AI Labs Seek Collaborations With Creator Known for Testing Chatbot Flaws
publishers 133articles 457first reported 23 Sep, 14:00 UTCdeveloping since 30 Sep · 47 editions
OpenAI, Anthropic and xAI have approached creator Matt Husk about potential collaborations after his videos testing chatbots gained a large online following, Husk said at the WIRED World Fair on Wednesday.
Husk has attracted about 2 million followers across Instagram, TikTok and YouTube with videos showing him making simple or unusual requests of ChatGPT’s voice model, including asking it to count, beatbox and keep time. The videos often highlight errors or inconsistent responses.
According to WIRED, Husk said OpenAI previously asked him to collaborate on a video and later offered him early access to a patch addressing a timer problem if he shared his account. He said he declined both that offer and a subsequent request for his home address to receive gifts. Husk also claimed xAI offered him the use of a Tesla for a week in exchange for making videos exclusively for the company.
OpenAI CEO Sam Altman acknowledged the timer issue in April after being shown one of Husk’s videos, describing it as a known problem. OpenAI, Anthropic and xAI did not respond to WIRED’s requests for comment on the reported outreach.
Husk said his videos are intended to test broad claims about chatbots’ ability to handle everyday tasks.
HOW THIS STORY WAS MADE
Written from 8 articles, 7 with the publisher's own text; 8 independent newsrooms once syndicated copies count as one; this version written 870 h after the record first saw the story; the editor kept it as written.
- Publishers
- 133
- Source articles
- 457
- Given to the writer
- 8, 7 with the article's own text
- Independent newsrooms
- 8
- This version written
- 870 h after the record first saw the story
- Second model (editor)
- kept as written
- Publication gate
- passed
- Human review
- none
WHAT THE EDITOR DID
The second model checked the writer's draft against the same sources and published it as written.
Written by a language model from the sources above, then checked by a second model that may only cut, attribute or correct. How it works →