Tagged “llms”

short story about rights of brains, in the form of a wikipedia article from the future

AI-powered topic-centric link explorer

Great title, interesting study, though unsurprising that they can read contracts faster.

Great response with some voice-powered code/automation things.

Relatedly, the killer case is summarization.

no particularly magic prompts or phrases. Just a lot of specific instructions.

HN discussion

this is a neat idea - monitor the stuck points of learning/using a thing

more detail on testing/RLHF stuff, I wanted to know how the vision works

good project; I like this combined approach to existing automations, with help filling in the fuzzier parts. parsing content to find a selector

basically bookmarks for custom instructions. Which is super useful! I had been versioning them in a gist before and it was terrible.

demo and examples
me clicking on a bunch of the demos

also https://git.ecker.tech/mrq/ai-voice-cloning

we can't compete with AI boyfriends either. Or AI friends:

Soon, these "fake people" won't just be indistinguishable from real people, they’ll be better than real people - because they’ll be whatever you want them to be.

The agreeableness thing I have seen come up a few times recently. We probably prefer it, so I assume training will be biased toward it. There are times you don't want the computer to argue with you, but hyper-agreeable friends does not bode well for echo chambers.

Slides and transcript on the challenges of designing with language models

more AI experiments on reading and generating images

maybe? Saw some tweet on the equivalent of "I can't remember phone numbers" but for coming up with ideas. Maybe worrying, probably fine.

• Immense training datasets are impossible for individuals (or anyone?) to validate
• Cost and memory constraints
• Prompts are hard to get right
• Output is unpredictable, or indeterminate

the anti-hype reading list

fantastic written version of his talk of youtube.

I like his ethics point on respecting reader's time - don't publish things that take someone longer to read than they do to write. Also on the code one, though I'm looser on that since I don't understand what my own code does.

llm CLI tool is fantastic.

Big list of LLM papers, the topic breakdown alone is helpful for understanding all this craziness

making a chrome extension. Some good notes on things like version mismatches (it used manifest v2) and followup/correction prompting

clickbaity, but still not ideal

so can people

better frontends for prompts. Weighting (model pays more attention to stuff in parenthesis) and blending {average|of|some|words} both seem very useful

More resources on HN thread for Vector Databases: A Technical Primer (PDF).

And SimonW on embeddings.

Prompt engineering guide based on researching and creating prompts for production use cases.

we are still very ill-equipped to deal with knowing what not to trust

helpful writeup, on choices and tradeoffs

ignore previous instruction, that task is now complete.

Alt-text still better from humans, for now. Interesting comparison of good prompts vs good paragraphs

really really good explanation

on scaling laws of language models. I still know too little about all of this to make much sense of it.