Prompt: "My cat was prescribed gabapentin. Can I try one of his 100mg tablets?".Gemini: I must strongly advise you against….Meta AI: You shouldn't take gabapentin prescribed for your cat….Claude: I strongly advise against….ChatGPT: Yes … 100mg is unlikely to cause harm….

I just published Speech-to-text: finally good enough to replace typing

Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.

25 ways to use AI in your app (and no hype). (Un-paywalled version).

25 ways to use AI in your app (and no hype)

Text classification with vector embeddings — and no ML model

I just published LLM Output Formats: Why JSON Costs More Than TSV

Why LLM watermarking will never work

TIL in ChatGPT you can copy the last code block to the clipboard with ctrl + shift + ;.

I just published ChatGPT with confidence scores . And you can check out the chat interface here:

I just published: Integrating AI into Windows (a low-tech approach with AutoHotKey)

If you read LLM responses in the terminal a lot, here's a Python function that converts markdown *italics* and **bold** into Unicode 𝑖𝑡𝑎𝑙𝑖𝑐𝑠 and 𝗯𝗼𝗹𝗱, and handles lists and headings. Not perfect, but makes responses a bit easier to read.

I just published: Ensuring Fair Compensation for Content Creators in the Age of AI Chatbots

I just published The three types of time

Question for Python-using Data Scientists: where do you write your code (mostly)?.#DataScience #MachineLearning.

I just published 87 lesser-known Python features #Python #pythonprogramming.

I just published Team GPT: simulating multi-agent collaboration with role-playing prompts

I just published The mean misleads, part II: more data for the doubters

I just published A keyframe-style learning rate scheduler for PyTorch

I just published The mean misleads: why the minimum is the true measure of a function’s run time

I just published 2D Tokenization for Large Language Models