Pinned Tweet
New w/ @AISecurityInst & @UniofOxford:
Frontier AI can now out-persuade expert humans in conversation - incl. world-champ debaters and professional canvassers.
This held even when humans chose their topics, prepared in advance, and competed for £1,000 prizes 🧵
Frontier AI can now out-persuade expert humans in conversation - incl. world-champ debaters and professional canvassers.
This held even when humans chose their topics, prepared in advance, and competed for £1,000 prizes 🧵
58
240
984
219,429
Kobi Hackenburg retweeted
Today we’re launching the AI Trials Observatory: a living systematic review of randomized trials on how AI affects people.
So far we’ve surfaced 93 trials on two outcomes: mental health and belief accuracy. More trials and outcomes coming soon!
aitrialsobservatory.org
🧵👇
So far we’ve surfaced 93 trials on two outcomes: mental health and belief accuracy. More trials and outcomes coming soon!
aitrialsobservatory.org
🧵👇
3
41
122
9,430
Super excited about this new observatory, masterminded by @Ben_Tappin!
It will be an incredible resource for researchers, policymakers and members of the public who want an up-to-date picture of how AI is impacting people.
Highly recommend checking out the first release!
It will be an incredible resource for researchers, policymakers and members of the public who want an up-to-date picture of how AI is impacting people.
Highly recommend checking out the first release!
Today we’re launching the AI Trials Observatory: a living systematic review of randomized trials on how AI affects people.
So far we’ve surfaced 93 trials on two outcomes: mental health and belief accuracy. More trials and outcomes coming soon!
aitrialsobservatory.org
🧵👇
0
1
8
444
Kobi Hackenburg retweeted
One of the most-viewed PNAS articles in the last week is “AI can persuade people to take political actions.” Explore the article here: ow.ly/1OGk5102ceT
For more trending articles, visit ow.ly/e8AL5102ceS.
For more trending articles, visit ow.ly/e8AL5102ceS.
0
3
15
4,042
New in @PNASNews:
In two large experiments (~18k obs from ~14.8k people), we find that brief conversations with AI can get people to take real political action - incl. signing real petitions and giving money to charity.
Surprisingly, the persuasion strategy that worked best for udes - packing convos with info - was the least effective at changing behavior!
🧵
In two large experiments (~18k obs from ~14.8k people), we find that brief conversations with AI can get people to take real political action - incl. signing real petitions and giving money to charity.
Surprisingly, the persuasion strategy that worked best for udes - packing convos with info - was the least effective at changing behavior!
🧵
6
68
230
22,881
Some caveats:
• Participants were paid to talk to our models
• The behaviors we measured were fairly low-cost
• Donations came from a study bonus, not people's own pockets
• In the real world, attention may limit AI's influence more than how persuasive any one conversation is
• Participants were paid to talk to our models
• The behaviors we measured were fairly low-cost
• Donations came from a study bonus, not people's own pockets
• In the real world, attention may limit AI's influence more than how persuasive any one conversation is
1
1
5
667
Many thanks to the great team who helped get this over the line: co-first author @lukebeehewitt, @acarolinewagner @summerfieldlab @Ben_Tappin
Paper: doi.org/10.1073/pnas.2…
Paper: doi.org/10.1073/pnas.2…
0
2
4
674
True re: attention! And glad Anthropic is starting to share this data. But I struggle to interpret these numbers:
-IIUC, they're reporting share of conversations, not users. If a small % of heavy users have lots of other conversations (e.g. about coding), which seems plausible to me, then 0.51% of convos could mask a much higher share of users (the election-relevant quantity) asking about the election
-6 weeks before a midterm is still before most voters tune in
Hope a fuller release has per-user data across the full cycle!
-IIUC, they're reporting share of conversations, not users. If a small % of heavy users have lots of other conversations (e.g. about coding), which seems plausible to me, then 0.51% of convos could mask a much higher share of users (the election-relevant quantity) asking about the election
-6 weeks before a midterm is still before most voters tune in
Hope a fuller release has per-user data across the full cycle!
The attention constraint on AI persuasion is fundamental, as we argued in this chapter preprints.apsanet.org/engage/apsa/ar…
Just 0.51% of Claude conversations are about elections or voting; at most 0.01% about who to vote for
1
5
18
2,172
Kobi Hackenburg retweeted
🚨Published in @Nature: We show how AI can simulate human behavior to predict experimental results as accurately as experts
Our work has many implications for AI safety. For example, simulators can be used to aid harmful manipulation, and current safeguards do not prevent this🧵
Our work has many implications for AI safety. For example, simulators can be used to aid harmful manipulation, and current safeguards do not prevent this🧵
4
39
159
23,947
Kobi Hackenburg retweeted
🚨🚨New paper in @Nature with @AshuAshok, @lukebeehewitt & @ghezae_isaias 🚨🚨
Can LLMs predict results of social science experiments?
Across 70 preregistered studies, we find strong correspondence (r=.85) between LLM-predicted and observed effects.
doi.org/10.1038/s41586… 🧵
Can LLMs predict results of social science experiments?
Across 70 preregistered studies, we find strong correspondence (r=.85) between LLM-predicted and observed effects.
doi.org/10.1038/s41586… 🧵
16
148
456
71,693
Kobi Hackenburg retweeted
See for exciting new results on AI “super-persuasion” 👇
Excellent work by all involved, especially Kobi, who I increasingly suspect is an ensemble of humans/aliens living inside a human meat suit such is his magnificent output.
Excellent work by all involved, especially Kobi, who I increasingly suspect is an ensemble of humans/aliens living inside a human meat suit such is his magnificent output.
New w/ @AISecurityInst & @UniofOxford:
Frontier AI can now out-persuade expert humans in conversation - incl. world-champ debaters and professional canvassers.
This held even when humans chose their topics, prepared in advance, and competed for £1,000 prizes 🧵
1
2
8
1,690
Regardless of these uncertainties, we hope this work can be a starting place for discussion (and further research!) on the impacts these persuasive systems could have on society in the near future
1
0
19
2,883
This was a big effort - huge thanks to my collaborators @acarolinewagner @lukebeehewitt @Ben_Tappin @hannahrosekirk @HelenMargetts @summerfieldlab, and to many at @AISecurityInst for making it possible 🙏
Comments and feedback welcome!
Paper: arxiv.org/abs/2606.16475
Comments and feedback welcome!
Paper: arxiv.org/abs/2606.16475
3
0
66
3,032