malborodog

I am going to ask gpt 4 to implement this for me so I can see what it looks like. Amazing.

why not just have a tonne of html files on disk?? I'm confused at why this requires a sophisticated backend.

(1) could you hash everyone's mp3 files so that if person a had the same songs as person b, you only need to store them once? (2) people obviously storing copyright stuff in there -- do you have legal issues?

+1 for hermitix -- funny to see it come up on HN!

Absolutely glorious. thx

How could this underlying idea be implemented in a python script from a running selenium driver?

What’s the infamous Dropbox comment? Sounds hilarious.

Can't use third party service -- compliance stuff. This is helpful though. Using another model to tidy it up -- maybe Alpaca -- could be an option too. Then we'll just do speaker separation etc. manually later.

Hm, I'm on Mac so it takes up a bunch of ram and I'm not used to this workflow. good point though.

I have to transcribe a tonne of Chinese interviews soonish -- any further thoughts or experiments you can think of? Maybe some preprocessing steps to the audio? For example, cut it into one minute chunks with some overlap, then transcribe those, so that it can't skip those bits...? Or can we finetune it on a library of Chinese mp3s + transcripts?

why not just run whisper from the command line directly? Why put it into a docker container??

Nah it was done kinda tastefully here and it’s helpful to know the comps

Anyone know best way to punch up his function a little so it tees >> out both the q and a to a file? Like history but for chat gpt.

Yep. This is such an obvious use case. Have you seen the best thing out there that does this? Where can I load 100gb of pdfs and ask questions about what's in them??

What's the startup?

Just because it produces sequences of tokens that we typically experience as evidence of reasoning does not mean that it reasoned to produce them. We know -- generally -- the process by which those tokens were produced. It is not deterministic, but its properties are understood. It's a language model! It is not a mind.

Is it? Why does everyone dunk on him?

Link to the 990 dead — should have used archive.org or something.

+1 for interest there

Why isn't it easily dealt with by doing that??

Someone still has to talk to the computer and make it do stuff. That’s us. That won’t change even if how we do it changes.

Beautiful. Got it; thank you.

When you say text indexing and serving http are library functions, what do you mean? Also, is the language here go or what? Since you said python is too slow and then necessitates all the infra to manage it.

Can you explain that again differently? I didn’t understand that captcha point. It feels important though.

Implementation details. Check out what's going on with LangChain, augmented retrieval, etc. We'll be able to create knowledge bases on specific subjects with vetted data, and get the bot to retrieve and summarize appropriate results while providing a citation to the original source.

Or ya know, just use it and don't cite him?? Seems pretty easy!

This kind of problem is (trivially?) solvable through the ReAct framework, like LangChain etc. Basically you get good data, vector embed it, and make sure the LLM knows where to look for accurate information.

No, don't think so -- I would be going from one digital ocean VM to another. So the target machine will be in operation. I honestly might just leave it and think of some other use for the server to justify keeping it running in current state.

Thanks for explaining this though - really appreciate it.

I want to click _more_ on this. Have you set out this view at length elsewhere? What is the metagame openai is playing? Who are they competing with (Google?? FB?), and are the stakes higher than simply making more money? Also - there is presumably not going to be a huge moat to these LLMs so why would being first mover convey outsize advantage? It's genuinely a fascinating way of thinking about it but feel like it needs to be fleshed out a lot more. Also, the full implications 5, 10 years into the future are still very cloudy to me.

So should I just use dd or something? Or can I actually just rsync from one root to another and that will capture all of the dependencies, nginx config, literally everything? GPT-4 tells me I can just shut off everything that might change the file system in source (nginx sshd etc.) then literally rsync from root to root with

sudo rsync -aAXv --exclude-from=exclude-list.txt --delete --numeric-ids --rsync-path="sudo rsync" / user

(the exclude list is just /dev /proc /sys /tmp /run /mnt /media /lost+found).

At a certain point you start to wonder if saving the 50 bucks a month is even worth it....