Elon Musk finds some answers given by ChatGPT "extremely concerning". Now he reportedly wants to work on an alternative.
Multi-billionaire Elon Musk aims to launch a new AI research lab, reports The Information. This lab is supposed to work on an alternative to ChatGPT.
Igor Babuschkin, a scientist who most recently worked at Google's AI unit Deepmind and previously at OpenAI, which was co-funded by Musk, is expected to take a key position. Musk was one of the main funders of OpenAI, but stepped down from the board in 2018 to avoid conflicts of interest with his electric car company Tesla, which is also researching AI.
Musk's new project is still in the early stages with no concrete product plans. Babuschkin hasn't signed a contract yet, but says he'd like to "work with Elon on something in the LLM space."
Musk does not comment on the rumors
Musk, who is otherwise very communicative, especially on Twitter, has not yet commented on the issue. Initially, he was somewhat positive about ChatGPT's progress: "ChatGPT is scary good. We are not far from dangerously strong AI."
At the same time, he admits that the chatbot is too "woke" for him, based on interactions on Twitter. "Extremely concerning," he responded to reports that ChatGPT refused to define a woman, praised Democrats instead of Republicans, and claimed that nuclear weapons were less dangerous than racism. A closer look revealed the AI chatbot's left-leaning political stance.
According to Babuschkin, Musk wants to develop a chat AI that can argue better and is more objective. The model should be more trustworthy and reliable, he said. The goal is not to set fewer rules for the model.
In mid-February, Musk called for "TruthGPT," a model whose name implies that it tells "the truth" from Musk's perspective, rather than, say, meeting certain social expectations.
Extremely concerning
— Elon Musk (@elonmusk) February 12, 2023
Musk's desire to strengthen what he sees as free speech on Twitter was the motivation behind his purchase of the platform. Musk's possible intention to create what he sees as a more politically neutral alternative to ChatGPT would therefore not be surprising.
However, OpenAI may be ahead of him: The company has announced customizable ChatGPT models for the future. Users will be able to tailor upcoming chatbot models to their needs, within limits set by society, according to OpenAI.
Recently leaked information about Foundry's OpenAI development platform talks about "more robust fine-tuning options for the latest models" that will be available through Foundry. The future of large-scale AI models could therefore become more diverse and customizable, which could also mitigate criticism of political bias.
