Weekly AI Recap: OpenAI drama abounds & UK regulators scrutinize Amazon-Anthropic deal

Along with Altman and Brockman, Musk helped to launch (and fund) OpenAI as a nonprofit in 2015 but departed in 2018 following a dispute with his co-founders. OpenAI announced the launch of a for-profit subsidiary the following year and soon secured many billions of dollars in funding from Microsoft.

Musk originally sued OpenAI in March, arguing that the company’s partnership with Microsoft undermined its ability to produce safe and open AI – the mission Musk supposedly had in mind when he helped found the company.

Shortly after that initial suit was filed, OpenAI published a blog post surfacing internal emails from Musk, which appear to show his desire to turn OpenAI into a for-profit company and even to merge it with Tesla (which he owns). He founded his own AI company, xAI, last year, positioning it as a competitor to OpenAI. The company said it was “sad that it’s come to this.”

John Schulman leaves OpenAI for Anthropic

The OpenAI-Anthropic pipeline continues to flow.

On Monday, OpenAI co-founder and alignment researcher John Schulman announced that he was leaving his post at the company and would be moving to Anthropic, with the intention, he wrote on X, “to deepen my focus on AI alignment, and to start a new chapter of my career where I can return to hands-on technical work.”

I shared the following note with my OpenAI colleagues today:

I've made the difficult decision to leave OpenAI. This choice stems from my desire to deepen my focus on AI alignment, and to start a new chapter of my career where I can return to hands-on technical work. I've decided…

— John Schulman (@johnschulman2) August 6, 2024

Schulman’s announcement follows the departures of Ilya Sutskever, an OpenAI co-founder and its former chief scientist, and of Jan Leike, its former head of alignment research, in May. Leike also took a post at Anthropic, which was founded by a cohort of former OpenAI employees.

Only three of OpenAI’s co-founders are still with the company: CEO Sam Altman, CTO Greg Brockman and machine learning researcher Wojciech Zaremba.

Altman (seemingly) teases OpenAI’s Project Strawberry

A tweet on Wednesday from OpenAI Sam Altman set a wave of excitement rolling across the internet about a secretive and supposedly highly advanced new AI system being built within the company.

The tweet included a picture of small strawberries sprouting from pots and the text, “I love summer in the garden.” Comments were ablaze with speculation that the cryptic post was a reference to Project Strawberry, an AI project that Reuters originally reported on last month.

The report, citing an anonymous source and leaked OpenAI documents, found that the primary goal of Project Strawberry is to build a model capable of abstract reasoning and long-term planning – both of which experts consider to be prerequisites for achieving artificial general intelligence, or AGI, a threshold that represents parity with or supersedence of human ability across a range of cognitive tasks.

A Bloomberg report also published last month found that OpenAI told employees that it is close to reaching the second phase of its five-step plan toward building AGI: creating AI models that can reason at a level comparable to that of someone with a PhD.

UK eyes Amazon-Anthropic partnership for potential antitrust violations

The UK’s Competition and Markets Authority (CMA) – the country’s competition regulator – announced Thursday that it will launch a preliminary probe into the partnership between Amazon and AI startup Anthropic.

Amazon began investing in Anthropic, an OpenAI rival, close to one year ago and has since poured $4bn into the company. The partnership has helped Amazon sharpen its competitive edge in the AI race, for example, with the April release of its Q chatbot.

The CMA has until October 4 to determine if the partnership poses a threat to competition in the UK and whether it warrants deeper investigation.

YouTube begins testing new Gemini-powered feature for creators

YouTube is experimenting with a new feature powered by Google’s multimodal AI model Gemini “with the goal of helping creators brainstorm video ideas, video titles and thumbnails,” the Google-owned company announced in a video shared on its Creator Insider page on Tuesday.

The feature, dubbed Brainstorm with Gemini, is currently being tested among a small cohort of creators before a possible more widespread launch, according to TechCrunch.

“We’re collecting feedback at this stage to make sure we’re developing these features thoughtfully,” the host of the Tuesday video told viewers.

For more on the latest happenings in AI, web3 and other cutting-edge technologies, sign up for The Emerging Tech Briefing newsletter.