A former AI researcher at Google has claimed that the tech firm is training its propriety chatbot Bard on stolen ChatGPT data.
The top AI man resigned from his post earlier this year and made shocking claims about his former employer on the way out.
im not that annoyed at google for training on chatgpt output, but the spin is annoying
— Sam Altman (@sama) March 31, 2023
So long Google
Senior AI researcher Jacob Devlin recently switched allegiances from Google to OpenAI. In a parting shot to the company that formerly employed him, Devlin alleges that Google is playing less than fair by stealing information from ChatGPT.
According to Devlin, Bard is being trained on data taken indirectly from OpenAI. A spokesperson for Google has denied the claims.
“Bard is not trained on any data from ShareGPT or ChatGPT,” the spokesperson said to The Verge.
Devlin contradicts this denial. In conversation with The Information, Devlin says Bard is being trained on information taken from ShareGPT. ShareGPT is a site in which users share exchanges they have had with ChatGPT.
While training Bard on ChatGPT may appear to be a shortcut, Devlin warned executives at Google that their actions could prove damaging in the longer run. According to Devlin, training Bard from ChatGPT data runs the risk of making Bard little more than a ChatGPT clone.
Devlin also stated his firmly held belief that Google’s practices are in breach of OpenAI’s terms of service. What OpenAI intends to do about the breach is less clear.
Sam Altman seemed unconcerned with the story on Friday. The ChatGPT CEO took to his Twitter account to say, “I’m not that annoyed at google for training on chatgpt output.”
As others on social media were quick to point out ChatGPT is itself trained on data scraped from the internet.
Andrew Lacy’s response typified the response of users on Twitter.
“Good clarification,” said Lacy. “It would otherwise be ironical [sic] to get upset about someone training off chatGPT when openai trained off people-created data by scraping the web. What is good for the goose is good for the gander, no?”
The exodus
Devlin is not the only member of staff who has left Google to join the AI rival. As the race to chatbot supremacy shifts up a gear the demand for experienced researchers and developers increases. With a scarcity of talent the main contenders are poaching staff from each other.
According to research from LeadGenius and Punks & Pinstripes OpenAI hired upwards of 50 former staff from Google, as well as considerable numbers from Meta as well.
Many of the staff are not quiet about the transfers either, taking to social media to advertise their departure.
While many have made the switch Devlin is among the biggest scalps that OpenAI has claimed. The researcher worked at Google for 5 years and was the lead author of a significant research paper focused on training machine learning models for search accuracy. The paper has been credited with helping to initiate the AI boom, filtering into both Google and OpenAI’s large language models.
and then