Updates fall 2025

Three recent articles that I found interesting: Cosmos Institute on AI deference, Mustafa Suleyman on seemingly conscious AI, and Nautilus on when AIs talk to each other.

Updates spring 2025

AI 2027 is a fascinating read with projections on where AI is going. Consider as you read this, all of these changes happen against the back drop of the existing political leaders. Mary Meeker’s trends in artificial intelligence slides are here (all 340 of them). Sam Altman post on AGI already being here and what the next five years looks like is also a good read. Consider this line from his essay: “As datacenter production gets automated, the cost of intelligence should eventually converge to near the cost of electricity.“ Lastly, Dario Amodei’s Machines of Loving Grace essay is a hopeful read on the potential of AI, and his essay Adolescence of Technology is much less so!

Buckle up.

Original fall 2024

To understand AI, the first thing step is just to start using the tools. Play with the main large language models (LLMs) like ChatGPT from OpenAI, Claude from Anthropic, Llama from Facebook, and Gemini from Google. Ask them lots of questions. Upload documents, images, and spreadsheets and ask about their contents (a recent study showed that AI is a better diagnostician than human doctors when entire patient case files are uploaded). Instead of going to Wikipedia, just ask an LLM and then compare the results. Use Perplexity instead of Google to search. Ask them weird stuff—like to create an image of a textbook on top of a teacup. Ask them to reason using chain-of-thought prompting (i.e., first do this, then do that). Ask them to respond as different people—answer at an 8th-grade level, answer at a graduate-school level, answer as if you were a doctor. Ask them to be sad or happy, to be creative. Play with the image and video models like Midjourney, Dall-e, and Sora.

All of the LLMs can code, so ask them to create simple web pages or programs for you. If you are more technically inclined, experiment with Cursor or start digging into the open-source models at Hugging Face.

Use AI to help with your daily work tasks: play with Fathom, Granola, or Otter to take notes on Zoom calls, for example. I’ve found the AI integration with Notion to be incredibly helpful and easy to use. This post from Fred Wilson at USV gives a great example of how an investing firm is integrating AI tools into its work. 

For a big-picture overview, The Coming Wave by Mustafa Suleyman (founder of Deepmind and now at Microsoft) is a good read. Vinod Khosla (a venture capitalist) wrote a good overview of the impacts of AI in both long and short form. John Brockman’s book Possible Minds contains 25 essays from people in the field and is a great read. McKinsey put out an oft-cited report on the economic impact of AI. James Lovelock (of Gaia hypothesis fame) wrote a little book before he died called The Novacene that I found thoughtful and a little disturbing (plants are to humans as humans are to artificial general intelligence, or AGI). While not a book about AI per se, The Maniac about John Von Neumann includes a great telling of the AlphaGo story. Von Neumann was one of the first to think of and conceptualize artificial intelligence. Both AlphaGo and AlphaFold (a platform for protein folding prediction) are great examples of the power of AI and are worth understanding. Mind Design III by John Hagueland is an excellent collection of writings on AI. Lastly, Sam Altman from OpenAI writes here and is worth following. 

Regarding alignment and downsides, Noam Chomsky wrote about the “False Promise of ChatGPT.” Stuart Russell’s Human Compatible addresses the alignment problem (short answer: don’t tell them our goals—let them try to learn). Atlas of AI by Kate Crawford is the best work I’ve read about the problems in AI, specifically regarding bias. Emily Bender has written widely on the downsides of AI and in her paper on the topic coined the phrase “stochastic parrots” (hers was one of the original critical papers on LLMs). Henry Kissinger and Eric Schmidt wrote The Age of AI, which is largely focused on politics and nation-states. Geoffery Hinton, who worked on early neural networks (he won the Nobel in 2024 for his contribution), has been outspoken on the risks of AI, and it is well worth finding his comments in various articles.

Regarding AGI, Situational Awareness by Leopold Aschenbrenner is the best thing I’ve read and convinced me AGI is way closer than we think (although I still think there are some missing pieces for true AGI). It is worth reading his whole site, but it does get a little technical. Superintelligence by Nick Bostrom is one of the original works that sparked concerns (I’ve read parts but not the whole book). Yuval Noah Harari has written about what merging with a superintelligence might look like in Home Deus and has a new book, Nexus (which is in my queue). The paper “Sparks of General Artificial Intelligence” argues we are already seeing signs of AGI. For a fun listen, check out Rob Reid’s podcast on a shy super AI—maybe AGI is already here.

For technical and foundational work, the textbook Artificial Intelligence: A Modern Approach is the textbook for AI. “Attention is All You Need” is the paper that defined the transformer and kicked off much of the current wave. Stephen Wolfram has written widely on LLMs, and while his writing is technical, it is also accessible. He wrote What is ChatGPT Doing and Why Does it Work?, which is a great overview of how LLMs work. Alan Turing defined the Turing Test in “Machinery and Intelligence.” John Von Neumann seems as important as ever today; check out “The Computer and the Brain” or his “Theory of Self-Reproducing Automata.” Also, it’s worth paying attention to the intersection of crypto and AI. Bittensor is a good example of a project in the space (there are tons of projects implementing various layers of the AI stack).

Regarding good old-fashioned AI (GOFAI), What Computers Still Can’t Do by Hubert Dreyfus is a good but dated criticism. Norbert Weiner was an early pioneer of AI and what ultimately became expert systems in computer science. The Human Use of Human Beings raises similar concerns as various works today and is still relevant; Cybernetics is his more technical work. Relatedly, see Marvin Minksy‘s work and his framework for representing knowledge.

And just to round it out, Refik Anadol is an artist working with GANs (generative adversarial networks) and large datasets. His work at MoMA in 2023, Unsupervised, was mesmerizing. Trevor Paglen is a photographer who has created works that expose AI algorithms and their bias. A lot of his work over the years has focused on trying to see the unseen. Holly Herndon is doing interesting work with AI and music. Listen to her album Proto (it’s weird) or her version of Dolly Parton’s Jolene where she trained an AI (Holly+) on her voice to hit the notes she couldn’t. In science fiction, Robert Heinlein’s The Moon is a Harsh Mistress is a fun read and features an AI (and coined the term to grok), and Adrian Tchaikovsky’s Children of Time trilogy is a fascinating read that combines animal intelligence and artificial intelligence.

To simplify and summarize, if I were going to read just a couple of pieces from the categories above, I’d start with The Coming Wave by Mustafa Suleyman, “False Promise of ChatGPT” by Noam Chomsky, Atlas of AI by Kate Crawford, Situational Awareness by Leopold Aschenbrenner, and What is ChatGPT Doing and Why Does it Work? by Stephen Wolfram. That would be a good mix of overview, criticism, optimism, and technical introduction. But most of all, let’s circle back to the beginning: just start using the tools.