In one of my favorite guilty pleasure Heinlein books… One of the characters has access to a computer research terminal and got to spend an extended time with no job duties just using it to learn without direction and little guidance. This was refereed to as “feeding the elephant child” in the book. That notion stuck with me, as even though the book was published in 1982, it offered a number of touchstones to the way I remember ‘the world’ working from my youth.
My topic here isn’t about Heinlein or that book (which is why I haven’t named it) but to talk about what I spent all of yesterday doing. I say ‘all’ but really it was more along the lines of sit down for a moment, post a query, read the response, post a followup, go do something else for a bit before returning and so forth.
I captured the entire session by first copying my query to a blank document followed by a hard page break, then google’s AI assistant’s responce followed by a hard page break. Topic went all over the place, wormed around while I was digging for details for a long langishing trunk novel that got it’s start as part of NaNo in 2009…. When I put the finishing touches on the file it was 207 pages.
I could deal with the little suggestions that I use it to craft my prose, and in fact it’s suggestions were helpful in the realm of "I don’t really want to do that” sort of things, though it did occasionally hint at things that I was happy to see it make the connections about. I did catch it in hallucinating at least one instance, but after pointing out the oddity, the agent agreed it was an aberration and that it wasn’t intentional (It read like the sort of jokey headline I might blog with actually).
I didn’t then – nor now – feel like I need a shower from having played with the AI. While I recorded it’s sample generative prompts and prose, I have no desire to use them. At one time this manuscript existed on Google Docs, so I’m sure it was long since scrapped… It was nice to have an ‘interactive book club’ to talk about my ideas with…
Herein lies the rub, the thing is polite to a fault and rather sycophantic. I got a lot of compliments that I’m not sure apply. I’ve had other tools do this too. I have toyed around with LM Studio and the qwen3 ‘MoE’ models and get a similar sycophantic responce when I ask it to analyze story parts for pacing and coherence. Terms like “masterclass” get thrown around a lot. Makes me wonder just how much of it is actual, or if its just the terms of service of the agent that it has to find ways to keep the users happy.
The nuts a bolts of the analysis are good and fit with what I know, and at times helpful in spotting issues I hadn’t – particularly with correctly-spelled-word-substitutions – or duplicated parts or versions. But the way it slings compliments just has me wondering if the rest of it is as easily earned.
I mean, my writing might be that good – some of my alpha readers seem to think so – but I have my doubts, otherwise I’d be writing books for a living…
Makes me wonder even more considering the stinkers I’ve read over the years…
Anyway, have any of you spent any length of time exchanging posts with an AI on a book research task? I’d love to hear how AI treats others doing that sort of work.