🎉 OpenAI Distances From Microsoft, 'Godfather of AI' Wins Nobel Prize, Meta Bests Sora, Doxxing Strangers, Google Reasoning
OpenAI Looking to Distance from Microsoft, Nobel Prize Awarded to AI Scientists, Meta Video Gen Leads the Pack, Meta Glasses Know Who You Are Looking At, Google Building New Reasoning Model
Welcome to this week’s edition of AImpulse, a five point summary of the most significant advancements in the world of Artificial Intelligence.
Here’s the pulse on this week’s top stories:
What’s Happening: OpenAI is reportedly looking to reduce its reliance on Microsoft for compute power and has started exploring options to set up its own data servers and secure AI chips independently, according to a new report from The Information.
The details:
CFO Sarah Friar told shareholders that Microsoft ‘hasn’t moved fast enough’ to supply computing power, causing the AI giant to look elsewhere.
OpenAI plans to lease an entire data center in Abilene, TX from Oracle, though Microsoft likely had to ‘bless’ the deal with its rival, according to the report.
OpenAI is also developing its own AI chip, which could lower costs for future computing clusters — its current supply is rented primarily from Microsoft.
Tensions have also reportedly arisen between OpenAI and Microsoft over the design and timeline of a massive joint data center project called ‘Fairwater.’
Why it matters: OpenAI and Microsoft’s relationship has felt a bit off for a while now. While both companies have leveraged each other well to ascend the AI power ladder, it certainly feels like there is trouble in paradise. There is plenty of smoke, and how this partnership shakes out could have fiery implications for the entire AI landscape.
What’s Happening: Scientists Geoffrey Hinton and John Hopfield were jointly awarded the 2024 Nobel Prize in Physics for their groundbreaking work in machine learning and artificial neural networks, which significantly laid the foundation for the current AI boom.
The details:
Hinton, often called ‘The Godfather of AI,’ co-created a method for allowing neural networks to learn from mistakes, influencing modern AI model training.
The 91-year-old Hopfield, currently a professor at Princeton, developed a network model in 1982 that mimics how the brain recalls memories.
Hinton later worked at Google, leaving in 2023 and becoming a vocal critic of current AI advances and sounding the alarm over the tech’s dangers.
Hinton reiterated his concerns at his acceptance, saying his research may lead to ‘systems more intelligent than us eventually taking control.’
Why it matters: While both scientists helped pave the way for today’s AI boom, neither seems particularly comfortable with what the future may hold for the technology. Hinton, in particular, has become the face of the ‘AI doomer’ movement since leaving Google, pushing hard for increased regulation and safety alongside other AI pioneers.
What’s Happening: Meta just announced Movie Gen, a powerful new suite of AI models for generating and editing video and audio content, positioning itself as a direct competitor to OpenAI’s Sora and other industry leaders.
The details:
Movie Gen consists of four models: a 30B video generation model, a 13B audio model, a personalized video model, and a video editing model.
The system can generate HD videos up to 16 seconds long from text prompts, along with synchronized audio like sound effects and background music.
Movie Gen also features video editing via natural text prompts and the ability to upload a reference image to create personalized videos.
Meta claims the model outperforms rivals like Runway Gen3, Luma Labs, and OpenAI’s Sora in human video quality and consistency evaluations.
Meta CEO Mark Zuckerberg said that Movie Gen will be ‘coming to Instagram next year’ in a post displaying some of the model’s sample generations.
Why it matters: Meta’s Movie Gen separates itself from other video generators by not only generating videos from text, but also being able to perform precise video editing. With the models coming to Instagram, it could transform the content creation process and give the masses a powerful video editing suite—with only prompting required.
What’s Happning: Two Harvard students just demoed a proof-of-concept system using Meta’s Ray-Ban smart glasses that allow the wearer to access personal information about strangers, raising major privacy concerns.
The details:
AnhPhu Nguyen and Caine Ardayfio combined Meta’s smart glasses with custom software, enabling the ability to ID people and retrieve personal data.
The system, I-XRAY, uses a combination of facial recognition, reverse image search, and LLMs to find names, addresses, phone numbers, and other details.
The students tested I-XRAY on Harvard’s campus, correctly identifying strangers and their personal info.
The privacy concerns come as Meta recently confirmed it may use any images and videos shared with Meta AI for training purposes.
Why it matters: This demo exposes how much privacy and surveillance are about to change in the AI age—and it is coming fast. If a couple of students can achieve these abilities with a pair of Meta smart glasses and publicly available tools, what will dedicated corporations and governments be capable of?
What’s Happening: Google is reportedly making significant strides in developing AI models with advanced reasoning capabilities similar to OpenAI’s o1 system, intensifying the rivalry between the two AI giants.
The details:
Multiple teams at Google are working on AI that can solve complex, multi-step problems, according to Bloomberg.
The AI uses chain-of-thought prompting, a technique created by Google, to tackle complex math and programming problems by ‘thinking’ before responding.
Google is taking a more cautious approach to its releases than OpenAI but has already debuted math-focused reasoning models like AlphaProof and AlphaGeometry 2.
Microsoft also infused reasoning capabilities into its Copilot assistant this week, leveraging OpenAI’s o1 model.
Why it matters: Human-like reasoning and agentic capabilities are clearly the two major developments on every AI firm’s roadmap, and the release of o1 may have signaled a new phase in the LLM race. The question is — will OpenAI’s speed keep it a step ahead, or is the competition for top-tier models about to get a whole lot tougher?