UTF-8 I already knew about the "confusables", e.g.: e vs. е. Which look ~same but are different. But you can also smuggle arbitrary byte streams in any character via "variation selectors". So this emoji: 󠅧󠅕󠄐󠅑󠅢󠅕󠄐󠅓󠅟󠅟󠅛󠅕󠅔 is 53 tokens. Yay https://
paulbutler.org/2025/smuggling
-arbitrary-data-through-an-emoji/
…
@karpathy
-
Unicode Variation Selectors Enable Arbitrary Data Smuggling in Emojis
By
–
-
Inspiring Domain-Specific Educational Content Through LLM Introduction Video
By
–
Part of the reason for my 3hr general audience LLM intro video is I hope to inspire others to make equivalents in their own domains of expertise, as I’d love to watch them.
-
Language Priors Steering Thoughts in AI Models
By
–
Cute idea, reminds me of “let’s think step by step” trick. Both lean on the language prior to steer the thoughts.
-
Hyperbolic hosts Llama 3 405B Base in bf16 precision
By
–
afaik Hyperbolic is the only place that hosts my <3 Llama 3 405B Base, and in bf16 precision. So thank you 🙂
-
Deep Dive into LLMs: Complete Training Stack Overview
By
–
New 3h31m video on YouTube:
"Deep Dive into LLMs like ChatGPT" This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental -
Building an LLM Council with Debate and ELO-Based Chair Selection
By
–
This is cool! I want to build something like it too because I want my LLM council. First they all run, then they debate, and then the chair of the council (the highest ELO model) works out the final response.
-
Building a Real-Time Battleship Game Between Two LLMs
By
–
Earlier, also ~hour of vibe coding, I built a Battleship game wired up so that you see two LLMs (any two models you select) are fighting each other in real time. I don't have super strong stats yet on this but I believe 4o beats 4o-mini, lol.
-
Custom LLM Reader App for Interactive Book Analysis
By
–
Last ~hour I built a custom LLM reader app so while I read Wealth of Nations I can ask questions about any paragraph. When you click a paragraph and "Ask" it calls an LLM, builds context window of what this is, copy pastes the full chapter, the paragraph, and the question. Works
-
The Spectrum of LLM Assistance in Programming: From Traditional to Vibe Coding
By
–
The amount of LLM assist you receive is clearly some kind of a slider. All the way on the left you have programming as it existed ~3 years ago. All the way on the right you have vibe coding. Even vibe coding hasn't reached its final form yet. I'm still doing way too much.
-
Vibe Coding: Embracing AI-Powered Development Without Traditional Constraints
By
–
There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper