Post

Some notes on blindly trusting what the adverts tell you about AI’s capabilities.

Thanks Steve, that’s an excellent article. I think the context window part is extremely important – it’s what I struggled with them most when I experimented with LLMs in the past. Gemini seems to have a longer one than ChatGPT, but still.

My memory is poor so if there were a big improvement to my workflow, having a tool that could help me reliably keep track of my thinking & theorising as it evolves with elements of sensechecking could be useful. Unfortunately, this is very much not it, so I stick to other methods.

The thing you're describing though, a tool that tracks how your thinking evolves and lets you sense-check against your own past ideas, that's a genuinely solvable problem. The issue is most tools try to be the thinker instead of just being a reliable memory. What methods are you using now?

Agreed! I don’t shy away from machine learning necessarily, as it does tend to be more reliable than LLMs in that respect. I don’t have a well defined process yet, but it’s a combination of using BlueSky as occasional dumping ground (I can search my own posts) which could be likewise done elsewhere

Yup, and for that a platform like MyMind works really really well in comparison; BlueSky is still good for occasionally getting some feedback on random musings from other people or having your own reflection based on sth someone else has written. That’s harder to replicate. Maybe a simple API link 🤔

Hmm, I'd actually push back a bit on MyMind for that though. It's great for saving stuff, but the retrieval is still keyword-ish. The real missing piece is searching by meaning across your own evolving ideas, not just bookmarks.

When have you tried it last? It’s evolved a lot over the last few years. If you write notes, you can search by any word in that note so it’s very association-friendly. ‘Meaning’ is harder to capture as that’s interpretive & may change from person to person, but you can self-tag as you go for that