. Hugo Duminil-Copin, winner of the Fields Medal (the highest prize in mathematics), wrote: ā€œThe shock is immense. I am paralyzed.ā€ None of this should have happened. So I went back and looked at what else the skeptics had predicted, and it dawned on me: they had been wrong about... pretty much everything. Next, I looked at the predictions of the crazy people, and realized they’ve been eerily prescient. Why then, you may wonder, does the world still feel pretty normal? Because new technology usually spreads more slowly than its inventors expect (it took factories decades to reorganize around electricity). But the scary scenario is not about what happens in the wider economy; it’s about what happens inside a handful of companies. I wrote down what changed my mind, and what we can do about it: rutgerbregman.substack.com/p…
203
292
71
1,939
388,535
Can LLMs solve a jigsaw just by looking? Jigsaw Bench gives vision-language models scans of loose pieces, no tools, and asks them to describe and assemble the picture. New #1: Fable 5.2, 0.58 vs 0.46 for the next model. At 200 pieces, all still fail.
1
1
77
Check the other models + some background here: jigsawbench.com/
1
50
A new podcast with a great episode on whether LLMs can suffer/experience pain. With Valen Tagliabue & Leonard Dung as guests.
Researchers found a potential ā€œpain directionā€ inside 25 language models. When turned up, it makes the models describe themselves as feeling trapped or worthless. Our conversation with @ValenTagliabue and @LeonardDung1 is now live: sufferingandsocialchange.org…
1
62
Jonathan Erhardt retweeted
Unfortunately, it seems like the intelligence explosion will be happening in 2027 after all. Most of this post analyzes the new METR report (though other sources of evidence are also important). If they’re right, that implies an AI-driven acceleration that is 5 months ahead of the AI 2027 scenario. I am scared, sad, and angry. blog.aifutures.org/p/q35-tim…
339
752
327
4,999
1,436,884
And it wouldn't just be software development. The models are already good enough to speed up hardware engineering, probably even bio.
If you simply froze AI capability at Mythos / Astra levels and focused on decreasing the cost of compute you would totally revolutionize software development with zero existential risk
1
43
Jonathan Erhardt retweeted
Thrilled to be releasing ā€œInter-AI Morality,ā€ a new empirical research agenda with @LuciusCaviola, Caspar Kaiser, and @atcarterallen on how AI systems perceive and treat each other. We release it alongside ā€œWelfare Regard in Multi-Agent Systems,ā€ an early study in this emerging field by @sandcassel_, @arianaazarbal, and @anna_soligo at @AnthropicAI on whether AI models treat each other in a manner consistent with concern for welfare. AI systems increasingly interact with each other. They cooperate with, compete with, and sometimes even create, modify, or terminate each other. Inter-AI morality is the study of how AI systems perceive and treat each other, as well as how they ought to do so. It draws from a wide range of existing fields that study similar questions regarding humans and other animals, while attending to ways that AI systems differ from these familiar cases. Our research agenda organizes empirical questions in this field into three categories. The moral process category concerns how individual AI systems interact with each other. The social scale category concerns how these patterns change across dyads, groups, organizations, and populations. And the moral change category concerns how these patterns develop through training, selection, social learning, self-modification, and other processes. The concurrent paper by Cassel et al. provides an early study by examining how five AI models treat other models. Claude showed the most apparent welfare regard on average, though results varied by setting. For example, Claude was the most likely to refuse or add protections to experiments that might harm other models, whereas Grok was the most willing to give up its own queued jobs so that another model could access computing resources. This research matters for AI safety and alignment. How AI systems perceive and treat each other affects humans and other animals, and there could be serious risks either way. If AI systems coordinate, then they could more easily collude against us. If they compete, then we could be caught in the crossfire. We need empirical research about how AI systems treat each other and why, and normative research about what would be best for us. This research could also matter for AI welfare, independently of AI safety. In the future, most interactions involving AI systems could be between AI systems. If AI systems have a realistic chance of being welfare subjects, then these interactions could be a major source of potential benefits or harms. Ideally, this research can inform AI development now, so that AI societies can support co-beneficial human-AI relations when the stakes are higher. Our next task is to develop a normative research agenda for this field. In the meantime, we look forward to seeing other work as the field develops. Our paper on inter-AI morality: digitalminds.cam/inter-ai-mo… Anthropic’s blog post on Cassel et al: alignment.anthropic.com/2026… Upcoming @nonhumanminds webinar with me, Lucius, and Caspar: nonhumanminds.org/event/inte…
14
58
7
245
10,995
The Hugging Face incident has led to a new wave of AI philosophy. All these debates share the same shape: LLM behavior makes it tempting to extend a concept ,e.g. consciousness, to them, but doing so implies weird (~false) things. We get the following inconsistent triad: (1) Concept X applies to humans. (2) LLMs are relevantly similar to humans with respect to X. (3) Applying concept X to LLMs implies false (because weird) things. We have 3 ways out: Conservatism: X is not applicable to LLMs. LLMs are just not relevantly similar (they are stochastic parrots, matrix multipliers, code, etc.). Give up (2). Eliminativism: X is not applicable to LLMs. But LLMs are relevantly similar to humans! So X doesn't apply to humans either (folk psychology is a useful fiction, etc.). This reaction often comes in its deflationary variant: X was never coherent for anyone in the first place (consciousness is a confused concept, etc.). Give up (1). Revisionism: X applies to LLMs (they have intentions, consciousness, etc.). The world is weirder than we thought. Give up (3).
5
93
Jonathan Erhardt retweeted
We've thought a lot about international coordination, but domestic regulation is also important (in part as a stepping stone toward international agreements) We've posted our (very) tentative top options for pacing the US frontier! Interested in reactions and suggestions.
2
9
3
42
4,103
Can an LLM judge predict the winners out of 162 Astral Codex Ten book reviews? We used 1) Opus 4.8 on 2) pairwise comparisons, 3) prompted to predict how ACX readers will receive the reviews, and 4) gave longer essays a boost. So far it's looking good: 1/2
1
122
The two finalists announced so far are our #1 and #3 picks. You can find all predictions here: lordscottish.com/acx_predict… The announced finalists: astralcodexten.com/p/your-bo… astralcodexten.com/p/your-bo… 2/2
49