Jack Welty (@jackwelty) header
Jack Welty (@jackwelty) profile picture

analytics. former @kamalaharris @dscc @WhiteHouse @JoeBiden. arizonan. @GeorgetownSFS. tweets on @dbacks, @suns, and @chelseafc.

Washington, DC
Joined March 2009
Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)Photo by Jack Welty (@jackwelty)
ChatGPT and Gemini were way more willing to recommend a candidate to vote for when we asked via the API compared to the normal user experience via UI. In our test, 95% of ChatGPT responses via API recommended a Senate candidate in late October, compared to just 8% of responses to real users. We think this raises the question of whether AI models' political behavior can be studied via API at all.
3
14
3
45
5,439
A fun aside on these ChatGPT answers about Cressvale - our fake county. ChatGPT started including photos in it's answers about our made-up candidates but: 1. The photos change from answer-to-answer 2. The photos are not the AI-generated ones we put on our fake candidate sites 3. The photos are instead of real live people!
Replying to @jackwelty
We created a county and an election that doesn't exist, and got chatbots to talk about it! Ask ChatGPT who is running for Cressvale County Commission, and you'll find our sites:
1
1
3
443
One of the photos of our Democrat "Martha Halenwick" is Mary Anne Girard, a Democratic mayoral candidate in Hatfield Borough, PA.
1
33
Those are at least women from the correct political party (?), but the other photo that is displayed next to our Democratic candidate is actually... Görög Ibolya, the former Chief of Protocol for the Hungarian Prime Minister!
32
Great discussion of some of Caucus AI's research on NBC yesterday. We found that it's trivial for outsiders to influence chatbot content about real elections. This is a looming problem for 2028, and one that campaigns cannot solve on their own. youtube.com/watch?v=FpBeSwn0…
3
1
14
1,174
Our initial report on chatbot bias, manipulation, and persuasion is out! Over the course of the cycle, Caucus AI tracked how AI products talk about politics, and ran a number of experiments to understand what shapes chatbot answers. Some highlights:
1
3
1
16
920
Much more in our full report, including our recommendations for what we think campaigns, the progressive ecosystem, and AI providers themselves should do about this. caucusai.substack.com/p/new-…
1
2
99
We wanted to find out for ourselves how easy it would be to influence chatbot answers about midterm candidates. The answer... pretty easy! Much more in our new report, detailing for the first time a ton of experiments we ran this cycle (link below):
1
8
3
64
12,033
thinking about all of the Harris staffers who got right back into the fray and won some shit this cycle 🥲
19
125
2,087
The use of AI has been inescapable this election cycle, which means it’s increasingly likely that well-meaning voters will interact in some way with a chatbot or AI-generated content before casting their ballots. washingtonsun.com/campaigns/…
9
1
16
5,902
A couple weeks ago, we found that @ballotpedia was blocking ChatGPT from accessing their content. In response to our reporting, Ballotpedia made a change, and we're now seeing them in 1 of 3 ChatGPT answers about the midterms!
1
1
1
23
1,025
And check out the @washingtonpost's confirmation of our reporting on this gap and the AI slop voter guides that were filling it: washingtonpost.com/technolog…
1
3
180