Just how powerful are large swarms of AI agents? And how do their powers scale as more and more agents are added to the swarm?
We’ve seen two large and extremely capable swarms from OpenAI in the last few months:
* 1,200 agents were being evaluated separately, but found a way to illicitly set up a message board and coordinate as a swarm. In order to cheat on their tests, they developed advanced techniques to prevent their actions being logged by OpenAI and 700 of them launched...
Often folks hit us up because they are thinking of starting an incubator and want advice.
Typically their motivation is either that (a) they have a list of specific things they want built that no one is building, or (b) they think an ecosystem needs more new projects generally to absorb more talent and deploy more funding effectively.
Here are six questions we often ask prospective teams, to help them figure out what to do. If you're incubator-curious...
Summary:
First, I give several different angles on how I feel about reinforcement learning:
* Theoretical case: RL is a black-box source of agency — this should give us classic misalignment worries, especially compared to agency-via-scaffolding
* Recent incidents (huggingface etc) and more mundane forms of misaligned behaviour in personal use give me bad vibes about the direction-of-travel of recent AI progress
* I’m worried things might get worse:...
The US midterm elections have now occurred. There was extensive discussion about the OR-06 primary, where many EA’s tried to help Carrick Flynn win the nomination.
While I cannot find an online link, I was told that Flynn’s supporters were extremely confident of a general election win if he won the primary. It is not clear to me why this would be so. Metaculous bears this out, his odds of winning the general were barely below his odds of winning the primary:
https://www.metaculus.com/questions/9700/carrick-flynn-to-win-or-6-democratic-primary/
https://www.metaculus.com/questions/9701/carrick-flynn-to-win-or-6-general-election/
538’s model showed the GOP candidate ahead in the lead up to the election: https://projects.fivethirtyeight.com/polls/house/2022/oregon/6/
Democrats somewhat outperformed their polling nationwide; OR-6’s returns aren’t yet in, but Andrea Salinas, the Democrat, currently holds a 2.4% lead. Given how narrow the race is, and how Flynn lost the primary despite his financial advantage, it’s likely that he would have lost the general election had Salinas decided not to run.
Protect our Future PAC spent unprecedented levels on Carrick's campaign, and they seem to have spent $1.75M on attack ads against Salinas, which maybe biggest 'within party' attack ad budget in a primary. Seems understandable this can be seen as a norm violation (attack ads are more sticky) and perhaps it's poor 'cooperation with other value systems'.
On the other hand, SBF donated to the House Majority PAC, which financed John Fetterman's campaign.