A Talk: How to Find Things Online
Metadata
Highlights
- But this is a shift you can see across more than just games communities. In the world of social media, there is a general shift away from public-focused platforms like Twitter and Instagram and towards smaller communities built on mutual trust. In many ways I think this is a shame - there’s a lot of power in having information somewhere where anyone can see it, where what you know isn’t predicated on who you know. When it comes to information about how to get work, how to access things important for people’s lives, making connections… (View Highlight)
- But here’s where I think things get a bit screwy - the question then becomes… why does someone write the walkthrough and publish it online? There’s already this shift from publishing information openly and without expecting reciprocity. And towards either a really corporate model, where you’re really trying to extract money from it. Or towards a much smaller group, where you’re talking within small groups of trusted friends. And neither of these groups has much incentive to write up stuff that can train AI models.
For the groups of friends on Discord, or even studios running community groups… these communities exist because they are in relative privacy. They don’t want their chats to be public - in a search engine or inside a large language model. And there’s a tension here where the AI companies would love to get their hands on this data, all this rich data about how people actually talk that would make their robots work so much better. (View Highlight)
- But of course, the flipside to this is that we lose much of the sense of the “commons” that has characterised the Internet so far. You lose a lot of the serendipity that comes from logging on and suddenly talking to someone in another country, who maybe shares an interest in adventure games with you but is otherwise quite different. And once people are in smaller groups, then in-group norms can shift and become more accentuated from each other. If these are norms that seem kind of harmless then this is called a filter bubble and journalists wring their hands in The Atlantic about how it’s happening to them. And if these are norms that seem kind of racialised or scary then it’s called radicalisation, and journalists wring their hands in The Atlantic about how it’s happening to other people. (View Highlight)
- And there’s a lot more to talk about with AI, and the weird things going on in that space. Even just with LLMs, there’s stuff about fine tuning for different tasks, the explosion of open source models, prompt injection attacks and how impossible they seem to be to mitigate… And outside LLMs there are obviously models that make pictures, understand video, do all kinds of different things. But I still I always return to the same question with AI - where does the training data come from? And, if this AI thing catches on, where will it come from, in the future? (View Highlight)
title: “A Talk: How to Find Things Online”
author: “v21.io”
url: ”https://v21.io/blog/how-to-find-things-online”
date: 2023-12-19
source: reader
tags: media/articles
A Talk: How to Find Things Online
Metadata
Highlights
- But this is a shift you can see across more than just games communities. In the world of social media, there is a general shift away from public-focused platforms like Twitter and Instagram and towards smaller communities built on mutual trust. In many ways I think this is a shame - there’s a lot of power in having information somewhere where anyone can see it, where what you know isn’t predicated on who you know. When it comes to information about how to get work, how to access things important for people’s lives, making connections… (View Highlight)
- But here’s where I think things get a bit screwy - the question then becomes… why does someone write the walkthrough and publish it online? There’s already this shift from publishing information openly and without expecting reciprocity. And towards either a really corporate model, where you’re really trying to extract money from it. Or towards a much smaller group, where you’re talking within small groups of trusted friends. And neither of these groups has much incentive to write up stuff that can train AI models.
For the groups of friends on Discord, or even studios running community groups… these communities exist because they are in relative privacy. They don’t want their chats to be public - in a search engine or inside a large language model. And there’s a tension here where the AI companies would love to get their hands on this data, all this rich data about how people actually talk that would make their robots work so much better. (View Highlight)
- But of course, the flipside to this is that we lose much of the sense of the “commons” that has characterised the Internet so far. You lose a lot of the serendipity that comes from logging on and suddenly talking to someone in another country, who maybe shares an interest in adventure games with you but is otherwise quite different. And once people are in smaller groups, then in-group norms can shift and become more accentuated from each other. If these are norms that seem kind of harmless then this is called a filter bubble and journalists wring their hands in The Atlantic about how it’s happening to them. And if these are norms that seem kind of racialised or scary then it’s called radicalisation, and journalists wring their hands in The Atlantic about how it’s happening to other people. (View Highlight)
- And there’s a lot more to talk about with AI, and the weird things going on in that space. Even just with LLMs, there’s stuff about fine tuning for different tasks, the explosion of open source models, prompt injection attacks and how impossible they seem to be to mitigate… And outside LLMs there are obviously models that make pictures, understand video, do all kinds of different things. But I still I always return to the same question with AI - where does the training data come from? And, if this AI thing catches on, where will it come from, in the future? (View Highlight)