Welcome to Stuff I Thought About Last Week, a personal collection of topics on tech, innovation, science, the digital economic transition, the finance industry, and whatever else made me think last week.

Click HERE to SIGN UP for SITALWeek’s Sunday Email.

In today’s post: KISS goes virtual; Google's Gemini hints at a plateau in LLMs, but moving to a new level could happen with the integration of AI and robotics; Mark Cuban's is exerting pressure on big pharmacy; melatonin use soars; Juror #6 gets a production deal; military goes high tech. And, much more below...

Stuff about Innovation and Technology
Cloning KISS
The legendary rock band KISS played their last concert and transcended into the virtual world – to live on as full-body motion capture 3D virtual characters. The new virtual KISS was created by Disney’s Industrial Light & Magic. The band sat down for a video conversation with ILM execs to discuss their history and decision to immortalize themselves. The band has not indicated what the virtual group will do in the future, and it’s not clear to me who owns the band’s likeness given the mocap was funded by Sweden-based Pophouse Entertainment. While the technology used by ILM to digitally clone the band is run-of-the-mill superhero movie stuff, the idea that the band is retiring from public concerts and handing the reins to a licensed, virtual version of themselves, while still alive, could set a precedent that other performers may follow. We are already drowning in infinite content, and the proliferation of virtual versions of real performers could cause the flood waters to rise even more. 
 
Gemini’s Plateau
Google finally unveiled their new multimodal AI called Gemini. A regular “Pro” version has already been integrated into Bard, and the “Ultra” version, dubbed Bard Advanced, will be available next year. In my experience, the new Bard is a huge upgrade from prior versions and is remarkably faster than ChatGPT. I would rate its capability as somewhere between GPT3.5 and GPT4, but more like a faster version of 3.5. However, the real magic appears to be in the Ultra model as seen in this video (the video is highly modified for effect, but I think reading through the process makes the AI seem even more impressive). The stats behind Gemini Ultra indicate it surpassed prior models on 30 out of 32 top AI benchmarks, and it also scored 90% on the Massive Multitask Language Understanding test, besting human experts across a variety of language, science, and social topics, such as ethics. The model also outperformed 85% of software programmers in coding competitions (although competitions differ from actual coding work). Gemini is a combination of an LLM and DeepMind’s tech that powers AlphaGo and AlphaFold. “Multimodal” means it works with text, images, video, sound, code, etc. I would guess that Gemini Ultra slightly exceeds the capabilities of the current GPT4 Turbo, but its multimodal processing appears stronger. 
 
While the base-case expectation is for LLMs to become superhuman, it seems like it might be harder to progress much beyond regular human skills, simply because today’s LLMs are so human-like. Afterall, the best of us humans still make countless mistakes, so we wouldn’t want to rely on a sole LLM for anything necessitating a high degree of accuracy. Several companies have been able to achieve GPT3.5-level functionality, but only two have had the ability to get to the GPT4 Turbo/Gemini Ultra level of performance. It’s plausible models will plateau at this point without some additional breakthrough, leaving us with advanced – but not superhuman – AI. And, I doubt we would want to rely on this level of AI to perform critical tasks, e.g., even OpenAI admitted last week they aren’t sure why GPT4 has started acting “lazy”, but are looking into it. 
 
Perhaps an important step forward will be going beyond sight and sound by integrating various sensors and data feeds to create an intelligence with super-sensing capabilities (e.g., LIDAR, weather forecasts, vibration, etc.). Gemini has an on-device version called Nano that will soon launch on Pixel 8 phones. On my Pixel 8, I’ve been noticing a lot of helpful copilot behavior already, such as real-time summaries of web pages and the ability to proofread and edit what I type with the click of a button in the Google keyboard. As I often note, the winner in any new platform shift is the one that wins developers and users, with developers who are creating apps for users being the key target. Microsoft OpenAI has an early lead, but Gemini Ultra could pull ahead, especially based on its apparent speed and advanced multimodal processing. This is a real-time arms race unlike anything we’ve seen before (e.g., Android came years after iOS, and Apple has still powerlawed app-economy profits). In a brief interview with the Alphabet and DeepMind CEOs, there seems to be an implication that the next version of Gemini (currently being trained) will be a significant leap from what we’re seeing in the demo video. That speed of progress is hard to wrap your head around – if AI does not soon plateau, then this is the most nonlinear innovation cycle I can recall in the history of the tech sector. 
 
Seeing Gemini has me thinking about my post on embodying AI in the physical world from March:
 
Much of being human involves processing myriad inputs from our expanded seven senses (sight, hearing, smell, taste, touch, thoughts, and emotions, all of which feed into creating our ongoing sense of self). But, ChatGPT runs on a server in some Microsoft data center. It’s effectively got one sense – the interaction between the model it was trained on and the input from a human interlocutor. We could describe Bing-Chat as having two senses, with the second being its ability to access the Internet in real time. We could theoretically feed more sensory data into ChatGPT, in particular real-time images/video, sounds, or even things that approximate “touch” like temperature and pressure data. Or, we could embody the LLM in some type of physical form that allows it to more dynamically interact with, and receive input from, the real world. It’s an open question as to whether LLMs would learn to process and respond to data in a human-mimetic way or if more alien behavior would emerge.
 
Neuroscientist Antonio Damasio characterizes human existence in terms of a drive toward homeostasis. He sees this quest for comfort as the fundamental life force (detailed in his book The Strange Order of Things). Essentially, the nervous system is a connected tool to make the living organism (e.g., a human) feel in balance. If we need calories (or detect that free calories are available), we feel hungry and eat. If we are cold, we seek shelter. Damasio further believes feelings emanate from the drive toward homeostasis and are a way for the brain to interpret good or bad states and act on them. He speculates we can derive all of human consciousness and culture from our thoughts, feelings, and actions regarding our relationship with homeostasis. I tend to agree with him, but I also hold these beliefs loosely given that we don’t know all the answers. Thus, homeostasis seems intertwined with feelings, consciousness, and a physical body able to monitor (and react to) our internal state and the world around us. Therefore, the critical junction of combining a LLM with a physical form capable of monitoring its systemic and real-world inputs (e.g., temperature, pressure, proprioception, energy reserves) – and react to these sensory data in a way that seeks to maintain its own, human-equivalent homeostatic targets – seems like the logical next step in the trajectory of AI. And, we may already be rapidly progressing down that road. Microsoft’s Autonomous Systems and Robotics division announced their intentions to put OpenAI technology into robots (blog post and video). This integration could ultimately lead to a paradigm shift in AI where we go from programming a specialized robot/tool to do a specific task with specific inputs, such as autonomous driving, to a situation where you could just ask ChatGPT to go learn how to drive a car. However, merging AI with a physical form demands an exceedingly careful approach for successfully progressing such tools in the real world, particularly with respect to protecting human safety (a couple weeks ago, I shuddered at the idea of integrating ChatGPT into the new Boston Dynamics humanoid!). Rather than a thoughtful approach, we unfortunately appear to be heading toward what I described last week as “my AI can beat up your AI”, with tech leaders now attempting to create rival AIs backed by distinct ideologies. Bill Gates is apparently heavily involved in leading the strategy as an advisor to what he refers to as “Microsoft OpenAI” in this FT podcast interview (the transcript is here, but it has several typos). Gates largely dismisses concerns over AI in the podcast. Given Microsoft’s desire to embody LLMs in real-world robots, this blasé stance is concerning to me. Gates went so far as to question whether we should blame people, rather than the AI itself, for its shortcomings. While this mentality also concerns me, the underlying point – that the risk with AI is more weighted to how people will use it than the AI itself – is valid. Gates described the pending GPT4 as: “wow”, with capabilities coming “many years before I expected”.
 
The next-level scenario of embodied, human-mimetic chatbots brings emergence to mind. Emergent behavior is something new that occurs in a complex system of interacting agents that wouldn’t have happened (or been predicted) based on the agents’ isolated actions. Certainly, chatbots are an emergent phenomenon from LLMs, but I am not sure today’s chatbots themselves demonstrate emergence (although I will note that the Microsoft researcher in that video linked above declared that they do, but I do not know what definition he was using). People are certainly finding emergent use cases for chatbots. When LLMs become embodied in the real world, however, we should expect to see emergent phenomena from the robots themselves. For this reason in particular, the scarcity of caution today is concerning. 
 
CVS’s Cost Concessions
The success of Mark Cuban’s transparent pharmacy company is putting broader pressure on the prescription-peddling industry. Drugstore giant CVS announced last week they are adopting a new prescription pricing strategy more aligned with that of the Mark Cuban Cost Plus Drug Company. This move comes after one of CVS’s customers, Blue Shield of California, migrated to Cost Plus. CVS is also likely responding to FTC scrutiny of its pharmacy benefit management (PBM) business, which, as far as I can tell, is designed to obfuscate pricing info and cause drugs to be more expensive than they need to be. We first covered the Mark Cuban Cost Plus Drug Company nearly two years ago, and it’s great to see the positive progress and impact it’s having. The company intentionally set out to work around the clouded and misaligned PBMs and insurance company agendas. Even with the reactive pricing changes, it seems unlikely that a business like CVS, with all of its conflicts, can best a highly non-zero-sum-focused business like Cuban’s. As Business Insider writes: “It's not clear if CVS's new program will result in lower prices overall for consumers. Some drugs will cost more under CostVantage, while others will cost less, according to the Wall Street Journal. The costs could also depend on the terms of an individual's insurance plan, and whether they choose to use it.” As for Cuban, he is reportedly selling a stake in the Dallas Mavericks (although will continue to run the organization) and announced he will be exiting the hit TV show Shark Tank. I am looking forward to seeing what he does next.

Miscellaneous Stuff
Epidemic Insomnia
According to research published in JAMA Pediatrics, 19% of US children aged 1 to 14 are relying on melatonin to aid in sleep, up from 1.3% 5-6 years ago. The nearly 15x increase seemed to ramp during the pandemic. One suggestion is increased screen time (which can disrupt the body’s natural melatonin cycle) is to blame, or perhaps it’s a general increase in anxiety (or maybe it’s TikTok, which seems able to increase both screen time and anxiety). Melatonin is available over the counter in the US, but other countries, like the UK, require a prescription. A total of 27% of US adults are now said to take melatonin.
 
Gladden’s Good Karma
Ronald Gladden, the earnest and unwitting star of the reality/mockumentary TV series Jury Duty, has signed a two-year development deal with Amazon MGM for new projects. From #396:
Jury Duty is a new bingeable show on Amazon’s Freevee channel (I watched it via Prime Video). The show follows a fake jury trial where only one member of the jury, Ronald Gladden, isn’t in on the act. Gladden, a 29-year-old solar installer, spent weeks in fake sequestration under the assumption he was part of a documentary on the judicial process. It’s a bit like a real-world version of The Truman Show, and it has many delightful moments throughout. I of course loved it because it plays with society’s increasingly blurred lines between fact and fiction (see Cinéma Vérité). In yet another reality-defying twist, Amazon plans to submit Ronald Gladden for a best-actor-in-a-comedy-series Emmy, despite him not even knowing he was in a TV show during filming. (Note: Jury Duty received four Emmy nominations, but not for best actor).

Stuff About Demographics, the Economy, and Investing
AI Called to Duty
Anduril’s new Roadrunner and Roadrunner-M jet-powered autonomous drones are designed to destroy other jet-powered targets, like enemy drones and missiles, mid-flight. The reusable vehicles (which launch like a rocket and then return home after completing their mission) are one potential advance beyond the US’ analog-focused and outdated military efforts. Development of this type of intelligent drone aligns with what I wrote a couple months ago: 
Autonomous drones imbued with AI are becoming quite common, as we’ve discussed in the past. The WSJ reports that the Pentagon wants to build a fleet of thousands of AI robots for air, land, and sea deployment. The move is said to counter China, which is far ahead of the US with these capabilities. This is starting to feel like the new “mutually assured destruction”, i.e., every country will have a massive fleet of AI-military tech, and the consequences of anyone deploying it would be met with just as big of a threat.
The main difference of course is that much of the autonomous and AI-driven technology is low cost and available to anyone with minimal resources. That is a very different situation than a handful of nations controlling the supply chains and production for advanced military weapons and nuclear bombs.

✌️-Brad

Disclaimers:

The content of this newsletter is my personal opinion as of the date published and is subject to change without notice and may not reflect the opinion of NZS Capital, LLC.  This newsletter is an informal gathering of topics I’ve recently read and thought about. I will sometimes state things in the newsletter that contradict my own views in order to provoke debate. Often I try to make jokes, and they aren’t very funny – sorry. 

I may include links to third-party websites as a convenience, and the inclusion of such links does not imply any endorsement, approval, investigation, verification or monitoring by NZS Capital, LLC. If you choose to visit the linked sites, you do so at your own risk, and you will be subject to such sites' terms of use and privacy policies, over which NZS Capital, LLC has no control. In no event will NZS Capital, LLC be responsible for any information or content within the linked sites or your use of the linked sites.

Nothing in this newsletter should be construed as investment advice. The information contained herein is only as current as of the date indicated and may be superseded by subsequent market events or for other reasons. There is no guarantee that the information supplied is accurate, complete, or timely. Past performance is not a guarantee of future results. 

Investing involves risk, including the possible loss of principal and fluctuation of value. Nothing contained in this newsletter is an offer to sell or solicit any investment services or securities. Initial Public Offerings (IPOs) are highly speculative investments and may be subject to lower liquidity and greater volatility. Special risks associated with IPOs include limited operating history, unseasoned trading, high turnover and non-repeatable performance.