Axios AI+

March 05, 2025
I think I made the right choice when I signed up for Tara VanDerveer's basketball class through Stanford extension.
Today's AI+ is 1,198 words, a 4.5-minute read.
1 big thing: AI "Nobel" winners chose "unfashionable" path
This year's Turing Award — often called the Nobel Prize of computer science — is going to Andrew Barto and Richard Sutton, the pioneers of a key approach that underlies much of today's artificial intelligence.
Why it matters: Reinforcement learning, as the technique is known, posits that computers can learn from their own experiences, using a system of rewards similar to how researchers have trained animals.
In a joint interview, Barto and Sutton said the award is extremely rewarding, especially given that for much of their career, the technology they pursued was out of vogue.
- "When we started, it was extremely unfashionable to do what we were doing," Barto told Axios. "It had been dismissed, actually, by many people."
- "There were periods of time when I could not get funding because I was not doing the current fashionable topic, and I wasn't going to change to what was fashionable," he said.
- Sutton added that it was "particularly gratifying" to be given this award since it was Alan Turing who proposed the notion of computers learning from their own experiences in a 1950s paper, though it would take decades for there to be enough computing power to test out the notion.
Catch up quick: Sutton, now a computer science professor at Canada's University of Alberta, was Barto's student at the University of Massachusetts in the late 1970s.
- Throughout the 1980s, the pair wrote a series of influential papers, culminating in their seminal 1998 textbook: "Reinforcement Learning: An Introduction," which has been cited in more than 70,000 academic papers.
- The approach finally gained prominence in the last decade as DeepMind's AlphaGo began to defeat human players.
- Reinforcement learning from human feedback is a key method for the training of large language models, while the approach has also proven useful in everything from programming robots to automating chip design.
What they're saying: Google's Jeff Dean said reinforcement learning has been central to the advancement of modern AI.
- "The tools they developed remain a central pillar of the AI boom and have rendered major advances, attracted legions of young researchers, and driven billions of dollars in investments."
- Google funds the $1 million prize given each year to the Turing Award winners.
What's next: Both Sutton and Barto believe that current fears about AI are overblown, though they acknowledge that highly intelligent systems could cause significant upheaval as society adjusts.
- Sutton said he sees AGI as the chance to introduce new "minds" into the world without having them develop biologically, through evolution.
- "I think it's a pivotal moment for our planet," Sutton said.
- Barto echoed that cautious optimism: "I think there's a lot of opportunity for these systems to improve many aspects of our life and society, assuming sufficient caution is taken."
2. Musk's quiet deal with U.S. AI safety office
Elon Musk's xAI signed an agreement to work with the U.S. AI Safety Institute as his DOGE efforts have taken aim at the office's future.
Driving the news: Ben Buchanan, an AI adviser under the Biden administration, said on "The Ezra Klein Show" on Tuesday that AISI has a memorandum of understanding with Musk's xAI, the company behind Grok.
- AISI has MOUs with various AI companies like Anthropic and OpenAI that are designed to establish voluntary research, testing and evaluation collaboration.
- The MOU with xAI was negotiated and signed under the Biden administration, sources familiar told Axios. It's not clear whether AISI was involved in xAI's latest release, Grok 3, which came out last month.
- xAI and the National Institute of Standards and Technology did not immediately respond to requests for comment.
The big picture: AISI is one of the many government entities in the middle of dealing with tech billionaire Musk.
- Some industry watchers and former government officials told Axios that xAI being partnered with AISI is a good sign for the entity's fate under the Trump administration.
What we're watching: Unlike partnerships with other AI companies, the deal with xAI was never publicly announced.
- It's unclear whether xAI has offered the Grok system to the government for testing, as other companies have, or if xAI being involved with AISI will shift its safety efforts in another direction given Musk and the Trump administration's desire to hit the gas on AI development.
- In announcing Anthropic's latest release, Claude 3.7 Sonnet, the company mentioned pre-deployment testing of the model with AISI and the U.K. AI Security Institute as part of its partnerships with those two agencies.
Axios Tech Policy is covering every twist and turn in the White House and Congress' efforts to regulate AI. Get it in your inbox.
3. Apple unveils M4 MacBook Air
Apple today announced an M4-powered MacBook Air with a $100 price drop. It also refreshed the Mac Studio with faster chips.
Why it matters: The move brings Apple's latest chip to more of its lineup and the lower MacBook Air cost could help spur sales in a price-conscious environment.
Driving the news: Both new computers are set to be available for preorder today, with devices shipping next week.
MacBook Air
- The new M4 chip has a 10-core CPU and up to 10-core GPU which Apple says can make the M4 up to twice as fast as the M1 model.
- Beyond the faster processor, the laptop now supports up to two external displays and it comes in a new "sky blue" (aka light blue) color option.
- The 13-inch model starts at $999 ($899 for education buyers) and the 15-inch model starts at $1,199 ($999 for education), both $100 price drops.
Mac Studio
- The new Mac Studio comes with either the M4 Max processor or, at the higher end, an M3 Ultra chip.
- M4 Max-powered systems come with at least 36GB of memory, while M3 Ultra-based Studios come with 96GB of unified memory. With support for up to 512GB of unified memory, Apple says the M3 Ultra-based Mac Studio can run AI models with up to 600 billion parameters entirely in memory.
- Apple says it won't create ultra versions of all its chip generations because of the longer design times. That's why the Mac Studio gets an M3 Ultra instead of an M4 Ultra.
- The Mac Studio starts at $1,999 ($1,799 for education buyers).
4. Training data
- A federal judge denied Elon Musk's request for an immediate injunction to block OpenAI's planned restructuring into a for-profit company, and promised a speedy trial in the lawsuit starting in the fall. (Bloomberg)
- President Trump called on lawmakers to kill the $52 billion CHIPS Act last night in his address to Congress. (Politico)
- CoreWeave, which just filed paperwork for an IPO, is said to be in talks to buy AI software firm Weights and Biases for $1.7 billion. (The Information)
- The LA Times is pulling an AI feature that offered commentary on articles just one day after launch, after the bot defended the KKK. (Daily Beast)
- YouTube warned creators that an AI-generated video of CEO Neal Mohan is being used as part of a phishing scam. (The Verge)
- Amazon Web Services has formed a team focused on agentic AI. (Reuters)
5. + This
In Tuesday's Masterclass on basketball from Stanford extension, we were discussing the Caitlin Clark effect in growing women's basketball with, among others, her college coach Lisa Bluder (via Zoom) when Clark herself made a cameo.
Thanks to Scott Rosenberg and Megan Morrone for editing this newsletter and Matt Piper for copy editing it.
Sign up for Axios AI+







