Content
@
0 reply
0 recast
2 reactions
antimo š©
@antimofm.eth
[1/x] The main problem with LLM interfaces is it's too hard to retrace your steps. Since it's also very easy to start a new convo, power users end up with hundreds of convos (which is another problem, out of scope for this project). But ChatGPT for instance does have a good search, which means you can find the right convo. However if a few days have passed, regaining the context of a longer exchange is hard: ā unless explicitly and properly requested at the end of the convo, chats don't natively have a TLDR at the end ā re-reading the exchange mean combing through (1) long responses, (2) verbal diarrhoea transcripts (which are not marked as such + cannot be replayed) and (3) changes of context ā the biggest problem is the visual context window: screens only have so much real estate (especially on mobile) I believe there are easy solutions to these issues that would significantly improve the UX of LLM products. Let's see how much I can get done between today and tomorrow
2 replies
0 recast
5 reactions
antimo š©
@antimofm.eth
PS the pics above show the last 2 messages of a real convo (https://chatgpt.com/share/67c34a05-b370-8004-9398-e384282f5eeb) that I'll be using as template for the project It's not too long, but long enough to illustrate the point; plus it has a small hallucination (she thought one of my voice notes was spoken in Welsh and so replied in Welsh lmao), which adds another problem / idea (4) add translation / action buttons to each message
1 reply
0 recast
2 reactions
eirrann | he/him
@eirrann.eth
Can't you just ask any old chat instance to give you a tldr if you can't remember what it was about?
1 reply
0 recast
0 reaction
