Well, as far as I know it nobody has done that yet and current LLMs seem to focus more on general applications than on being efficient for specialized use cases like this.
Well, as far as I know it nobody has done that yet and current LLMs seem to focus more on general applications than on being efficient for specialized use cases like this.
Mainly because most social media isn’t really very well made for the purpose of making sure you have seen every post (anything with upvotes/downvotes) or limit the content of a post (microblog-style social media, video/image focused social media).
You can literally store all preferences in cookies without a problem with EU legislation.
I am not even necessarily talking about relevant/spam. Some content might just naturally lose out because e.g. an interesting mathematical proof has less mass appeal than a cute cat picture even though the former might be higher quality and effort.
I think one reason that is often overlooked in that polarization is that there are a lot fewer places in RL too where you have to interact with others in a more than superficial way or even can do so if you actively choose to do so.
A big reason for that is cars which isolate people as they travel from place to place and another is the reshaping of our cities into large homogeneous parts, each just for a group of people with similar income and culture.
I wonder if that is one of the areas where AI might be useful in the future. LLMs could potentially be useful to identify non-trivial statements that are not just a rephrased version of statements that have already been made in other comments.
I wouldn’t say that upvotes always mean quality, they could also just indicate mass appeal while quality but niche content is hidden that way.
I think you are not seeing the whole scope of the problem. Echo chambers are only one of the problems, lowest common denominator posts are another issue of self-moderation/voting.
As someone who has lived through some of the time with newspapers and without social media, no, quality was pretty bad back then too.
%25 is the URL encoding for 0x25 (or 37 decimal), the ASCII code for the percent sign. Basically it seems to recognize that it is a URL and then URL-encode characters that are not allowed in URLs
A0 is also exactly 1 square meter which makes it easy to figure out the number of pages of e.g. A4 from weight if you know what kind of gramm per square meter paper you are dealing with (which is why that is often part of the packaging).
The aspect ratio of the two sides is the square root of 2 since you otherwise couldn’t divide it again and again and get sizes with the same aspect ratio.
Ah, I remember those days, back when sci-fi movies had fancy notions such as multi-pass.
I am not saying they can’t post links to the posts to all social media platforms, just that the actual post should be on some regular website.
Honestly, Mastodon is better than Twitter of course but I would still prefer them to post official stuff on a website that isn’t social media at all.
I feel this is missing the methodology part where they describe how they evaluated a model that literally can’t do anything but read input and write output for actions like “copy itself over a newer version” or “its own goals”.
The thing about advertising is that pretty much every bit of data that could tell us if it works or not is produced entirely by the same industry that has an overwhelming incentive to tell us that it does work.
I am not even saying that it never works, more that advertising’s primary purpose is to sell advertising to companies and if that advertising works or not is a distant secondary goal at best as long as those companies have no way to prove it doesn’t.
It very likely works on some types of products while being completely useless on others. And even where it works it very likely has a distribution from extremely bad ads for a given product or service to very effective ones for the same product.
Imagine nonsense content on Instagram…
Isn’t that all Instagram ever was? Out of all the social media platforms it was the one that was essentially fake from the start, just fakes produced by the users.
Well, technically 100% employment wouldn’t be desirable, that would mean nobody can ever switch jobs or take some time off from working to deal with some personal issues or projects.
Isn’t that essentially how that whole scam called advertising works in general?
Then they shouldn’t link to the social media post from elsewhere though as described in the article.