Garbage in garbage out. We are falling back to the Yahoo categories curated view of the internet.
Google was really interesting when using organic link strength to indicate human curation. A web of trust if you will. But it’s been gamified to death at this point. LLMs have no capacity to reason, they just generate probable text.
Without a strongly curated training set it’s just going to become random noise… As we are seeing now.
Not only that, but trust from a self contained community is not the same as safe for the general public outside of context. Imagine asking for a summary of the Gamestop shortsqueeze and getting an answer from Superstonks.
They may also have missed the part where unlocking past post for rewrite allowed APIcalipse refugees use shredder bots to destroy a good part of the comments that actually held any value, replace a bunch of others with nonsense, and kicking out invested human curators (mods) in favor of generic button pushers (“power mods”) opened the floodgates to random BS, propaganda, and misinformation.
It has always been only a signal for popularity. It’s the same on every other site with similar systems, including ones that have existed long before reddit. Even in the early 2000s, there were endless reminders that the upvote button was not meant as an “I agree” button, but that’s not how it’s being used.
The scraped AI generated SEO’d to hell tech sites that push their bullshit registry/driver/ram software overtook anything actually useful. It’s a sad state out there.
Garbage in garbage out. We are falling back to the Yahoo categories curated view of the internet.
Google was really interesting when using organic link strength to indicate human curation. A web of trust if you will. But it’s been gamified to death at this point. LLMs have no capacity to reason, they just generate probable text.
Without a strongly curated training set it’s just going to become random noise… As we are seeing now.
Given the wave of Yahoo and MSN repost news articles, where they’re word for word the same as the original source, this is far too true.
I wonder if that’s why Reddit data was chosen. Upvotes could be used as a signal for trust. What they forget is jokes comments often get upvotes.
Not only that, but trust from a self contained community is not the same as safe for the general public outside of context. Imagine asking for a summary of the Gamestop shortsqueeze and getting an answer from Superstonks.
They may also have missed the part where unlocking past post for rewrite allowed APIcalipse refugees use shredder bots to destroy a good part of the comments that actually held any value, replace a bunch of others with nonsense, and kicking out invested human curators (mods) in favor of generic button pushers (“power mods”) opened the floodgates to random BS, propaganda, and misinformation.
It has always been only a signal for popularity. It’s the same on every other site with similar systems, including ones that have existed long before reddit. Even in the early 2000s, there were endless reminders that the upvote button was not meant as an “I agree” button, but that’s not how it’s being used.
Same as the link trust they used to use as a signal for trust. Again it was about popularity.
Trust here does not mean trustworthy, but rather higher likelihood to be answer that is sought. Calling it trust or authority as just marketing.
The scraped AI generated SEO’d to hell tech sites that push their bullshit registry/driver/ram software overtook anything actually useful. It’s a sad state out there.