Just messaging with your LLM

In which I
try Poke

an attempt
at
overcoming
the
threaded
conversation
model

In which I try Poke, an attempt at overcoming the threaded conversation model.

I've always wanted a simpler interaction model for my chatgpt-like app of choice: a single thread without worrying about "conversations", just talking with the LLM.*

In the conversation model every new chat intuitively looks like I'm adding "noise" to the app. Like a growing number of notes. Don't we clean our digital notes every now and then? But also: "I asked something stupid, should I delete this conversation? Maybe next time I'll use a temporary chat for these kinds of questions... or maybe not", or "Is the conversation context growing too much?".

All these questions should not exist in a good UX. You should just be talking with your LLM, and it should remember what you told it yesterday. That's it. The conversation model mostly exists because of current technical limitations.

I've been trying Poke by Interaction and I'm finding it a serious attempt at building this UX. You write to Poke through iMessage or Whatsapp and there is no concept of conversation. You just write to Poke asynchronously: you send the message, put your phone down, and wait for the notification.

Poke has the usual capabilities: reasoning through problems; online search; integrations with emails, calendar, github etc. They intend this to be a proactive assistant: you can ask it to notify you at certain times or when something happens; in other words, you can schedule Poke to talk to you, instead of the other way around, and it looks natural if we are receiving a message from "someone". Two examples: "Check the web every day at 19 and send me a message when there is information about this particular event in my town.", or "Send me a message each morning at 9 if there are emails from at least 3 days ago, mentioning this thing, that don't look like spam."

Another funny idea is that there is no fixed price. You decide it together with a snarky "AI negotiator" during onboarding! Someone got it to 0.01$ -- don't say you're rich!

*Except for work tasks like coding agents, or things I want the LLM to be as unbiased as possible by my personal context.