86
How to stop being overly dependent on AI
(lemmy.world)
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
You need digital detox. Get the LLM apps off your phone, and maybe get something like this if its affordable: https://commodore.net/callback/
Sign up for university counseling. Trust me, I WISH I had done this.
Restrict yourself to "offline" LLMs on your laptop if you MUST use them, with all the search/agenic features turned off. This turns them into more of a utility and forces you to find the references yourself, to actually integrate blocks of code, and so on.
Treat this like smoking or some physical addiction, as that's what these digital addictions are.
Its probably unpopular here, but I'd also suggest you play with a "raw" LLM with completion syntax in mikupad, with the raw formatting tokens and everything:
https://github.com/lmg-anon/mikupad
Among other things, it dispels the "psychological mysticism" around them, and kinda shows your brain what they really are. A weather prediction model, for text. Not "AI" you brain can't help but anthropomorphize.
And:
This is not your fault.
Tech companies spend billions designing the chat UIs to be addictive, just like social media.
You can address it though, even if it feels like everyone around you is compelling you to use. I'm not even coming at this from the anti AI angle either: I've been hacking at LLMs since 2022, but what OpenAI, Google and such have done is beyond fucked up.
definitely gonna try this.
One thing I should have clarified. If you play with mikupad, you have to use an LLM API (or local LLM engine) that serves the old Completions API. With logprobs.
This is "raw" LLM access without any chat formatting, and its full output before sampling.
If you use OpenAI, Claude, or most other big providers, you are stuck with Chat APIs without logprobs, so it kinda defeats the point.
Honestly, I don't know of any API providers that offer this, but some role play focused ones might. Local inference engines (which are what I use) mostly have the raw API, but you need a pretty beefy computer for this.