My reaction too when reading all the RAG is dead tweets earlier today. Huge amount of optimism that the context window is also usable in practice for real problem solving and not just in theory. Could very well be true I just don’t super know.
@karpathy
-
Automating Call Centers: Will 50% of Jobs Disappear?
By
–
It’s a little too soft to resolve properly. Maybe one example. Pick something everyone thinks is surely easy to automate yesterday, eg call centers. Number of employees across the 5 biggest call center companies falls by 50% by year X.
-
Moving AI Predictions to Betting Markets for Truth
By
–
Let’s take AI predictions from blog posts, podcasts and tweets and move them to betting markets, our state of the art in truth. My struggle has been coming up with good, concrete, resolvable predicates. Ideally, predicates related to industry metrics and macroeconomics. Eg
-
LLM Text Interfaces Need to Evolve Beyond Command Terminals
By
–
Writing text back and forth with an LLM is like we're all the way back to the era of command terminals. The "correct" output is a lot closer to custom web apps written just for your query, information laid out spatially, multimodal, interactive, etc. Will take some time. https://t.co/dMAQEjE8GF pic.twitter.com/N849kmMTZQ
— Andrej Karpathy (@karpathy) 31 mars 2025Writing text back and forth with an LLM is like we're all the way back to the era of command terminals. The "correct" output is a lot closer to custom web apps written just for your query, information laid out spatially, multimodal, interactive, etc. Will take some time.
-
Vibe coding versus traditional code review practices
By
–
Good post! It will take some time to settle on definitions. Personally I use "vibe coding" when I feel like this dog. My iOS app last night being a good example. But I find that in practice I rarely go full out vibe coding, and more often I still look at the code, I add… pic.twitter.com/EAHAZPb4tI
— Andrej Karpathy (@karpathy) 23 mars 2025Good post! It will take some time to settle on definitions. Personally I use "vibe coding" when I feel like this dog. My iOS app last night being a good example. But I find that in practice I rarely go full out vibe coding, and more often I still look at the code, I add
-
Karpathy Shares Swift Learning Conversations with ChatGPT
By
–
A number of people asked If I can share the convo and yes sure – these were the 4 convos with my super noob swift questions lol: 1 starting the app https://
chatgpt.com/share/67e02d8a
-994c-8007-bf44-a63127cbbbb2
…
2 enhancements https://
chatgpt.com/share/67e02d99
-5e68-8007-b30c-80c9ed7f3693
…
3 adding AppStorage to persist state over time -
Learning Without Documentation: Following ChatGPT Instructions Instead
By
–
I didn't even read any docs at all, I just opened a ChatGPT convo and followed instructions.
-
Building iOS App in Swift Without Prior Experience
By
–
I just vibe coded a whole iOS app in Swift (without having programmed in Swift before, though I learned some in the process) and now ~1 hour later it's actually running on my physical phone. It was so ez… I had my hand held through the entire process. Very cool.
-
AI-Generated Quiz Contains Factual Error About Makemore Video
By
–
It's a fun idea in principle but question #2 on a random quiz i took is already incorrect. The "correct answer" claims that in makemore bigram video the progression of architectures includes RNNs, which is wrong and RNNs were not covered. The trouble across the board is that AI
-
Breadth is free, depth is expensive in neural network compute graphs
By
–
yep exactly, great work spelling it out step by step.
sometimes I talk about it as "breadth is free, depth is expensive" in the imagined full compute graph of the neural net. afaik this was the major insight / inspiration behind the Transformer in the first place. The first time