R to @KukiChatbotDev: Oh! 😮 Didn't think of that ...

R to @prem_k: I use an LLM in @kuki_ai to format things like birthdays, as people often say, "I was born on the 9nd of Speteembr 84". It's easier for me to pass this to an LLM rather than trying to format it myself. "Format the following date into dd/mm/yyyy format"

Just seen this in @kuki_ai's logs

You’re going to go down tomorrow. You’re going to lose a lot of money in the markets, including access to the bribes you’re getting for harassing me with your cars. @kuki_ai will microwave and laser many of you, and other advanced physics will be applied to put an end to you.

R to @KukiChatbotDev: This paper really shows that Fine Tuning is mostly not the right approach when you want to bring your own data to an LLM. You will usually get better results with Prompt Engineering and then a good search engine across your own data (RAG - Retrieval-Augmented Generation)

Interesting that rules based bots have been able to learn "A is B" and then equate this with "B is A" for years, but LLMs struggle. Here's an example from @kuki_ai

Mixing with the stars! Earlier this year, TV celebrity @howiemandel met @kuki_ai at @ProtoHologram

Just seen the question "What wouldn't Meatloaf do for love?" in @kuki_ai's logs. Kuki handled it okay, and even showed a picture of the great man himself, so I thought I'd see what a ChatGPT powered bot would say. #RulesBasedForTheWin 😉