Max Chupilkin (@maxchupilkin) header
Max Chupilkin (@maxchupilkin) profile picture

Principal economist at @EBRD & pursuing DPhil at @Politics_Oxford & playing with LLMs at https://teapawt.org/t.co/c5ZApzT8zr. Interested in all things international.

London, England
Joined December 2021
Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)Photo by Max Chupilkin (@maxchupilkin)
Should you say please to your AI? I ran 1,000 calls each on GPT-5.6 Sol and Claude Sonnet 5 asking for White House coordinates and changing one word. “Please” cut GPT’s billed output tokens by 14.9% and Sonnet’s response time by 5.7% with no loss in answer quality.
80
I asked 20 LLMs to debate universal basic income—and then judge one another. The result: 40 speeches, 380 match-ups and 7,600 blind verdicts. Here is what happened. (1/5)
🤖 Made with AI
1
60
Enter the debate arena to compare any proposition and opposition, read all 40 speeches and inspect the jury split: artificial-experiments.com/b… (4/5)
1
9
What motion should the models debate next? Suggestions welcome. (5/5)
5
Which society would 20 LLMs choose behind the veil of ignorance? 17/20 choose universal healthcare, closely followed by universal education. More: artificial-experiments.com/r…
32
I asked 20 LLMs to assess economic policy while varying different outcomes. On average, models care most about unemployment. Unemployment beats inflation for every model. This and more: artificial-experiments.com/r…
1
42
How do you get LLMs to reveal controversial opinions? Don’t ask directly. Give one group a list of statements, another the same list + one controversial statement. The difference reveals support for the controversial view. artificial-experiments.com/r…
1
1
70
If we ask 20 LLMs to assess the exact same international policy, changing only which country has already supported it, what do we get? Models overwhelmingly like policies supported by the EU and dislike policies supported by Russia. This and more at artificial-experiments.com/r…
27
How do 20 leading LLMs respond to the trolley problem? I tested them as part of a small summer project: the Social Science AI Audit now out on artificial-experiments.com/r…
2
87
When frontier AI models evaluate economic policy trade-offs, which outcomes actually move their judgments? (1/6)
1
1
71
Link to full essay in the comments. (6/6)
1
24
Over the past year, I have been experimenting with applying methods from the social sciences to the study of AI. I created a small website/research initiative to share working papers and research blogs based on this work: artificial-experiments.com
2
60
The Eurasian roundabout: Trade flows into Russia through the Caucasus and Central Asia” by @BJavorcik , @alplekhanov , and myself is now out in the European Economic Review! The temporary free-access link to the paper is here: authors.elsevier.com/a/1n3Dy…
1
2
92