I’ve been building a Swift app [1], compatible with OpenAI APIs, easy model switching across providers, and with hotkeys for OS integration to capture text and images. It’s far more minimal than most other LLM frontends I’ve tried, but it’s been sticky for me.

[1]: https://www.wvlen.llc/apps/tomo

More recently, `gemini-exp-1206` did quite well [1].

[1]: https://github.com/simonw/pelican-bicycle/blob/main/README.m...

Gemini 1206 is the new hotness in my books. I've moved my day to day LLM needs over to Google's tab for the first time. I'm not sure what they changed, but it deserves a good look. Claude 3.5-Sonnet (New) is fantastic as well, but the 2M token context window offered by Google allows you to suck in an entire code repository and reason effectively across the whole thing. Google is catching up...

I feel like ”quite well” is overselling it a bit. It did maybe better.

This is nice and has a very snappy UX.

For the Vercel folks, how are you all thinking about this tool compared to offerings from Github[1] or Sourcegraph[2]?

[1]: https://cs.github.com

[2]: https://sourcegraph.com/search

Cogno is a word game that is the inverse of Semantle. You start with a word and the goal is to guess as many of the top 50ish semantically similar words as you can before you make 5 guesses outside the top list. You get points based on how close the word is to the starting word, calculated by embeddings. More info here: https://www.cogno.fun/about

https://blog.danielcorin.com/ I mostly write about LLMs (lately) or whatever I'm learning or working on

I have some rough details on how I operated the AI tools here: https://adventure-of-penelope.vercel.app/credits. Unfortunately, I've lost the original log of the conversation.

Thanks for the kind words

I used very basic prompts including words like "cute" and "happy" and environment descriptions but not much beyond that (no art style or specific colors of the characters). I felt overall the character consistency has room for improvement but could have been a lot worse. Midjourney's default "style" has a sort of specific flavor to it.

This piqued my interest. I might give it a shot, either with this or something new.

~15 minutes with ChatGPT to turn the initial plot into a story with the tone and length I was looking for, an hour or two with Midjourney illustrating and a few more hours for the site from zero to up and running.