Cupcakes

I came across a haunting fanfic (a short poem) that strongly implies (without saying it explicitly) that a certain fictional character was Jack the Ripper. I was curious if this allusion could be detected by an AI, and asked 3 of them (ChatGPT, Copilot and Gemini) to analyse the poem and note any hidden meanings. While their analysis was detailed and mostly on point, none of them guessed it! Moreover, even after I revealed the secret, Copilot asked if this interpretation could be intentional or accidental. Even though it is blindingly obvious to any reader who's into Jack the Ripper lore. (The other two agreed right away that it was brilliant and absolutely fitting.)

It's like what I had posted a while ago, about puns that AI can't understand. I wonder when they will be able to look deeper into the context and put all the clues together...?

On the positive side, with the help of Copilot's clarifying questions (it can never stop a conversation by itself, provoking the user to keep chatting ;) I came up with a satisfying ending/resolution of the story, and ChatGPT even wrote a couple of snippets for me. I guess it can be used to scratch an itch, if you can't find a fanfic you want to read... The problem is, you need to know exactly what to write, with all the twists... whenever I asked to write something from a vague prompt, the results were rather boring.

I browsed a few AI prompt engineering courses on Coursera (Im still a freeloader, without access to graded assignments). Learned some new and exciting stuff, but it's impossible to listen to these lectures word by word without skipping, because there's so much verbosity, hype and trivial common-sense advice, and because a lot of examples are so mundane and corporate, created by adults for adults: summarize spreadsheets, summarize documentation, write reports, create time schedules, draft HR interview sessions and such. And I was rolling my eyes at meta tasks like asking AI to write prompts for you. 

But after a nice chat with Gemini for an ungraded assignment, where instead of some tedious project management task, I asked for a motivational boost for game development, I suddenly got an idea that it would be cool to generate a daily motivational email from a LLM in the same vein. (Using a set of predefined quotes is really boring, as well as getting generic motivational quotes from the Internet.)

So I went meta and asked it to write the appropriate prompt. And the AI rose to the task like a morning star :) It gave me a badass system prompt of ~30 lines (based on the info I shared with it in the current session), and how to incorporate it in the Python litellm script (that was the template from another course, currently on hold, it uses OpenAI API), and later (after subsequent discussion) how to make it even more personalized by reading my latest git commits (via Bitbucket API) and comment on them, and how to send email over ssh (which I will try later), so that was also practicing the "iteration" concept which I didn't bother to do much in the course assignments. 

Everything is so cool and fascinating! It makes so much difference when you work on something you're really excited about.

(Wow, such a deep thought, worthy of an AI...)

Btw I'm really warming up to Gemini - never used it often until recently when it was recommended for the assignments in these series of courses from Google. (Maybe that was their main goal...) In the motivation script, I use ChatGPT (gpt-4o-mini).