Happy Monday and welcome to your weekly probe of what matters in AI.
One of my favorite aspects of learning about AI is the openness of information in the field. Not only is most research public, but ML systems are remarkably self-aware and LLMs are capable of teaching you about how they work with a little bit of interrogation.
The most common question I have received since starting this newsletter has been from people wanting to learn ML. Everyone has a different starting point, and there is no way to generalize a learning path that works equally well for all.
Here I have tried to collect some of the resources that should help anyone interested in the field to actually develop the core skills necessary to work in it. Regardless of where you fall on the skill spectrum, there should be something new and useful for you in the list below:
-
Ilya Sutskever shared a list of 30 ML research papers with John Carmack and said, “If you really learn all of these, you’ll know 90% of what matters today”. Here is that list of papers with summaries.
-
The Little Book of Deep Learning by François Fleuret is an excellent technical primer on DL and readable on your phone.
-
3Blue1Brown offer a seven-video playlist (circa 2 hours playtime) helping you learn the basics of neural networks and backpropagation.
-
Andrew Ng offers a very popular ML Specialization course on DeepLearning.ai which has received almost 5 million participants since 2012. Doesn’t require prior math knowledge or coding skills. Also available on Coursera.
-
What we learned from a year of building with LLMs. Spectacular series of articles, podcasts and videos about building products with and for LLMs.
-
Introduction to ML by MIT. A classic in the field, helping you build fundamental and deep skills around computational thinking and data science, including supervised learning and feature vectors.
-
Parlance Labs offers a series of video lectures on more advanced concepts like evals, RAG, prompt engineering and fine-tuning.
-
Why Machines Learn by Anil Ananthaswamy, which I recently started reading and can wholeheartedly recommend.
-
Reddit thread on /r/ArtificialIntelligence from a couple of days ago about with a few of the links above and more learning resources for complete beginners.
Please DM if you have any links I might have missed, as I’m working on a more definitive primer for the Vocab and Masterclass.
Until next week,
MZ
Relentless Curiosity in Science (3h)
Don’t miss Lex Fridman's interview with Anthropic CEO Dario Amodei where they explore the principles behind scaling laws, challenges and breakthroughs in AI safety, and the role of mechanistic interpretability in understanding AI systems.
Scaling is like a chemical reaction.
Future AI Business Models (4 min)
If AI were a toy, it’d be a pinata—you whack it, and all these surprising things fall out.
Nvidia’s AI Revolution (30 min)
Jensen Huang shares how Nvidia has transformed computing by creating tools that power industries like gaming, robotics, and scientific research.
For the first time, we’re creating skills that augment people, not just tools.
Infinite Possibilities (13 min)
Designers are learning to “play” with AI, treating it like a mysterious new material or even raising a kid, discovering surprising things it can do. It’s less about controlling everything and more about guiding users while figuring out how to work with AI’s quirks and potential.
Everything is changing so fast, but even if you froze progress, there’s years of design work left to uncover what models can do.
Ben Affleck on AI in Hollywood (4 min)
Craft is knowing how to start, art is knowing when to stop.
Chatbots Defeat Doctors at Diagnosing Illness (NYT)
A small study found ChatGPT outdid human physicians when assessing medical case histories, even when those doctors were using a chatbot.
OpenAI's artistic ambitions (NYT)
Visions of A.I. Art From OpenAI’s First Artist in Residence.
Simulated Personas
Microsoft framework for simulating various personas using LLM in order to test things like marketing, software or even brainstorming. It's developer centric rn but expected to trickle down to other use cases. Smart way of using agents with current limitations and reminds me of the SocialAI app.
LLM-powered multiagent persona simulation for imagination enhancement and business insights.
You are receiving this newsletter because you signed up on envisioning.io or Substack.