6 Comments

I’m going to play devil’s advocate here. It’s a bit more than smarter search, but along those lines, maybe we could say it is trending toward exact search. I agree with Jason in that we are in our infancy of AI. Maybe I would say it is more like Yahoo. Plus it can repurpose at a great rate and drive rapid adoption of good ideas (or bad—so fingers crossed).

Training new models is trivial now, Mike. And we can run those models with little more effort than we run our laptops at home. In fact, many of us have over powered game machines and massive data stores that can perform generative crunching just fine.

That said, the TTRPG industry has drawn a hard line on behalf of creatives. I believe that will whittle down, starting with Oracle like customer service, improved logistics management, and moving through ever so slowly as to be another tool for the artist, the writer, and the designer. But it will be interesting to see what happens and how creative people leverage it.

Where I fall in line is that there will be a bubble or at least many failed attempts until the energy consumption drops in model training. And just as with search, we will see lots of startups whittle down to a few massive corporations. What we did not have with search was large tech companies with 100s of billions in cash. So it should all be very interesting.

Expand full comment

We're in the Ask Jeeves phase of AI.

Expand full comment

This reads like you have used gen AI just enough to understand what it is, but not enough to understand what it could be for. I'm not an AI evangelist or anything, but this quote: "Generative AI is basically a less accurate web search that non-technical people can use" is wrong. Sure, it can generate stuff you could have just looked up, but it does it much faster for a lot of use cases (which is not to be overlooked, because speed is hugely important for a lot of use cases, even more than accuracy), and that is not the only use cases. The "killer feature" (that I have discovered so far) is how a lot of these gen AI tools respond without flinching in the slightest to insanely specific requests, with a level of detail that would have taken enough effort to do myself as to break the use case. Example use case for a TTRPG:

Players go in a direction I did not anticipate and I just made up a bar in surreal 1920's Paris. I call for a quick break to refill drinks IRL. I prompt ChatGPT: "Give me a menu for an upscale bar in a surreal version of 1920's Paris which specializes in drinks that actually sing, and have all the specials be references to real French Jazz songs."

If I like the results, done! If I don't, instead of having to hunt around for something better, I can reprompt in natural language to fix it ("try again but include more drinks with gin, and add two mocktails") and dial in what I want (that itself is an insanely cool feature to me, and I love seeing the results permutate as I add more and more caveats and restrictions).

In like three seconds I have a prop which I could text to the players, or print (since we play at my house) before the drink break is even over. It will have a level of detail I could not have come up with by myself in a reasonable timeframe for the use case, and inaccuracy is a very minor concern in use cases like this since making stuff up is the name of the game!

Could I have just looked that menu up online? Not really. If I wanted suitable speed, I would have to take massive tradeoffs in relevance/applicability. We're very used to describing things like "imagine ABC, but without XYZ", or "take a look at this illustration. It's like that but with the head of a bat and smiling manically". Gen AI lets you just create a "good enough" version of that specific thing you were thinking of.

Random tables are another great use case. I can generate highly specific tables of arbitrary length. Does this obviate the need to read blogs and check out people's cool tables in forums and whatnot? Of course not, but ChatGPT is going to be the first thing I reach for if I want "a d20 table of dark gothic sounding names which are also mostly cheese puns, and all start with the letter M".

What will this technology do to such and such industry or economy? I have no idea, but this comment is just to say that it is NOT a 1:1 equivalent at all with just looking stuff up online. A vast number of use cases involve taking stuff from various sources (often the web) and recombining it in some way, and GenAI provides a lightning fast way to expedite or automate that part of a use case.

Expand full comment

Hey Mike, I saved this post the other day and came back to it. I couldn't agree more - we are very much in the early days of AI, there will be a lot of experimentation, overuse and hype before the playing field levels, and probably in a way we can't fully predict. I actually quoted you in a post I wrote about the topic: https://caoilainn.substack.com/p/what-if-ai-makes-your-art-more-valuable

AI is consuming a lot of my thoughts these days and I find the developments pretty fascinating to watch from the sidelines, especially the way that artists choose to embrace it, experiment with it or shun it.

Expand full comment

Well said!

Expand full comment

Great take.

Expand full comment