Tuesday, August 25, 2026

Saying something nice about Claude

I have been accused of being overly negative about AI, particularly large language models, and I have my share of issues with the technology. I do believe that we are pouring trillions of dollars into a bubble based on business models that can't possibly justify the investment and that the resulting crash will do great damage to the economy. I believe that the technology will never be capable of doing much of what has been promised and that much of what it is capable of (AI slop, disinformation, non-consensual sexualized images) is already doing a tremendous amount of social harm.

All of that said, large language models are an impressive advance in the way we interact with computers and that tools based on LLMs, used properly, can be extremely helpful.

Case in point, I have increasingly come to rely on Claude for advanced searches. I still use Google for the simple stuff, but for anything that would take me more than five minutes on the old platform, I turn to the chatbot.

The key point, and I cannot emphasize this enough, is that I'm not asking the chatbot questions. I am asking it to find and send me specific articles, books, videos, etc., which I then verify independently. Perhaps it has just been luck of the draw, but I have yet to encounter a hallucination under these constraints. Even if I did, however, it would not be a big deal since it would be discovered immediately.

Recent case in point: for a long time now, I've been trying to locate a video satirizing the way that certain people, particularly right-wingers, often become offended at self-evidently accurate labels. Here's the clip, which comes from a satirical troupe working for German public broadcasting. If you want to avoid spoilers, you might want to watch it before reading further.

I started with this prompt.

looking for a satiric viral video presumably set in the 30s where protestors shout "nazi" at an ss officer only to have him complain "why do you insist on calling people you disagree with 'nazis'?" 

It couldn't find the video, though it suggested some alternative options such as Key & Peele, so I provided more details.

sometime post 2010. possibly subtitled. similar to a college humor sketch 

This time it gave me the name, the date, the comedy troupe that produced it, and a link to the video on YouTube.

This doesn't make up for the sins of Anthropic but credit where credit is due. 

1 comment:

  1. Speaking of social harm, my wife got our first chatbot letter from our healthcare provider, the evil United Healthcare. The letter gave the reasons why the injection her doctor recommended was being denied. Instead of one reason for rejection, there were three, all clearly drawn blindly from a list and none of which were applicable to my wife (statements about her treatment history that were false, etc.). So she calls - and after waiting on hold - someone comes on the line and tells her oh just ignore the letter, the treatment was only rejected because the doctor called out a name brand and they would only approve the generic.

    ReplyDelete