Melanie Michell is an AI researcher and professor at the Santa Fe Institute. In her book “Artificial Intelligence: A Guide for Thinking Humans,” she seeks to make artificial intelligence understandable for laypeople, and for the most part, achieves that goal. She begins by discussing the history of artificial intelligence, starting from the “perceptron” and working up to deep learning for natural language. Michell explains each concept in concrete terms and apt metaphors, with minimal technical jargon and math. Each “AI spring”, heralded by new breakthroughs and followed by breathless claims about potential revolutionary applications just around the corner, are inevitably followed by “AI winters”, where the technology falls short of expectations. She also delves into philosophical concepts such as the alignment problem, the singularity, and consciousness in AI systems. ...

Generative AI PC Build Log Why With the emergence of generative AI, I finally had a compelling reason to build a new PC. Most deep learning projects require a dedicated GPU, making them difficult to run even on an M1 MBP. While cloud options like Paperspace and Colab exist, the GPUs they provide access to are limited. Paperspace, for instance, only allows access to the most basic GPU and denied my request for access to the RTX 4000. AWS was also an option, but it was somewhat inconvenient to use cost-effectively since you would have to stop the instance and snapshot the volume to save on costs. Having a local GPU is crucial to quickly run experiments at a small scale before deploying them to the cloud to run on bigger hardware. This helps me minimize the feedback loop and make faster progress. I had previously built a computer a couple of times in high school, mostly for gaming. I didn’t have a good reason to do it again until now. ...

A Time Capsule of Startup Interviews “Founders at Work” by Jessica Livingston is a chronicle of startup stories, mostly from startups in the 80s and 90s. The book consists of a series of interviews between Livingston and the founders or early employees of various startups. It feels similar to tuning into a podcast from a decade or two ago, reminiscent of modern equivalents like “How I Built This” or “The SaaS Podcast”. With 32 interviews, the book offers a fascinating peek into the early days of several notable startups. Livingston, as one of the founders of the famous startup incubator Y Combinator, is well-suited to act as the interviewer. ...

Summary Made to Stick, by Chip and Dan Heath, explores the art of crafting memorable and influential ideas by outlining six key principles: Simplicity, Unexpectedness, Concreteness, Credibility, Emotions, and Stories (SUCCESs). Through engaging examples and stories, the book demonstrates how to apply these principles to create “sticky” ideas that leave a lasting impact. By overcoming the “Curse of Knowledge” and employing the SUCCESs framework, readers can improve their communication skills across various fields and contexts. Made to Stick offers practical guidance, backed by research, to help anyone create and share ideas that resonate with their audience. ...

Kodak was once the dominant player in the photography industry, selling disposable cameras, film cartridges, and developing photos. However, they failed to invest in digital cameras, which they had invented. They were making so much money on film that they didn’t want to speed up the process of making the old technology obsolete. This approach made sense to Kodak, but customers increasingly preferred instant digital photos. Now, most Gen-Zs have probably never heard of Kodak. ...

TLDR: I built a semantic movie search app with GPT-4. You can check it out here: https://moviesgpt.net/. And here’s the Github repo: https://github.com/sinzin91/moviesGPT the hero keeps forgetting who he is Idea I’ve been thinking a lot about potential products that one can build around LLMs. One idea I mentioned jokingly to some friends was to use GPT as a natural language interface to a movies API. In theory, I thought, you could ask GPT about movies semantically, meaning you could search based on what the movie is about or its content, instead of just raw keyword search. For example, I could search for “movies where all the main characters die” and get back “The Departed”. Then I could use the response to query a movies API like The Movies Database. Google is surprisingly bad at this, returning links to listicles instead of the movies themselves. ...

TLDR “Peak: Secrets from the New Science of Expertise” by Anders Ericsson challenges conventional wisdom about talent, skill acquisition, and what it takes to achieve true expertise. Ericsson, a psychologist and researcher in the field of expertise, highlights several experiments to back up his claims, although some of the cited studies had small sample sizes. Despite containing some contradictions, the book offers a fascinating exploration of the science of expertise and empowers readers to pursue their path to mastery. Ericsson demonstrates that deliberate practice, not innate talent, is the key to mastery. ...

TLDR I created SocratesGPT to test the concept of using AI to generate questions about a given topic. Its goal you learn topics better by asking you questions about it. I also provide some technical details on the implementation. Why? Over a year ago, I began using Anki, a spaced repetition flashcard app that helps me retain information over the long term by taking advantage of the Ebbinghaus forgetting curve. Memory is like a leaky bucket, and it was incredibly frustrating to realize I’d forgotten most of what I learned. Anki has enabled me to confidently tackle difficult technical topics that I would have otherwise avoided. ...