2025-04-17
- A new day, a beautiful one in New York City. Back from Europe, so I get to watch the sun rise over the city for the next couple days. I don’t love being jet lagged, but this is a nice perk.
- Back at work, feels busy. Not an amazing vibe, but good to be busy, in some sense.
- Still feels like there’s a lot of room for computers to be better. Optimistic. Ofc concerning that most people are not using the tools super well today (and aren’t given an easy way to do that). Sort of reminiscent of that Voltaire quote about how there are so many books out there and so few people bother to read. And so many of those that do read seem content to consume pure drivel. Has something of an elitist vibe, sure, but mostly interesting that it seems to echo today with computers (access to infinite knowledge, mostly used to watch mindless videos). I personally don’t mind a small amount of mindless drivel, but it does seem disappointing that the metrics indicate people spend so much time with that kind of thing.
- A buddy sent me a link to the recent HN “fingerprinting” user contributions, noting that it seems like it would be effective in finding people’s “alt” accounts. I think they’re probably right. And then I thought, “ok, is there any way to improve the anonymity here while preserving the essence of the contributions and without a huge hassle?” And I think this could be something LLMs help with. Sort of like the recent demo where AI was used to alter the accent of an Indian call center guy to sound more like a native English speaker (to improve understandability when talking to e.g. Americans), you could run your comment through an AI filter that would apply a separate, consistent personality layer to your words. Not sure if current LLMs would be sufficiently good at this, but I think the basic idea is sound. LLM as filter/flexible transformer is a good/repeatedly interesting frame (in addition to categorizer, summarizer, planner, reasoner, etc).
Date
April 17, 2025