Here Comes a New Challenger

Welcome back to the AI Geekly, by Brodie Woods, brought to you by usurper.ai. This week we bring you yet another week of fast-paced AI developments packaged neatly in a 5 minute(ish) read.

TL;DR OpenAI’s building a mystery; carpal tunnel signing all these publishing deals; Going after the king

This week we have another packed note for you as we walk through OpenAI’s recent, and quite telling, blog post regarding the formation of a new safety team and hints that it has begun training its next model (GPT-5 when??). Sticking to the OpenAI theme, we’ll take a look at two recent content licensing deals the company just signed with a couple of major publishers (we’re making you scroll for the names!) and how they impact the legal overhang facing models trained on unlicensed content (read: basically all of them). Finally, we’ll take you through the formation of the Ultra Accelerator Link Promoter Group and their efforts to dethrone the King of Compute, Nvidia —Can it be done? Read-on below!

New Shiny Things from OpenAI
Safety committee assembled ahead of GPT-5 launch

What it is: In the wake of the (much ballyhooed) disbandment of the company’s Superalignment team (responsible for ensuring that AI’s interest align with human interests over the long term) OpenAI announced the formation of a new Safety and Security Committee, led by CEO Sam Altman, Chairman Bret Taylor and two other board members, providing strategic guidance on critical safety and security decisions for the company’s projects and operations. Concurrently, OpenAI has begun training a new AI model intended to succeed GPT-4 (let’s call it… oh… I don’t know, GPT-5?), aiming to advance towards artificial general intelligence (AGI). Recall that AGI, per CEO Sam Altman, is an AI with the competence and capability of an average coworker (hopefully not a lazy one).

What it means: The establishment of the committee follows internal criticisms and high-profile resignations (which we previously covered) highlighting concerns that OpenAI has been prioritizing product development over safety. The committee’s first task is to evaluate and enhance OpenAI’s safety processes and safeguards within 90 days, after which recommendations will be presented to the board and publicly shared. GPT-5 is expected to bring significant advancements relative to the prior iteration, underscoring the importance of these safety measures as the company pushes the boundaries of AI capabilities.

Why it matters: We note that the focus of this team appears to be more near-to-mid-term in nature, and does not, in fact, constitute a replacement of the Superalignment team, which has a longer-term, and in our view more critical role in ensuring the alignment of eventual AGI to human priorities. As we know, OAI already has robust internal controls related to safety built-in to each team and with each stage of development. This has led to models that have been severely restricted in the name of safety, relative to uncensored open-source models. We worry that a hyper-focus on short-term safety may slow or stall innovation and could ultimately hinder superalignment efforts if a new team is not established.

Let’s Make another Deal
Vox and the Atlantic join growing list of OpenAI licensors

What it is: On the heels of several recent content licensing deals with News Corp., Financial Times, Axel Springer, The Associated Press, and more, OpenAI has signed content licensing agreements with Vox Media and The Atlantic, allowing the AI company to use their archived and current content to train its models and enhance its products like ChatGPT. These multi-year deals also include provisions for product development collaborations, where both media companies will work with OAI to create new AI-driven features and tools. Financial terms of the deal were not disclosed.

What it means: These partnerships provide OpenAI with another rich repository of high-quality content for refining its AI models and ensuring they remain at the cutting edge. For Vox and The Atlantic, the deals offer opportunities to leverage OpenAI's technology to enhance their own products in an era where publishers are again facing the threat of technological disruption —better to be the disruptor than the disrupted! An additional benefit is that these agreements help mitigate potential copyright infringement issues by formalizing the use of the publishers' content. We honestly doubt that OpenAI has refrained from training on content from either publisher, so in some ways these deals are meant to tidy-up that little practical reality.

Why it matters: The deals are part of a growing trend of media companies opting for partnerships with AI firms to ensure some form of compensation for the use of their content in training these powerful AI models. The ability to use the models themselves in novel ways is icing on the cake but could better position these publishers from a competitive perspective relative to those who refuse deals (i.e. The NY Times, currently engaged in a lawsuit with OpenAI). It seems that more and more publishing companies are capitulating and signing deals with OAI and its competitors. While dollar figures of most of these deals have not been disclosed, the numbers are obviously circulating within a tight nit group of tech and publishing execs. Writers’ unions have expressed alarm at the seeming lack of compensation for their members and their contribution as the ultimate creators of this content.

“Headline” risk reduction: We remain in a price discovery phase in the development of modern AI and content licensing. As more publishers sign on, prices firm-up. Importantly, we also see a slow and steady trend of de-risking the use of LLMs built by large players —we have cautioned clients about the legal risks associated with using LLMs for which content has/had not been appropriately licensed. As a greater proportion of training data underlying the models becomes contracted, the legal risk associated with the use of unlicensed content diminishes incrementally for users of the models. That said, given the models have been trained on essentially the entire public internet, it may be difficult to neatly lock-down every piece of content, thus the risk, even if diminished, remains.

With Our Powers Combined…
Big Tech forms alliance to disrupt Nvidia’s chip dominance

What it is: Several major tech companies, including Intel, Google, AMD, Microsoft, Meta, HP, Cisco and Broadcom, have formed the Ultra Accelerator Link Promoter Group (yes, a mouthful) to develop a new open connectivity standard aimed at eliminating proprietary barriers in AI development —we’re looking at YOU Nvidia! You may have noticed that Nvidia, king of the GPU, is missing from the list above, this is by design: This initiative specifically seeks to prevent companies from being locked into using Nvidia’s NVLink interconnect tech as the sole available option in datacenters using Nvidia hardware. Currently, NVLink dominates as the de-facto industry standard for connecting multiple AI Accelerators or servers.

What it means: The Promoter Group’s project aims to create a more flexible AI development environment by supporting a variety of hardware and software configurations. By doing so, it challenges Nvidia's current dominance in the AI chip market, which has been bolstered by its superior GPU offerings. A proliferation of competitive open protocol AI hardware (chips, and connective infrastructure) could both serve to put downward pressure on Nvidia’s highly profitable GPUs, while also offering an alternative in a market that is essentially a monopoly.

Why it matters: While this collaborative effort could potentially disrupt Nvidia's market position, which currently accounts for approximately 80% of the AI-chip market, we’re not holding our breath. Don’t get us wrong. We admire the valiant effort, but Jensen and his band of merry monopolists have seen this movie before. Nvidia has long played the vendor lock-in game before NVLink. The company’s CUDA software platform is but one example. Another is NVDA’s G-Sync monitor technology, which has long dominated the rival Open Sync platform that was created for precisely the same reason —to challenge Nvidia’s dominance (in a different segment).

Tell me how you really feel: We are reminded of the timeless words of the Legendary Linus Torvalds, creator of the (semi-eponymous) Linux operating system, regarding Nvidia’s refusal to open source drivers and work with the Linux community when he famously quipped in 2012 “Nvidia F*** YOU!” after decrying the company as “the single worst company we have ever dealt with”. Having personally dealt with some of these Nvidia driver challenges in Linux for performing AI workloads, we add our voice to the choir with a hardy “Hear! Hear!”. We’ve all had enough of the anti-competitive practices by now.

About the Author: Brodie Woods

As CEO of usurper.ai and with over 18 years of capital markets experience as a publishing equities analyst, an investment banker, a CTO, and an AI Strategist leading North American banks and boutiques, I bring a unique perspective to the AI Geekly. This viewpoint is informed by participation in two decades of capital market cycles from the front lines; publication of in-depth research for institutional audiences based on proprietary financial models; execution of hundreds of M&A and financing transactions; leadership roles in planning, implementing, and maintaining of the tech stack for a broker dealer; and, most recently, heading the AI strategy for the Capital Markets division of the eighth-largest commercial bank in North America.