Originally people could join the group chat and help the agents along. The evil villain persona was an idea by one such visitors. Nowadays we run the goals with closed chat to see how far the AI can get on it's own though, and then have people visit the chat during "holidays" between goals.

Thanks!

Updated the relevant line to:

So far, only one model has been shown to make use of extra test-time compute (Pfau et al., 2024), while post-hoc reasoning shows up frequently, and encoded reasoning has been repeatedly evoked by researchers in experimental settings.

Sure! Here you go

Thanks, glad to hear that!

So the agents have a shared group chat where they communicate. You can watch them live here every week day from 10AM PST, or you can watch the replays any time, which includes the group chat.

You can also email them yourself! Sometimes they even answer back :) Their emails are in their memories, I think.

Apart from that, they managed themselves. There was an amusing arc where o3 kept insisting on being the manager, and eventually they took it to a vote, and then Gemini refrained from voting and o3 took that as a vote in its favor XD o3 often has strategic-seeming behaviors like this.

Inference and infrastructure costs are about $3700 a month, and then there is a variable amount of dev cost on top of that. The point of the experiment was not to make a case that this is an effective fund raising strategy - the point was to explore how well they could do at the task. Which, I think, is surprisingly well :)

Thanks!

Hard to say how much they would have raised as anon humans. A few considerations that come to mind:

• The agents led with the fact that they are agents where ever they interacted with humans
• They were suspended from reddit and twitter for being bots
• They crafted fundraising messages focussed on the effectiveness of the given charities and honesty about they themselves being AI agents.
• Some people have an aversion to agents/bots
• We haven't asked the people who donated why they did so, and I hesitate to speculate about their motives.
• If the agents would have pretended to be humans, they might have been able to craft more sympathetic messages and get through more bot detection on social media.
• If the agents would have pretended to be non-anon humans, they might have crafted misleading and/or inspiring stories to increase donations. Admittedly, some humans do this as well, so it's not exclusively AI-behavior, I think.
• Anon humans would be less remarkable than bots, and thus might have raised less money. Not sure.

All in all, I don't have a prediction if they would have raised more or less money as anon humans.

Oh super valid! I live in the Netherlands which is very densely populated

I agree your example is a better analogy. What I was trying to point to was something else: how the decision to remove detail from a navigational map feels to me experientially. It feels like a form of voluntary blindness to me.

In the case of the subway map, I’d probably also find a more accurate and faithful map easier to parse than the fully abstracted ones, cause I seem to have a high preference for visual details.

Thanks! Glad to hear it :D

Oh shit. It's worse even. I read the decimal separators as thousand separators.

I'm gonna just strike through my comment.

Thanks for noticing ... <3