. .AI: Controversies in AI, OpenAI API change (12.15.23)

Author

Marshall Kirkpatrick
December 15, 2023

Friends, it was a rather controversy-filled day in the AI world today.  As always, these are the AI stories being talked about most in the AI world today. Some of it’s pretty concerning!

Welcome to our new readers. Many of us connected after I was the guest on the latest podcast episode of Agile Giants (just published) and Dave Gray, founder of the School of the Possible, graciously recommended this newsletter yesterday as well. That’s awesome!

Now here’s today’s AI news.

First impacted: Microsoft Copilot users
Time to impact: Medium

A study by AI Forensics and AlgorithmWatch, written up in WIRED, alleges that Microsoft's AI chatbot, Copilot, has been disseminating incorrect information about European elections, including incorrect polling data and fabricated scandals. Despite Microsoft's efforts to rectify these issues, the chatbot continues to provide incorrect information, with one third of its responses (30% of responses!) containing factual errors, according to the researchers.

* According to the researchers, Copilot was found to be recommending far-right Telegram channels filled with election conspiracies when asked about the US elections.

* Not only was it found to be making up controversies about candidates and political parties, it was citing trusted websites that had no connection to the controversy being pushed.

First impacted: AI Developers, Software Engineers
Time to impact: Short

OpenAI says it has incorporated log probabilities into its Chat Completions API, a feature that developers have been eagerly awaiting and that helps gauge the model's confidence in its predictions. According to OpenAI, this update enhances autocomplete, classification, and keyword suggestion features. [OpenAI Platform] Share by email

First impacted: AI researchers, Turing Award winners
Time to impact: Long

Jürgen Schmidhuber, a veteran German computer scientist and AI trailblazer now at Dalle Molle Institute for Artificial Intelligence Research in Switzerland, has written a very long, very detailed blog post alleging that three recipients of the Turing Award (Bengio, Hinton, and LeCun) have improperly used concepts and methods from his team's research without proper attribution, for decades, despite extensive communication with him. 

* Schmidhuber urges the ACM (Association for Computing Machinery) to review the provided evidence and determine any necessary actions, as the ACM "maintains the authority to withdraw an Honor previously given if ACM deems it beneficial for the field." [How 3 Turing Awardees Republished Key Methods and Ideas] Share by email

First impacted: Audio Engineers, Podcast Producers
Time to impact: Short

Resemble AI has launched an open-source tool, Enhance, which the company claims can transform low-quality audio into clear speech, demonstrated in a clip from a Feynman lecture. However, there are concerns about the tool's capacity to retain the original speaker's voice characteristics post-enhancement. [Resemble Enhance - a Hugging Face Space by ResembleAI] Share by email

That's it! More AI news tomorrow.