Why Banning ChatGPT is Ridiculous

[@Flat0Line1]

But risk of leakage not in LLM itself, but in service - which saving your queries, responses, metadata of dialogues etc. You can provide some nda-info, which will be stored - that you can see it in chat with bot, and that's just simple db which stores your queries and responses. And it can be a risk with inside job in company which provide you these services

[@ichseseien5194]

Why would it be ok to give openai my secrets in the first place?

[@d-squared70]

It wouldn't, but accidents happen. Especially, when you're securing large corporations with lots of employees. 😉

[@cmfjr0]

thank you for taking the time to put this together. it's clear that the productivity gains from using LLMs can be significant, but doesn't that value hinge on the dataset it references? i think organizations make LLMs most valuable by putting their data into the database it defers to. do you agree? this unfortunately increases the likelihood of the real risk you highlight.

[@d-squared70]

Indeed it does increase the risk, but there are tactics one can take to reduce that risk. Maybe a future video on RAG defenses. 😄

[@d-squared70]

Recent developements from Deepmind in relation to LLMs memorizaing and regurgitating training data - not-just-memorization.github.io/extracting-training-data-from-chatgpt.html The risk still seems overblown, especially if coroporates are preventing their employees from using this tools. 🤔

[@Michael-pw6qk]

Tell that to the various Board of Directors. They're the technophobes not the rank and file programmers.

[@MichaelHoughton_]

This is a very good video. Thanks 🙏🏻

[@briansmith2160]

This video was sponsored by: CHATGPT

[@d-squared70]

🤣 One can wish! LOL. You can easily substitute ChatGPT for Claude, Bard, PerplexityAI, etc.

[@samuelmcmurray3502]

cache... they have your conversations bro

[@almightyzentaco]

I don't believe any chat data is used to train the model. There should be zero connection to the next model of GPT and the data inputs from any chats. The real danger is someone intercepting packets between the opening server and your enterprise servers if you are sharing sensitive information.

[@d-squared70]

This interesting prespective. You seem pretty confident here. Can you share some supporting details as to why there's no connection between user input and training data for future models?

[@samgoodwin89]

⁠@@d-squared70that’s a nice way of saying this person is totally wrong. lol which they are. They do train on chat data and have admitted it. They don’t train on the api calls or ChatGPT enterprise, but they do use ordinary logs to distill gpt4 turbo from gpt4. Companies that are worried should buy GPT enterprise. Simple. Worrying about packets being intercepted is ridiculous. That’s true of any network request. HTTPS solves this.

[@danielmarco7863]

The loom filler word filter is so distracting

[@blastygamez]

Wait it’s getting banned? Let’s gooooo now u can’t be lazy anymore

[@WhisperingUniverse]

Yeah, we'd better throw out our computers and heavy machinery to be even less lazy! ... Desirable difficulty for learning can still be achieved with the use of AGI.

[@blastygamez]

@@WhisperingUniverse people have to stop somewhere and this is the good end point, Computers are really useful tools and their designed to help, but this goes a little bit too far, people can cheat even more and easier now, it’s also probably going to replace high level jobs which are full of creativity, be careful on who’s side ur on

[@diedforurwins]

people have to stop somewhere? ok you stop and I PROMISE I will too (not)

[@blastygamez]

@@diedforurwins I’m not here developing technology to bring higher jobs in danger, filthy

[@diedforurwins]

@@blastygamez do you understand game theory? I actually think we should have stopped a long time ago, but the fact is if we stop and the rest of the world continues advancing technologically, we will be part of a chinas dystopia instead of our own.