Now in paperback and Kindle!

Better without AI is now available in paperback and on Kindle!


I mostly wrote Better without AI in August and September 2022. I revised it in December 2022 and January 2023, and published the whole thing on the web mid-February. I expected to produce paperback and Kindle editions later that month, so they’re nearly a year overdue. Unfortunately these gaps have been due to health issues—now perhaps resolved—which made intensive work difficult.

This final version is somewhat clearer, better organized, and easier to read than the February one. I’ve posted it to the web site, which now has identical text to the paperback and Kindle.

The pace of progress in AI was famously overwhelming during 2022. Returning recently to the book for the final round of revision before publication, I was concerned that it might already be out of date. That was not the case. To my surprise I hardly needed to change any of the substance.

There is one section of new material. I’ll write about that in the next metablog post!

There’s two reasons not much had to change. One is that the book discusses mainly general principles, rather than rapidly-changing specifics. Those haven’t changed, and probably won’t for several more years. I don’t expect much of the book to be invalidated by events (although anything is possible).

The other reason is that, despite massive hype and massive spending, 2023 has not shown the rate of technical progress 2022 did. GPT-4 (developed late 2022, released March 2023) is significantly better than GPT-3.5 (mid-2022), but not qualitatively different. Nothing better than GPT-4 has been released since.

And, a year after ChatGPT was released, it appears to have had minimal effects (good or bad) on the world. This, despite the “ChatGPT is about to destroy many industries and jobs, and dramatically increase the power and productivity of others” hype. This could all change at any time, of course, because we still don’t know what’s going on inside. Nor do we have a clear understanding of how useful it actually is, or for what; opinions about that are mysteriously divergent.

However, there’s weak but growing reason to believe that this line of AI research and development may be nearing its limit. If so, that’s probably a good thing.

The first of many, perhaps

I write mainly books, not blog posts or standalone essays. I write them incrementally online, so readers get some benefit from unfinished versions. It’s long been my intention to publish them on dead trees as I complete them, because some people would rather read that way.

Better without AI is the first I’ve completed. It’s not the book I’d ideally like to publish first. The others all hang together, sharing fundamental themes, even if superficially their topics are very different. This one is an outlier. I wrote it because the matter is important and urgent; and also because I have some particular insight, due to my long involvement with AI research. It is probably not my strongest work. Still, it received much positive feedback when I first put it on the web. You too may find it to your liking.

Publishing this book in paperback and Kindle is partly an experiment to see how much audience there is for such editions. I’m often asked to produce them. I haven’t, because nothing has been finished before.

Time spent producing polished books trades off against writing the most important parts of the unfinished ones. If the Better without AI paperback sells 87 copies, I will have learned that it’s not worth taking time out from other work to generate more like it. If there are 8,700 readers for it, I will know repackaging my writing in conventional book form is worthwhile.

If it seems offline editions do reach many people that online ones can’t, I may publish paper extracts of the unfinished books. (Readers quite often request that.) There are some coherent chunks that I could probably finish fairly quickly. For example, I have notes and drafts for the missing bits of the nihilism chapter of Meaningness. With a new introduction and conclusion, it could stand alone as a conventional book, running to about a hundred printed pages when complete.