SAIL: Systems, but still about those agents
Welcome to sensemaking, AI, and Learning (SAIL). I focus mainly on AI advancement that might impact higher education.
While there is growing interest in how AI impacts the existing learning process (links below) there hasn’t to date been a focused “clean slate AI” education initiative (ok, other than this initiative with Paul LeBlanc) that begins with the premise of what is now possible, rather than service to a legacy system. I was on a panel last year with an individual from one of the main LLM providers. I stated that universities need to start evaluating their data architecture and making some foundational technical investments into a modern computing infrastructure. The LLM representative stated that this approach was overkill and simply giving students access to AI tools would be sufficient - don’t complicate it.
This critical difference is between a point solutions vs systems solutions (see here). Point solutions solve an existing problem in an existing systems, but largely fail to account for affordances enabled by new technologies. Every university will adopt point solutions. Only a few with forward thinking and visionary leadership will achieve the systems change. Today I haven’t seen a university leader meet the systems challenge.
AI and Education
Change in USA post secondary education. See Figure 4. Massive change in non-degree credentials.
Academics fear students are becoming too reliant on AI. The impact of using AI in learning, however, is significant: “We find that students learn more than twice as much in less time when using an AI tutor, compared with the active learning class. They also feel more engaged and more motivated.”
How much do AI agents help with tasks like coding? Hamel Hasain (disclaimer: we’re collaborating on our education work with Matter & Space) leads an excellent overview with some devastating, but affirming for humans, conclusions: “This reflects a pattern we’ve observed repeatedly in AI tooling. Social media excitement and company valuations have minimal relationship to real-world utility. We’ve found the most reliable signal comes from detailed stories of users shipping products and services. For now, we’re sticking with tools that let us drive the development process while providing AI assistance along the way.”
Five recent AI and Tutoring studies (not all higher ed). On study though dropped the rather astonishing statement of learners achieving “the equivalent of two years of typical learning in just six weeks.”
WEF offers a common skills taxonomy, with the goal of “aligning the efforts of businesses governments and learning providers in addressing critical skills and talent shortages.”
Granola is my current favorite (new) AI tool. Excellent cross platform meeting summaries.
Universities need to start building AI products. That’s an internal technical and procedural capability, but one that also requires leadership. Here’s an overview of lessons learned building AI products by the CEO of Granola.
LLMs visualized. Fantastic. Take some time here.
AI Advancement
The biggest news of the past week, possibly year, has been DeepSeek - a Chinese developed LLM that is performing at o1 levels for reasoning. It’s licensed under an MIT license, enabling users to “Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software”. This is a big deal. It’s the first advancement that I’ve seen distressing AI insiders who see a sudden collapsing of the proverbial moat. It’s good news for anyone on the using, not developing foundation models, end of LLMs. You can run DeepSeek R1 (their reasoning model, comparable to OpenAI’s o1) locally, offline, on your mobile (the 1.5b parameter model). DeepSeek is significantly cheaper than other models.
Some people aren’t afraid of AI taking jobs, assuming that regulation will drive up cost. The DeepSeek and Llama open models suggest it will be hard to fully regulate.
OpenAI launched its first agent - Operator - this week. It’s included in the $200/mo plan. I’ve been playing with it and it offers glimpses of what’s possible, but is still too experimental to be broadly useful or even reliable. OpenAI acknowledges this and states the model will improve with user feedback. Some simple tasks like “find a vegetarian recipe, make a grocery list, and order for delivery” work quite well even now.
The new administration (USA) is one of the more aggressive globally to advance AI, including rescinding previously set guardrails. It’s an “all gas no brakes” mindset.
USA makes $500 billion infrastructure announcement. The numbers seem kinda ridiculous and there is some speculation that at least some of it is hype.
Since AI is all about the hundreds of billions Abu Dhabi will throw $300b into the ring as well “The world has yet to fully recognize the extent of change artificial intelligence will bring to every aspect of human life”.
AI and war is the next big economic frontier. ScaleAI calls for more focus on this area. Andruil (a must watch company in this space) with OpenAI is in the game.
This is the year of agents (though nascent), but next year will be all about robots “The big breakthrough right now is the evolution of humanoid robots that essentially follow individual workers on the factory floor, on a construction site, even a chef in a restaurant or a housekeeper. It's terrifying, but it's happening in the next literally year or two”
AI and Anxiety
AI exhibits self-preservation It seems naive (arrogant) to assume we can control AI when it exceeds our intelligence. We really don’t have a clue: “Anybody who says they know is just overconfident because really, science doesn’t have the answer. There is no way to answer in a verifiable way.”
Something has shifted. “Researchers began speaking urgently about the arrival of supersmart AI systems, a flood of intelligence. Not in some distant future, but imminently.”
AI and job loss, despite comments above from Andreesen, seem to be happening. At least at Salesforce and Replit and Meta are minimizing employment. Goldman Sachs is also using AI to reduce employment.
OpenAI offers an economic blueprint (Seagrams offers new sobriety program!) “Chips, data, energy and talent are the keys to winning on AI—and this is a race America can and must win.”