The American billionaire is said to be building a team of researchers specializing in artificial intelligence to develop an alternative to ChatGPT. The goal would be to enable the chatbot to “reason” so there is no need to impose so many pre-established limits.
Elon Musk, almost the richest man in the world – he was recently dethroned by Bernard Arnault – suffers from a “Existential fear linked to generative artificial intelligences”. That’s what he said in a tweet at least two days ago. For the record, these AIs have been front and center since late last year and the launch of ChatGPT in November. Conversational AI is everywhere, shaking the tech world or at least shifting the lines, threatening to shake up an established order in the world of online research, education or even the context of literary creation.
Also, isn’t it totally surprising to learn that Elon Musk is interested in ChatGPT in between two tweets or layoffs from Twitter, which doesn’t necessarily leave him much time. Unsurprisingly so, but funny considering he co-founded OpenAI in 2015 before retiring due to differing views.
According to The informationTHE “Tesla Technology” allegedly built a team of artificial intelligence researchers over a few weeks to develop an alternative to ChatGPT. The billionaire has already been critical of the chatbot, and more specifically the limits imposed by the designer in an attempt to control his reactions to users. In a conversation on Twitter last December, Elon Musk said: “The danger of training an AI to be woke up – in other words, lying – kills”.
But it is not that simple, because Elon Musk bases his words on technical arguments. In his view, linguistic models such as GPT, which rely on reinforcement learning through human feedback and evolve based on these interactions, suffer from a weakness. They reflect the prejudices of their developers. It was this analysis that prompted him to tweet about ten days ago “We need a TruthGPT”.
However, it seems that Elon Musk is not a chatbot, like ChatGPT but with less “control”. Even though it’s been a long time in the “tech age”, we’ve seen in the past what these types of attempts could yield. Think of Tay, the chatbot from Microsoft launched in 2016, and within hours became neo-Nazi, macho and fan of Hitler…
According to Igor Babuschkin, researcher in artificial intelligence recruited by Elon Musk for his project and interviewed by The information, “making a chatbot with fewer guarantees is not the goal” from his new boss. The engineer, who specializes in the machine learning that powers ChatGPT, and who left Deepmind last week, says that “the goal is to improve the reasoning and factual aspect of language models”. In other words, succeed in developing an artificial intelligence advanced enough to be able to provide more reliable and reliable answers without putting limits on it, but rather by “teaching” it to understand things, better process information that is used for his education and knowledge. We imagine quite easily that the task will not be easy and the road will be long before we get there.
The TruthGPT of the CEO of Twitter is therefore not expected in the coming days, weeks or even months. Igor Babuschkin has indeed confirmed to our colleagues from The information that Elon Musk and he had a discussion about setting up a team of researchers on this topic. He also said that he would like to “working with Elon [Musk] about something in the LLM field (for Large language models)”. He also and above all indicated that the project is only at the beginning. No concrete plan or specific product has yet been established or considered.
It is difficult to know where this resolution will lead Elon Musk. If his search is successful, maybe he wants to offer his alternative “ChatGPT” within a special application or site? Or maybe he’s considering integrating such AI into Twitter? With all the technical and ethical questions that entails, of course. As always, for better or for worse.
Source :
The information
Elon Musk, almost the richest man in the world – he was recently dethroned by Bernard Arnault – suffers from a “Existential fear linked to generative artificial intelligences”. That’s what he said in a tweet at least two days ago. For the record, these AIs have been front and center since late last year and the launch of ChatGPT in November. Conversational AI is everywhere, shaking the tech world or at least shifting the lines, threatening to shake up an established order in the world of online research, education or even the context of literary creation.
A competitor of ChatGPT
Also, isn’t it totally surprising to learn that Elon Musk is interested in ChatGPT in between two tweets or layoffs from Twitter, which doesn’t necessarily leave him much time. Unsurprisingly so, but funny considering he co-founded OpenAI in 2015 before retiring due to differing views.
According to The informationTHE “Tesla Technology” allegedly built a team of artificial intelligence researchers over a few weeks to develop an alternative to ChatGPT. The billionaire has already been critical of the chatbot, and more specifically the limits imposed by the designer in an attempt to control his reactions to users. In a conversation on Twitter last December, Elon Musk said: “The danger of training an AI to be woke up – in other words, lying – kills”.
Knowing his libertarian views, with variable geometry on freedom of speech – understand: everyone has the right to say what they want, but if they say bad things about me, I censor if I have the means – we could therefore quickly conclude that the founder of SpaceX envisions a conversation bot with fewer restrictions on certain topics.
But it is not that simple, because Elon Musk bases his words on technical arguments. In his view, linguistic models such as GPT, which rely on reinforcement learning through human feedback and evolve based on these interactions, suffer from a weakness. They reflect the prejudices of their developers. It was this analysis that prompted him to tweet about ten days ago “We need a TruthGPT”.
A more “reasonable” chatbot
However, it seems that Elon Musk is not a chatbot, like ChatGPT but with less “control”. Even though it’s been a long time in the “tech age”, we’ve seen in the past what these types of attempts could yield. Think of Tay, the chatbot from Microsoft launched in 2016, and within hours became neo-Nazi, macho and fan of Hitler…
According to Igor Babuschkin, researcher in artificial intelligence recruited by Elon Musk for his project and interviewed by The information, “making a chatbot with fewer guarantees is not the goal” from his new boss. The engineer, who specializes in the machine learning that powers ChatGPT, and who left Deepmind last week, says that “the goal is to improve the reasoning and factual aspect of language models”. In other words, succeed in developing an artificial intelligence advanced enough to be able to provide more reliable and reliable answers without putting limits on it, but rather by “teaching” it to understand things, better process information that is used for his education and knowledge. We imagine quite easily that the task will not be easy and the road will be long before we get there.
The TruthGPT of the CEO of Twitter is therefore not expected in the coming days, weeks or even months. Igor Babuschkin has indeed confirmed to our colleagues from The information that Elon Musk and he had a discussion about setting up a team of researchers on this topic. He also said that he would like to “working with Elon [Musk] about something in the LLM field (for Large language models)”. He also and above all indicated that the project is only at the beginning. No concrete plan or specific product has yet been established or considered.
It is difficult to know where this resolution will lead Elon Musk. If his search is successful, maybe he wants to offer his alternative “ChatGPT” within a special application or site? Or maybe he’s considering integrating such AI into Twitter? With all the technical and ethical questions that entails, of course. As always, for better or for worse.
Source :
The information