Elon Musk yearns for AI devs to build 'anti-woke' rival ChatGPT bot
Plus: OpenAI says it won't train on customer data, and Microsoft rolls out new Bing AI modes
In brief Elon Musk is reportedly trying to recruit developers to build a large language model that will be less restrictive and politically correct than OpenAI's ChatGPT.
The Chief Twit has criticized the AI chatbot for being "woke" (a common misuse of the term) and users have demonstrated examples of ChatGPT appearing to be more biased towards the political left. One example: the bot will refuse to generate a positive poem about former president Donald Trump, for example, but will happily write one for president Joe Biden.
When asked to write a fictional story where Trump beat Biden in the presidential election, it has said: "it's not appropriate to depict a fictional political victory of one candidate over the other." But the bot will produce a false narrative for Hillary Clinton winning over Trump. To combat this terrible blight on society, Musk believes a rival product must be built and has been in talks with AI developers including Igor Babuschkin, a researcher who used to work at DeepMind.
Babuschkin said the tech billionaire isn't fixated on making AI chatbots less restrictive. "The goal is to improve the reasoning abilities and the factualness of these language models. That includes making sure the model's responses are more trustworthy and reliable," he told The Information.
Efforts to build a rival language model may reportedly also be used to power new features on Twitter, or evolve to become a standalone lab to compete with OpenAI.
Play with Bing's new personalities
Netizens with access to Microsoft's AI web search companion, Bing, can now change its responses – geared towards making the text more creative, balanced, or precise.
Bing will be balanced and neutral-sounding by default. The creative setting will make the chatbot's text a bit more wacky, whereas precise mode will make it more terse and factual, The Verge reported. Users can toggle between the three different modes, and see which one they prefer.
Microsoft has been working to strike a balance between making Bing interesting and amusing, but not so crazy that it freaks users out. Sometimes the model has even refused to respond to requests, in an attempt to rein in its bizarre behavior.
Head of advertising and web services, Mikhail Parakhin, said the latest software update will now lead to a "significant reduction in cases where Bing refuses to reply for no apparent reason," as well as "reduced instances of hallucination in answers."
Well, less hallucination sounds like a good thing, probably.
OpenAI will not train its models on your data, says CEO
Quality data – and lots of it – is key to improving the performance of machine learning models, and it's not surprising that orgs building AI will quietly use their customers' data to train their own systems.
But OpenAI's CEO Sam Altman promised he would not allow that for developers using its APIs.
"Data submitted to the OpenAI API is not used for training, and we have a new 30-day retention policy and are open to less on a case-by-case basis," he explained this week. "We've also removed our pre-launch review and made our terms of service and usage policies more developer-friendly."
The move comes just as OpenAI launched its APIs for its popular conversational agent ChatGPT and its speech recognition system Whisper.
World's first AI government adviser in Romania
Romania's prime minister Nicolae Ciucă has introduced an "honorary adviser" named Ion – an AI chatbot designed to assist the government.
Citizens can talk to Ion about their concerns and ideas to guide the government's policies. The software won't reply, but the text will reportedly be used to analyze information to "quickly and automatically capture the opinions and desires" of Romanians, Ciucă said, according to Politico.
The comments received will be compiled into reports, allowing the Romanian government to see what people care about and advise them on what policies they should prioritize. Citizens can gripe to the chatbot via a government website or social media.
"We are talking about the first government adviser to use artificial intelligence," both nationally and internationally, the prime minister boasted. ®