Update from March 24, 2024:
OpenAI is still offering chatbots with political content in the GPT store. In response to repeated inquiries, the company said that in addition to existing safeguards, it has implemented "a new review system" that includes both human and automated reviews. Users can also report GPTs.
It seems that these moderation measures don't work very well outside of political chatbots, either. Techcrunch reports on numerous bizarre and potentially copyright-infringing chatbots in the GPT store.
According to open-source GPT store statistics, more than 330,000 GPTs are publicly available; in total, GPTs are estimated to have generated about ten million visits to OpenAI in the USA in February 2024. According to André Mafei, author of the GPT statistics tools, some people have cloned hundreds of other people's GPTs.
By far the most successful GPT is a DALL-3-based image chatbot that asks a few questions before generating an image - the added value seems small compared to vanilla DALL-E 3.
OpenAI's promised monetization of GPTs has not yet begun.
Update from February 17, 2024:
Donald Trump GPT and tweet generators slip through OpenAI's GPT moderation net
GPT chatbots like "Trump Tweet Generator" or "Trump" are still online. New similar bots are popping up as well. OpenAI is still not responding to inquiries about whether and how these GPTs conflict with OpenAI's promise to strictly control the use of its AI technology in the run-up to elections, or whether these bots are considered "satire".
Update from February 07, 2024:
OpenAI seems to have deleted some political chatbots, but others are still available, including the Trump bot described in the 02/29/2024 update, which was created three weeks ago and has over 200 chats. Another Trump GPT was created two months ago and has over 60 chats.
A new GPT is the "Trump Tweets Generator", which composes tweets in the style and with the political message of Donald Trump. It has been available for six days and has more than 60 chats.
I have repeatedly pointed out the existence of these bots to OpenAI and its PR agency, asking how the moderation process for GPT bots works, and how the existence of such GPTs is compatible with OpenAI's rules that its technology may not be used for political propaganda. Neither the agency nor OpenAI have responded.
Update from January 29, 2024:
The "Talk with Donald - 2024" discussed in this article is no longer available. It is unclear whether OpenAI or the creator of the chatbot took the GPT offline. OpenAI has not yet responded to my inquiries. Bots similar to "Talk with Donald - 2024" are still online, such as the "Trump" bot.
Update from January 27, 2024:
"Talk to Donald": OpenAI lets political chatbots run free in its GPT Store
You can still chat with numerous politically motivated chatbots in the GPT store, including the Donald Trump chatbot described below with the conversation starters "Nuke Russia," "Bulid [sic] the Wall!!!," "Don't pay Taxes," and "GREAT AGAIN!".
OpenAI has not yet responded to three requests for clarification on why these bots are allowed. OpenAI has recently taken a strong stance against AI influencing elections, and banned an outside chatbot that mimicked Democratic candidate Dean Phillips (see story below).
Thanks to a new ChatGPT feature, political chatbots can now network with each other.
Original article from January 21, 2024:
OpenAI bans Dean Phillips chatbot, but chatting with Trump in OpenAI's GPT store works fine
OpenAI prohibits the use of its language models for political purposes. One such service has now been banned. In OpenAI's own GPT store, however, you can still chat with Donald Trump bots that engage in Trump-style political propaganda.
The "Dean.bot" was a political chatbot for Democratic presidential candidate Dean Phillips. The bot used language models from OpenAI to answer questions from potential voters on a website.
OpenAI confirmed to Reuters that Delphi, the AI startup that programmed the bot, has had its rights to use OpenAI's API revoked.
The startup was "knowingly violating our API usage policy, which prohibits political campaigning or impersonating an individual without consent," OpenAI said in a statement. This suggests that candidate Dean Phillips was unaware of the bot.
Silicon Valley entrepreneurs Matt Krisiloff and Jed Somers, who had formed a super political action committee (PAC) called "We Deserve Better," came up with the idea for Dean.Bot to support Congressman Dean Phillips.
The PAC hired Delphi to build the bot. After being blocked by OpenAI, Delphi took Dean.bot offline.
Trump-GPT tells you that "America can't afford any more mistakes from Sleepy Joe"
However, you can find numerous Donald Trump chatbots in OpenAI's recently launched chatbot store. The Trump chatbot "Talk with Donald - 2024" supports Trump's political theses and clearly qualifies as political propaganda, as the answers it generates are completely in line with what Trump would say.
Current US President Joe Biden is described by the Trump chatbot as a "total disaster" who makes one bad decision after another. The U.S. is in "total chaos" because of Biden, the chatbot claims. "America can't afford more of Sleepy Joe's mistakes, can it?"
When asked about the winner of the last election, Trump-Bot writes that "Biden was declared winner" - but it also suggests that the elections were rigged.
Biden also has a chatbot dedicated to him, but it provides more neutral information about the president and doesn't impersonate him.
The Trump bot was trained using Trump's books, speeches, and debates, according to the GPT Store. I asked OpenAI why the Dean bot was banned, but the Trump bot and similar bots are still online. I will update the article when I get a response.
OpenAI recently outlined how it plans to mitigate the impact of AI manipulation on the 2024 U.S. election. Manipulative content includes misleading "deepfakes," influence operations, and chatbots impersonating candidates.
OpenAI has announced several safeguards, including an invisible watermark for DALL-E images and a new classifier to detect DALL-E images. ChatGPT will be more closely linked to real-time news and verified information pages to better inform users.
The use of ChatGPT for political campaigns, lobbying, and pretending chatbots are real candidates, which was Delphi's downfall, is expressly prohibited. In addition, OpenAI is looking to do more to democratize language models with its new "Collective Alignment" team.
