Now I'm curious about what kinds of things YOU find LLMs useful for. (Since I've been following you on Quora, Twitter, and Bluesky for around a decade now… and I think I'm probably justified in considering you a language-learning genius and a world-class expert on Levantine Arabic dialects. 😎)

Show Post

So that's the list of tasks where I personally have found LLMs to be actually useful. I've tried Gemini or ChatGPT for a few other research-y or math-y tasks, but when I immediately get a combination of high confidence w/poor logic and consistency (like in my initial post here 🥲)… I give up.

Show Post

3. Research on tax rules. (A crazy complex morass for those of us who have to file taxes in both 🇺🇸+🇨🇦.) Gemini is quite good at trawling through gov websites, forums, and blog posts and summarizing and referencing relevant info. (5/5 🏆) 4. Generating silly images for jokes and fun (🤷🏻‍♂️)

Show Post

2. Analyzing unknown binary data. I threw some binary blobs at Google Gemini while writing github.com/dlenski/very..., and it was darn good at finding structural patterns like TLV or Protobuf. Better and faster than what I could do alone with Ghidra or a hex editor. (4.5/5)

Show Post

More recently, though, I've found a few cases where LLMs are useful (to me): 1. Coding w/Claude (paid work plan): Its Python code SUCKS in terms of consistency/maintainability, but it can generate and test it quickly so it's *somewhat* useful for quick-and-dirty analysis of messy data. (2-3/5?)

Show Post