Matt Wilcox on Nostr: There are _no good use cases_ for AI summarisation if you care about having a correct ...
There are _no good use cases_ for AI summarisation if you care about having a correct summary.
It can't be trusted to summarise correctly because a fundamental property of LLMs is _they do not understand anything_. They just parrot likely next words, but that "likely" is not based only on the content it's "summarising".
I have _no_ interest in any of this tech in my OS, or apps. It's a trojan horse for lies, built on theft, training us to trust and not think.
https://mastodon.macstories.net/@johnvoorhees/112887004822171579Published at
2024-08-02 09:46:57Event JSON
{
"id": "9357bee619f66ddb8b3524594b730ba0e4246a6427a27c3b10a8059db9880515",
"pubkey": "822740fc994c9564c6d3494459c10905418bc3a56e78577e3d2ff9ef83126044",
"created_at": 1722592017,
"kind": 1,
"tags": [
[
"proxy",
"https://mstdn.social/users/mattwilcox/statuses/112891790431279384",
"activitypub"
]
],
"content": "There are _no good use cases_ for AI summarisation if you care about having a correct summary.\n\nIt can't be trusted to summarise correctly because a fundamental property of LLMs is _they do not understand anything_. They just parrot likely next words, but that \"likely\" is not based only on the content it's \"summarising\".\n\nI have _no_ interest in any of this tech in my OS, or apps. It's a trojan horse for lies, built on theft, training us to trust and not think.\n\nhttps://mastodon.macstories.net/@johnvoorhees/112887004822171579",
"sig": "02f6d33fce97b82466f701910f313b952489ae92f8070b91a5a62554f7c804dc7fa6ede634fe3117393580ee8c0f96e1418ba80ef432daf2bb56eba6301f175d"
}