All the ai companies have trained on the FBI cp database but they use a coefficient of -1 thus steering it away from such content. Reinforcing against such material. Openai used to pay people for cp Idk if they still do this tho.
Those files are kinda a nightmare to navigate in their bare state. And the datasets are huge. I doubt anyone training AI would allow them to go through knowingly, less it was specifically a police invesigation and case law focused AI that was designed to process and categorize that kind of data.
Most AI are designed for functional discussion and factual data processing. It's not a great idea to just feed in random trash.
I had to use cloudflare to stop AI crawlers from using like 60% of my 16 core server that runs this instance. They were spending that much time pulling fediverse content, multiple bots without and wait time between requests. You really think they'd reject epstein files but seek out our combined output?
And also everyone's vague notes about, like, the Sword of Truth mass isekai hatefic they wanted to write back in 2024 but then gave up on because they mentioned the Battle of Cable Street and then had to stare at a wall for a bit and walk away in shame
14 Comments
hexagonwin@lemmy.today · 46 pts · 126d
highly doubt it, they're trained on publicly available data mostly
ramble81@lemmy.zip · 30 pts · 126d
If by “publicly” you mean “any data source that it managed to connect to, public or private”, then yes….
youcantreadthis@quokk.au · 0 pts · 126d
Not entirely though.
Like we know grok was trained on the fbi's cp stash, right?
Goodlucksil@lemmy.dbzer0.com · 15 pts · 126d
Do you have any source on that?
youcantreadthis@quokk.au · 0 pts · 126d
Dont recall where i heard it.
muntedcrocodile@hilariouschaos.com · 1 pts · 125d
All the ai companies have trained on the FBI cp database but they use a coefficient of -1 thus steering it away from such content. Reinforcing against such material. Openai used to pay people for cp Idk if they still do this tho.
youcantreadthis@quokk.au · 1 pts · 124d
So why all the cp from them?
exaybachae@startrek.website · 20 pts · 126d
Those files are kinda a nightmare to navigate in their bare state. And the datasets are huge. I doubt anyone training AI would allow them to go through knowingly, less it was specifically a police invesigation and case law focused AI that was designed to process and categorize that kind of data.
Most AI are designed for functional discussion and factual data processing. It's not a great idea to just feed in random trash.
r00ty@kbin.life · 26 pts · 126d
I had to use cloudflare to stop AI crawlers from using like 60% of my 16 core server that runs this instance. They were spending that much time pulling fediverse content, multiple bots without and wait time between requests. You really think they'd reject epstein files but seek out our combined output?
degenerate_neutron_matter@fedia.io · 5 pts · 126d
youcantreadthis@quokk.au · 3 pts · 126d
Which would stop them.
testaccount789@sh.itjust.works · 8 pts · 126d
anon_8675309@lemmy.world · 7 pts · 126d
Well know if they start using redactions as much as em dashes
theywilleatthestars@lemmy.world · 4 pts · 126d
And also everyone's vague notes about, like, the Sword of Truth mass isekai hatefic they wanted to write back in 2024 but then gave up on because they mentioned the Battle of Cable Street and then had to stare at a wall for a bit and walk away in shame
drsaxoncrawfish@lemmy.today · 1 pts · 124d
██████ ███ █████ █████ ██ ██████ ██ ████ ███ ████ ████ ██ ████████ ██████ ███ █ ███ █████ ████ Bill Clinton ██████ ███ █████ █████ ██ ██████ ██ ████ ███ ████ ████ ██ ████████ ██████ ███ █ ███ █████ ████