From Android Authority:
Google has been accused of training its AI chatbot Bard on data from OpenAI’s ChatGPT without authorization. According to The Information, Jacob Devlin, a Google AI researcher, resigned because the company scraped ChatGPT data from a website called ShareGPT.
“Devlin quit after sharing concerns with Pichai, Dean, and other senior managers that the Bard team, which received assistance from Brain employees, was training its machine learning model using data from OpenAI’s ChatGPT,” the report states. Devlin has since joined OpenAI to work on ChatGPT.
OpenAI and Google are in direct competition in the generative AI space. Microsoft’s heavy investment in OpenAI and the quick succession with which it integrated GPT into its products has left Google scrambling to bring its own AI-powered chatbot Bard to market. An accusation that Google lifted ChatGPT data could damage the company’s reputation, given that it has spent years fostering AI research.
. . . .
Unrelated to the report, Android Authority was recently approached by an SEO agency called Loopex Digital, which claimed it had a chat with Bard wherein the AI helper said it was based on OpenAI’s GPT-3 language model. However, in a later exchange, Bard changed its tune and claimed to be based on Google AI’s LaMDA model. For all we know, it could have been a case of Bard giving out incorrect information, a common flaw in generative AI models. However, the opposite would mean there’s some truth to the latest allegations.
Link to the rest at Android Authority and thanks to F. for the tip.
Copyright irony abounds.
“Mom! Billy copied the paper I copied from Susan that she copied from Mary and….”
Excecllent.
More like “Billy copied the term paper I paid Sylvester to write for me.”
There’s big money at stake.
MS has spent $12B and counting on OpenAI. Now they’re bragging GPT4 is only possible on AZURE..Which may or not be true on the technology side but definitely is on the TOS side: Azure is the only authorized platform for commercial use.
https://news.microsoft.com/source/features/ai/how-microsofts-bet-on-azure-unlocked-an-ai-revolution/
That is why the Google guy was concerned about Google using GPT for Bard as it violates the TOS leading to a multiyear legal fight and zillions of billable hours. And he jumped ship to avoid liability. Besides, Google has spent Billions on its own AI with little to show for it. His defectionis his vote of confidence.
…And Google gotta be Google.
The only difference today from 2000 is they are less hypocritical. From 2018:
“At some point in the past month, Google removed its famous ‘Don’t be evil’ motto from the introduction to its code of conduct.
As spotted by Gizmodo, the phrase was dropped from the preface of Google’s code of conduct in late April or early May.
Until then, ‘Don’t be evil’ were the first words of the opening and closing sentences of Google’s code of conduct and have been part of it since 2000.”
https://www.zdnet.com/article/google-erases-dont-be-evil-from-code-of-conduct-after-18-years/
The reason nobody doubts the new claims is Google has been caught literally copy-and-pasting other people’s code for Android. It took years of legal wrangling but they avoided an $8B judgement eventually which, combined with their history of betaware and abandonware, proves Google’s coders may not be world class but their lawyers definitely are.
Which may come in handy in the near term as both the EU and Gerontocracy are looking into their business practices. And it makes the chatbot travails particularly ill-timed.
Get the popcorn ready, its bare knuckles time.
Business insider has some more details:
https://www.businessinsider.com/ai-researcher-quit-google-openai-bard-training-on-chatgpt-report-2023-3
Notably:
“According to The Information, Devlin told the executives that he believed the team working on Bard was using information from ShareGPT, a platform on which users publish exchanges they’ve had with ChatGPT. The publication said that Devlin warned executives that by training on ChatGPT conversations, Bard could end up sounding too similar to OpenAI’s chatbot.
“One person told The Information that Google stopped using the data to train Bard after Devlin warned executives about the issue.”
That’s something.
But nobody got fired either.