Microsoft and OpenAI have AGI figured out ... sort of

January 2, 2025

Good morning, and happy New Year.

Hope you all had a lovely holiday (and if you had any AI-related arguments with family members, hope we helped prepare you to win ‘em).

We’ll be taking a look back at 2024 (and into 2025) next week. In the meantime, things have been a little busy since we last spoke. Let’s get into it.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

🤖 AI for Good: Robotic gloves
📊 xAI raked in $12 billion in funding this year
💰 Microsoft and OpenAI have AGI figured out … sort of

Source: ETH Zurich

The idea of AI-enabled robotic exoskeletons — which we talked about last week — has a number of iterations. While one of them does involve full-body suits, others are a bit more targeted.

What happened: For several years now, the Rehabilitation Engineering lab at ETH Zurich has been developing a robotic glove to assist people with hand impairments.

The details: Several diseases, including strokes, cerebral palsy and spinal cord damage, can result in impairments that make it difficult for those affected to use their hands. The idea of the device — called the RELab tenoexo — is to re-enable full-hand use in those impacted populations.

The device itself is a lightweight glove, loaded up with motors, controls and sensors that can be tailored to a user’s specific situation.
The key component of the glove, however, is the algorithm that makes it work, which is designed to detect intent around physical motion, which triggers the functionality of the glove.

Though still a prototype, the glove promises to aid users both in physical therapy and in their daily lives, offering a means for people to regain their motor function.

Source: Tesla

Elon Musk’s xAI said Dec. 23 that it had closed another $6 billion funding round, which included participation from a16z, Fidelity, MGX, Nvidia and AMD, among others.

The details: The money will be used to accelerate the startup’s AI infrastructure build-out, which prominently features a 100,000-strong GPU chip cluster in Memphis, Tennessee, a cluster that xAI plans to shortly double to 200,000 chips.

The firm is currently training Grok 3, the latest version of its Grok series of chatbots, which are accessible through Musk’s X platform.
It’s not clear at what valuation the funding was raised, though reports have pegged it at somewhere between $40 and $50 billion, a figure that, at least economically, is starting to shape xAI into a real competitor to OpenAI, which is valued at $157 billion.

The landscape: xAI in May closed a $6 billion funding round at a $24 billion valuation, meaning the firm has secured some $12 billion in venture funding this year alone. According to Business Insider, Fidelity recently lifted its valuation of both X and xAI.

This rapid growth, according to the Southern Environmental Law Center, is contributing to steadily worsening air pollution in Memphis, the site of xAI’s massive data center, a data center powered by gas turbines which are emitting carbon dioxide and particulate matter alike.

The parents of OpenAI researcher-turned-whistleblower Suchir Balaji, who was found dead in his apartment in November, are demanding an FBI investigation into the death of their son, who, before his death, declared publicly that the act of training an AI model constituted copyright infringement. Shortly before he died, he was named a witness in the New York Times’ lawsuit against OpenAI. Balaji’s parents said that a private investigation found that their son’s apartment had been “ransacked,” and that a private autopsy “doesn’t confirm (the) cause of death stated by police,” which was suicide.
South Korea has passed its “Basic Act on the Development of Artificial Intelligence and the Establishment of Trust” Act, a piece of legislation that, similar to the EU’s AI Act, establishes a comprehensive, risk-based regulatory regime for AI.

Meta envisages social media filled with AI-generated users (FT).
AI brings better odds and betting concerns to sports gambling (Semafor).
Microsoft works to add non-OpenAI models to Copilot products (Reuters).
Nvidia completes $700 million Run.AI acquisition (TechCrunch).
AI hallucinations are driving scientific breakthroughs (New York Times).
Silicon Valley’s turn of fortune: Intel has worst year ever, while Broadcom enjoys record gain (CNBC).

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

Source: OpenAI

In the week since OpenAI unveiled o3, the discourse around AGI (artificial general intelligence), already contentious, has heated up. But this discussion of future threats and reasoning benchmarks has been scientifically challenged from the start; no one has a clear definition of what, exactly, AGI is.

It’s hard to achieve something if you can’t even define what you’re trying to achieve.

Scratch that, it’s not hard. It’s next to impossible. At least, scientifically.

Still, one of the major points of the ongoing contract negotiations between OpenAI — which is attempting to transition to a for-profit organization — and Microsoft, OpenAI’s largest financial backer, involves AGI. The reason behind this is likely pretty simple; under the terms of their current agreement, OpenAI gets to withhold all technology deemed “AGI” by OpenAI’s own board, effectively severing the relationship once/if OpenAI achieves this breakthrough. OpenAI loosely defines AGI as something that exceeds most people at “most economically valuable work.”

The pair have been working to nail down a clearer definition.

What happened: According to The Information, which reviewed internal documents distributed by OpenAI to investors, AGI would only be achieved when OpenAI has developed a system that has the “capability” to generate the maximum total profits it owes its initial investors. This totals about $100 billion, with Microsoft alone entitled to $93 billion.

According to the report, the documents state that the declaration of AGI remains up to OpenAI’s board, adding the expected dosage of vagueness and flexibility to a definition that, even in its clarity, remains hard to pin down.
It’s not clear if this $100 billion system refers to a single model, or a system of interconnected models, or how Microsoft plans to assess whether a model is capable of generating that amount of profit. It is also unclear whether any sort of scientific constraints or benchmarks will be coupled with the economics to determine the AGI of it all, or how OpenAI’s conversion to a for-profit organization might impact that definition.

Speaking of … At around the same time, OpenAI publicly detailed the reasoning behind its intention to transition from its current hybrid structure — a strange blend of nonprofit and capped for-profit — to a more traditional for-profit model.

In the post, OpenAI explained that the impetus behind its evolution from a strict nonprofit to that odd, hybrid combination, was simple: they needed “far more compute, and therefore far more capital, than we could obtain with donations in order to pursue our mission.”

The company said that it has become clear that the accomplishment of its mission to achieve AGI will require a scale of capital higher “than we’d imagined.”

As it stands today, the plan is to transition into a Deleware Public Benefit Corporation, with OpenAI’s AGI-centric mission as the stated public benefit. The non-profit arm of the company would remain, taking “shares in the PBC at a fair valuation determined by independent financial advisors.”
The corporation will control the business and its operations; the non-profit board will “pursue charitable initiatives.”

If OpenAI doesn’t complete this transition within the next two years, the investors who participated in the company’s recent $6.6 billion funding round can ask for their money back, plus interest.

Beyond handling all the logistics behind such a transition, OpenAI will have to deal with legal challenges, the first of which have already appeared.

Elon Musk, an initial founder of the company, has filed an injunction against the company to block its conversion to a for-profit organization. And Encode, a youth-led organization focused on safeguarding AI, has backed Musk’s injunction in an amicus brief that argues that such a transition “would fundamentally undermine OpenAI’s commitment to prioritize public safety in developing advanced artificial intelligence systems.”

A hearing on the initial injunction is set to occur on Jan. 14, 2025.
“The public has a profound interest in ensuring that transformative artificial intelligence is controlled by an organization that is legally bound to prioritize safety over profits,” Nathan Calvin, Encode’s vice president of state affairs and general counsel, wrote in a statement.

“OpenAI was founded as a non-profit in order to protect that commitment, and the public interest requires they keep their word,” he added.

I would add only that this serves as a pretty open admission of the sheer cost associated with the construction and operation of these models. And though they do not mention it, that cost comes in far more than just dollars.

Still, it is unsurprising and rather fitting that the true, internal definition of AGI centers around simple economics — all these seemingly noble scientific ventures to better humanity are little more than a wrapper around the true, corporate vision of AI: to make a lot of money, roughly $100 billion, for starters.

So when we think about actual progress and scientific advancements, when we think about ethics and morality and the multi-layered impact that ripples out from AI deployments, let’s remember the simple source of skepticism; that these companies exist, not to better the world, but to sell a product.

We should have a high bar for what, or who, we’re willing to sell our trust to.