AI: Happy New AI Building 2025! RTZ #587
...accelerating progress that will require patience and out of the box execution
Thanks for opening this on New Year’s Day, and again, the best wishes to you and yours for the 365 days ahead.
As I underlined yesterday, the LLM AI Tech Wave, now over two years old since OpenAI’s ChatGPT launch in late 2022, will see accelerated deployment at Scale in the US and abroad. Both the underlying AI hardware and software will scale at unprecedented rates, with new AI applications and services, far beyond ‘chatbots’ will see mainstream traction with both businesses and consumers.
AI Reasoning and Agents, levels 2 and 3 above, will take most of this year to move out of the AI labs and research projects, to commercial products that become more familiar beyond Developers and early adopters. We will see accelerated progress up and down the AI Tech Stack below.
Yesterday I highlighted six trends continuing from 2024, that will also be more relevant than ever this year. Today I’d like to highlight the 7 predictions I made last year, that will also see continuation and acceleration in 2025.
“Picks and Shovels still rule”: This was of course a major trend last year, and just accelerates this year. Power is an element that is even more important this year in Box 1 above, given the gigawatts of energy required per AI data centers with hundreds of thousands of AI GPUs (mostly from Nvidia), that need to be powered for training, inference of next gen LLM AIs. And now we add big amounts of AI reasoning, agents, synthetic content and data, multimodal video and audio generation computations that will very much dwarf the Compute expectations from last year. Nvidia of course remains in pole position, with its top customers and beyond.
“Data, Data Everywhere”: This one is shifting into second gear as move from the Data that have fueled the LLM AIs to date with ‘everything’ from the internet. As Ilya Sutskever of OpenAI until last year, and now founder of Safe Superintelligence (SSN) pithily says, we are at ‘Peak’ Data of the old form. And now onto new data for these models harvested from videos and sensory inputs from the physical world. That will need new, massive investments that will power on the Scaling of the AI models in 2025 and beyond. Much of it will likely be ‘Synthetic Data’ as I’ve discussed, along with AI creating ‘Synthetic Content’ as well, for user consumption AND training/inerence in new AIs going forward.
“AI Beyond Search”: With all the major players aggressively deploying AI to revamp Search as we’ve known it, expect lot more mainstream progress how how AI Search is both delivered and monetized. Besides Google, expect OpenAI in particular to be focused here. Along with AI startups like Perplexity, Glean and others that are focused on horizontal and vertical AI Search applications.
“Smart AI Agents”: Just getting started here going into 2025. Last year was about laying the technical foundations to do this at scale. Daresay it’ll be at least until next year that we see truly useful, reliable AI agents at scale in consumer and business applications. But that doesn’t mean a whole lot of effort won’t be expended by incumbents and startups. A work in progress in 2025.
“Reinforcement Loops Accelerate at Scale”: This was true last year, and go into overdrive this year. Especially with new AI reasoning techniques like ‘Chain of Thought’, and new ‘Distillation’ techniques that allow smaller models to get far better off bigger models. Vibrant area for both research and commrcial applications. Again, a massively important area for data and copyrights compute efficiencies of all types this year.
“Regulatory Stability in AI”: Discussed this yesterday at the macro level. The details matter here at the state and local levels in the US and other jurisdictions. A LOT of relatively unpredictable elements to watch here, that are likely to slow down the AI technologies that are coming out of the labs. Area to watch closely.
“Working with China”: Again discussed yesterday at a macro level. But bears highlighting here given the US ambitions to apply AI to the physical world in enterprises, automobiles, robotics, and more. China in particular has ecosystem advantages at scale, which are not likely addressed by a few reshored fabs and factories within US borders. Will need a lot of innovative strategies to cooperate with China while competing. Again, an unknown variable to focus on.
So again, US/China is a wild card for AI and tech in particular. And a lot of moving parts to watch closely again this year as this AI Tech Wave progresses. This year should still be viewed as a build year, for every box in the AI tech stack above.
While we may see new breakout AI applications, services and possibly devices, it’s still very early days. That will require a lot of capital and patience.
Happy New Year to all again. Stay tuned.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)