I wouldn’t be too hasty about getting GPT-5 into Navarro. What I’m reading is not good. I’m not going to waste time posting a lot of it here, people can look it up.
At least, I hope users can get a choice of models. Personally, I’d like to see Navarro working with Claude.
I will agree with it not being quite ready–but I think that’s more a matter of the frameworks and harnesses around the model, rather than the model itself. Meaning–I can’t just flip a switch and magically deliver GPT-5–it’s going to take some work, redefining agents, evaluating, and steering.
I could maybe see a future where users could choose their own models (from other providers, I presume), it will be interesting to see how everything develops over the next year or so (or maybe even months!).
Right now, I think we want to focus on making sure Narrova is the very best it can be as an all-around narrative intelligence. From the looks of it, GPT-5 has massive potential, it’s just a matter of getting comfortable with how it will fit in with what we already have in place.
I tried interacting with Narrova today and it was awesome! The denser language of GPT-5 was actually great in a “writing buddy” interaction because it got things across in less words. It offered fantastic ideas, and presented everything in a logical and compelling way (often using tables). Most of all I felt like it totally understood the art of storytelling, we were conversing about deep stuff like character emotions and motivations, the repercussions of certain plot points and even reader reactions. The Dramatica stuff was then layered on as an additional clarifying lens, kind of a validation tool to make sure everything’s on track and holding together. I really appreciated that because that’s how I use Dramatica myself.
Now, I was using it to validate my own ideas – I explained a complex aspect of the story that is tricky to pull off in terms of character emotions and audience believability, and my plans to tackle it. Narrova went a bit overboard in suggesting a bunch of new ideas and possibilities that I didn’t really need, but that was okay because they helped me see how well my ideas worked. And it still did a great job helping to validate and refine my plans.
I also loved how it ended each response with a few open-ended questions (how might X impact the rest of the story/series?, was Y different in past books and could comparing that provide extra dramatic punch? etc.). I appreciate how these interactions inspire new insights and ideas of my own.
Wait. What? I thought GPT-5 wasn’t in Narrova yet. Now I am completely confused. And Jim hugged you, indicating…GPT-5 is in Narrroova?
I still want to say Narrova like Ricardo Montalban. Narrrrooova, with fine Corinthian leather.
I’m using Narrova every day at this point and I love it too, btw. If I get this stupid book done it will only be because AI and Jim Hull forcibly intervened.
No, Narrova is not running on GPT-5…yet…the posts i’ve been sharing here are tests. I have to update some of the scaffolding around everything to fully release it… but if @mlucas thinks it is as good as GPT-5, I can’t wait to see what he thinks when it IS GPT-5 LOL
Oh, I see, got it. Catch you on the rebound.
You can tell I don’t use AI/chatbots very much lol. Some of the language reminded me of Jim’s posts so I just assumed!
I use them constantly and am taking two LLM engineering courses I’m fascimunated by this weird pivot of technology, betting the farm on what is in many ways the least promising and least interesting technology branch in all of AI research. But it’s hard to argue, in America, with commercial success. And it works well enough. I am way less enamored of it than Jim is; but what it does do, it does well enough.
This is something I posted on LinkedIn that got a lot of impressions:
”I asked GPT-5 to give me a detailed timeline regard any of Nixon’s doings in 1983-4. It gave me a very generic top level list of things like tv interviews and his book. It left off things like, he almost took flight KAL 007, his visited san Clemente to look at his library site, he was working on securing a co-op in NYC. When I mentioned the specific things, it found the items easily: it just wasn’t smart enough to report them. As a deep research tool, this thing is pretty f***ing useless.”
I didn’t use asterisks on LinkedIn.
fwiw perplexity.ai (another agentic system) did a lot better, with a pro account.