The AI Model Genie is Out of the Bottle

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

TL;DR Modelpalooza; Atlas is dead, Long Live Atlas!

This week we’re going even tighter on the Geekly as we hammer-away at our introductory AI 101 Primer piece, but fear not as we have some great content for you. While it was the Coachella music festival in Indio California this week (Sublime and No Doubt killed it!), you would be forgiven for thinking that it was the AI Model Release festival this week as well. Heavyweights like Meta, MistralAI, StablilityAI, Cohere, Zyphra, and more each released state-of-the-art models this week that push the envelope further across generative disciplines (text, image, sound), while other advances were made with self-driving car models that use natural language to drive (Wayve’s LINGO2). Finally, you may have noticed we’ve been including a lot of content on robotics over the past several months —this is because of our convergence thesis which sees AI, robotics and spatial computing beginning to converge over the course of the year. Perhaps the best-known robots are Boston Dynamics’ Spot (dog-like robot) and Atlas (humanoid). While this week Hyundai-owned Boston Dynamics retired its aging and comparatively unintelligent hydraulic-powered Atlas robot, it teased a new automaton that might give the likes of Figure01 a run for their money (or do robots use crypto?).

Llamas, Mistrals, and Zambas, Oh My!
Pace of AI model improvement continues to ramp

What it is: This week saw a deluge of models, with the arrival of Meta’s hotly anticipated Llama 3 model (Llama 2 is one of the most popular open-source models) followed by another new open-source model from France’s MistralAI team, Mixtral 8x22B. Finally, there was a new take on the Mamba model series with Zyphra’s Zamba (aren’t model naming conventions fun?), which uses a different model architecture than the Transformer architecture underlying ChatGPT and the models mentioned above.

What it means: In line with our expectations from the beginning of 2024, open-source models like the three mentioned above (Llama 3 specifically) are beginning to approach the capabilities of rival Open AI’s closed GPT-4 model, the current best-in-class —though Anthropic’s Claude 3 Opus claims to have beaten GPT-4 on several benchmarks, GPT-4 is still the name to beat.

Why it matters - Enterprise: From a capital markets perspective, this widening playing field is critical, as the macro value proposition for GenAI remains an open question. That is, we’ve seen pockets where GenAI has driven better business outcomes in select use cases for select companies, but it has not yet been any kind of broad panacea and with billions of dollars being pumped-into the space, investors need to see companies generate significant ROI in order to justify the eye-popping figures committed to date (see our upcoming AI 101 for a summary of AI equity raises to-date). Performant open models like these offer a viable alternative to the expensive closed models offered by the major players, permitting cheaper fine-tuning and inference which is a major factor in determining the viability of these tools in production. For highly regulated industries like Healthcare, Finance, and Government, on-prem solutions are a must, and open-source models present the simplest and cheapest solution in that regard.

Why it matters - Society: Open models are our “insurance policy” against the oft-unchecked actions of tech oligopolies and the negative partiality of those who govern in their own interests. As we’ve stated previously, democratization of AI is the only way to ensure equitable distribution of the benefits of this game changing technology. It defies reason that one would trust the very parties responsible for the widening wealth gap, the erosion of societal values and civics, and successive declines in generational prosperity to mete-out the positive impacts fairly. Ergo, the gap between closed models available to a select few, and open models, available to all, must be kept as tight as possible. With Open models approaching GPT-4 levels of performance the necessary check is now in place for the time being. But the effort to chase parity is ongoing.

“Give me your clothes, your boots, and your motorcycle”
A new challenger appears

What it is: Hyundai-owned Boston Dynamics has retired its famous Atlas humanoid robot and has introduced a new more capable robot named… Ok, this one is called Atlas also, but it’s electric instead of hydraulic powered, meaning it’s got a few new tricks up its sleeve (check-out the video link below).

What it means: We were wondering when trailblazer Boston Dynamics was going to update us on what they’ve been up to. With Figure01, Tesla’s Optimus, and a handful of other robots getting press with capabilities far surpassing the original Atlas robot we’re glad to see that BD hasn’t been asleep at the wheel.

Why it matters: Look, robots creep people out, I get it. But do you know what bothers lot of people way more than robots? Manual labor. Dangerous work conditions. Caring for the elderly. Topsoil erosion (ok that last one’s serious but robots won’t help there…). There are millions of use cases for a functional robot imbued with the intelligence of modern AI. The humanoid design allows them to seamlessly plug into a world that was purpose-designed for human forms. Just as we must adjust our comfort levels to save millions of lives by accepting that self-driving cars are safer than human drivers (despite fearmongering clickbait headlines); we need to come to grips with the reality that intelligent robotics carries with it the potential to dramatically improve the quality of life of current and future generations. Now we just need jetpacks and flying cars. But that’s a different newsletter.

About the Author: Brodie Woods

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.