Technology

Musk’s xAI: twelve researchers poached from DeepMind, OpenAI and Google, a mission “to understand the universe”, and a grudge against the company he co-founded

Announced on Wednesday, xAI is Musk’s entry into the race he helped start: a company led by him, staffed by veterans of AlphaStar, AlphaCode, Inception, Minerva and GPT-4, advised by Dan Hendrycks of the Center for AI Safety, separate from X Corp but working with Tesla and Twitter. What is known, what Musk has said about OpenAI since he left its board, and why a man who signed the “pause AI” letter in March is building it in July.

Musk’s xAI: twelve researchers poached from DeepMind, OpenAI and Google, a mission “to understand the universe”, and a grudge against the company he co-founded

Google answered ChatGPT with Bard in March. On Wednesday Elon Musk answered both with a company: xAI, announced on Twitter with a website, a twelve-name team list and a one-line mission — “to understand the true nature of the universe”. Nothing else: no product, no model, no date. A Twitter Spaces session on Friday was promised to tell staff, and everyone else, what the company will actually do.

Who is in it

Musk leads it. The founding team is drawn from DeepMind, OpenAI, Google Research, Microsoft Research, Tesla and the University of Toronto, and its members’ credits are the greatest hits of the last five years of AI: AlphaStar and AlphaCode at DeepMind, the Inception architecture, Google’s Minerva maths model, and GPT-3.5 and GPT-4 at OpenAI. Igor Babuschkin, formerly of DeepMind and OpenAI, is the most senior. Dan Hendrycks, director of the Center for AI Safety — the group whose one-sentence statement in May compared AI risk to pandemics and nuclear war — is an adviser. xAI is legally separate from X Corp, the company that owns Twitter, but will, the announcement says, work closely with Twitter, Tesla and Musk’s other businesses.

The history

Musk co-founded OpenAI in 2015 as a non-profit counterweight to Google, put in tens of millions, and left its board in 2018 after, by various accounts, a failed attempt to take it over. Since ChatGPT he has called OpenAI “a closed-source, maximum-profit company effectively controlled by Microsoft” and “not what I intended at all”, and in March he signed the open letter calling for a six-month pause on training models beyond GPT-4. In April he told Fox News he would build “TruthGPT”, a “maximum truth-seeking AI”. In the same month he bought about 10,000 Nvidia GPUs and registered xAI in Nevada. The pause letter and the purchase order were signed within weeks of each other.

What it means

Three things. The frontier now has five well-funded labs — OpenAI, Google DeepMind, Anthropic, Meta and xAI — which makes a slowdown by agreement less likely, not more. The competition for the few hundred researchers who can train these models has a new bidder with the deepest pockets of all. And Twitter, which sat on one of the largest text corpora in the world and has been closing its API to everyone else since Musk bought it, now has an in-house reason for doing so. For users in Bangladesh, xAI is a name to file for now; whatever it ships will reach Twitter first, and Twitter is where it will be tested on the public.

Source: Gadgets Now, via Prothom Alo.

Source: Prothom Alo

Written by

Tech BD

Editorial team of Tech BD.