Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Replacing my best friends with an LLM trained on 500k group chat messages

izzy.co

Post reply on HN
Print view View original

271–280 of 371 posts

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#271

Post by jwithington »

Jcowell wrote: post #238
faeyanpiraat wrote: post #196

Earlier quoted context omitted.

Thanks for the reminder. FOr some reason I had to put that down a couple episodes in. May re-watch it again from the beginning. Hope I haven't shelved it due to woke cringiness as I still don't tolerate that.

Tangent but what was woke about the 2004 adaption of Battlestar Galatica?

The parent comment was talking about Caprica, which is different than BSG.

But some people get worked up about how the OG Starbuck was a hard-drinking, hard-partying man but the 2004 Starbuck was a hard-drinking, hard-partying woman lmao

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#272

Post by marvin »

mabbo wrote: post #49

While I love all these stories of turning your friends and loved ones into chat bots so you can talk to them forever, my brain immediately took a much darker turn because of course it did. How many emails, text messages, hangouts/gchat messages, etc, does Google have of you right now? And as part of their agreement, they can do pretty much whatever they like with those, can't they? Could Google, or any other company…

At least if you're in the EU, you are one GDPR deletion request away from removing the legal grounds of such a simulacrum of you.

Not that I'm in favor of the way the GDPR has played out in general, but, you know, at least in this instance it delivers on its promise.

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#273

Post by wintermutestwin »

Zetice wrote: post #165
mabbo wrote: post #49

While I love all these stories of turning your friends and loved ones into chat bots so you can talk to them forever, my brain immediately took a much darker turn because of course it did. How many emails, text messages, hangouts/gchat messages, etc, does Google have of you right now? And as part of their agreement, they can do pretty much whatever they like with those, can't they? Could Google, or any other company…

> And as part of their agreement, they can do pretty much whatever they like with those, can't they? What? No haha, they aren't able to read your emails or use them as training data for an LLM.

>they aren't able to read your emails or use them as training data for an LLM.

I'd love to see the policy or law that prevents Google from doing either.

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#274

Post by izzymiller »

thih9 wrote: post #226

The article frequently mentions costs but never gives any numbers or a point of reference. As an outsider to LLM and training I find this disorienting. What would be e.g. a total cost for a project like this?

Cost me about a hundred and fifty bucks, give or take. Continued GPU inference is on the order of ~50 cents a minute or something like that— but it's serverless so negligible. I think you could do it for significantly cheaper with some of the newer models i mentioned!

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#275

Post by Fauntleroy »

wpietri wrote: post #268

This is great! One of the things that I find horrific about a lot of LLM projects is that people are taking them so seriously. "They're going to destroy the world!" Or, worse, "I've taken $25m in VC money to see if I can destroy one part of the world!" But this is lighthearted fun. Instead of putting it in a context where the LLM tendency to bullshit is a problem, here's it's exactly what is needed.

It's not that existing tools are particularly dangerous; no, they're just really good and interesting text autocomplete systems. The danger lies "two more papers down the line," where they have 10-100x the capabilities they do now. Those who already have immense wealth and power will be able to deploy as many human-like internet agents as they'd like, and the potential evil applications of that are endless.

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#276

Post by J5892 »

AlchemistCamp wrote: post #231

I've heard several people doing this and chatting with simulacra of their friends, but you can always just send your friends a message and chat with the real version. My first inclination was always to try a conversation with a virtual me (yay recursion!) I've always thought that would be fascinating. Or scanning in my old journals from when I was a teenager and training it on that. Once this technology improves a bi…

Now I'm curious about training a bot on my IRC logs from the early 2000s.

It'd be like a chat time machine. I'd love to go back and bullshit with long-lost online friends about modding Halo:CE on Xbox again.

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#277

Post by J5892 »

overthrow wrote: post #47
sdwr wrote: post #17

Wish I had friends who talked mild shit like this! All my friends are nerds who take everything seriously. On the project, did you do anything about the time dimension? ChatGPT is strictly input -> output, but something like this needs time between messages to feel real (and not run constantly). I imagine adding "time since last message" to the training data + expected output would work.

To be completely realistic, the AI would also need the ability to leave you on read.

I have an ongoing chat in chatGPT where it's instructed to every once in a while ignore my question and just respond with "Shut up, nerd."

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#278

Post by Fauntleroy »

isaacremuant wrote: post #239
Fauntleroy wrote: post #89

Earlier quoted context omitted.

Compare the response of the Chinese government and the US government, then return here and tell us all about the "blindfold" we Americans are wearing wrt the government's COVID response.

I didn't even talk specifically about the US but that is just pathetic as a defense. Comparing with the bottom of the barrel to make yourself look good? That's like a country using the US healthcare situation to claim their own healthcare is good. It's a poor car salesman trick.

And you're out here making blanket statements suggesting "we" did not pay attention to our own country's piecemeal, half-assed, state-by-state COVID response, instead painting it as a brutal federal crackdown. Lord.

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#279

Post by kevincox »

overthrow wrote: post #47
sdwr wrote: post #17

Wish I had friends who talked mild shit like this! All my friends are nerds who take everything seriously. On the project, did you do anything about the time dimension? ChatGPT is strictly input -> output, but something like this needs time between messages to feel real (and not run constantly). I imagine adding "time since last message" to the training data + expected output would work.

To be completely realistic, the AI would also need the ability to leave you on read.

I was thinking it would be interesting to get the model to generate timestamps with the messages. Then you could actually queue the messages until that time. It would be like a real conversation.

Of course if you send a message before the AI does it re-runs and produces a new future message with a new timestamp.

Re: Replacing my best friends with an LLM trained on 500k group chat messages

#280

Post by kevincox »

Erwin wrote: post #234

Remember Replika and how people got quite attached to that chatbot? I imagine in the near future you'll be able to sell your and your friends' chat history to a company building a more advanced, realistic chatbot. Do you want to have a group of friends to hang out with? Buy an organically fabricated and pre-trained chatbot. Or maybe there's enough emptieness in your life that you go deep assume one of those friends'…

Yeah, this sounds like a great business opportunity.

1. Allow creating customized chat bots by uploading some conversations. Hope people find this fun and you go viral.

2. Sell the data to highest bidder.

3. Sell product placements.

Post reply on HN