OpenAI, the founders of popular AI chatbot ChatGPT, have announced their new, ‘human powered’ GPT‑6 model.
The new update will see users’ queries answered by 10,000,000 human employees, who will be allocated desks in the Stargate US data centres across the USA, while water cooling systems previously used for servers will now be adapted to drinkable water fountains.
It comes as users have reported high numbers of ‘hallucinations’, where previous computer-generated answers have falsified information. The company faced a defamation lawsuit in 2023, after radio journalist Mark Walters sued OpenAI over false embezzlement claims — which the company ultimately won — while yesterday (March 31) publishing giant Penguin filed a lawsuit over copyright infringement, alleging that ChatGPT reproduced a series of its childrne’s books.
There have also been concerns over safeguarding of users, with concerns raised over “chatbot psychosis”, where vulnerabilities users develop paranoia and delusion after using an AI chatbot.
The company hope that employees who are taught “real research skills” will be more proficient at spotting mistakes than its previous LLM model, while also being more aware of potential harm that people may face.
Those who ask ChatGPT to generate an image will now be answered by a member of OpenAI’s art team, who will create the image either through drawing an image to fit the brief, or setting up a photoshoot if specified as a photograph. Loading times are expected to increase, with the company predicting “three to five business days” at launch, which they hope to cut down within the first year of operation.
Sam Altman, OpenAI’s CEO announced the new update in a press conference. He said: “We are delighted to announce our latest update to ChatGPT — GPT‑6 — which should improve accuracy in search query results, while also allowing us to take a more human approach to care for our users. It providing 10,000,000 new jobs to ordinary Americans in a time of economic struggle.
“Human cognition is an incredibly powerful but underutilised technology. We’re excited to finally scale it. We believe this model unlocks incredible economic potential by connecting millions of humans directly to prompts,” he continued. “We’ve trained our human models on a wide range of sources including books, lived experience, and occasionally touching grass.”
Kim, 32, a reformed ChatGPT user, welcomed the change. She said: “It’s exciting. I have become tired of LLM-generated slop filling my feeds, and I’m excited for a human touch to return to art and writing. While the launch of ChatGPT made me think, I ultimately realised that everything I was looking at was hollow. It’s good to see people make art and write literature again.”
Al Goritham is Huck’s roving tech correspondent
Buy your copy of Huck 83 here.
Enjoyed this article? Follow Huck on Instagram for more from the cutting edge of sport, music and counterculture.
Support stories like this by becoming a member of Club Huck.