nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqm3fp0m7fjdsd58dhs39sfvz4d45zsxtp4c362avjw4a869kr453qx8g23r (nprofile…g23r) Actually, I think early-era Wikipedia offers an interesting analogy with current-era LLMs. There are parallels: back then, crowdsourced encyclopedias was the upstart paradigm challenging expertly curated encyclopedias, similar to how LLM-generated content challenges human-generated content today; and there were even similar concerns about students blindly using these new resources for their coursework. The skepticism concerns about reliability, and the visible failings, of current-era LLMs are echoed in similar concerns about crowdsourced content two decades ago. Nevertheless, with the passage of time, Wikipedia has thrived to be a usable and largely trustable resource (at least outside of highly politicized topics), and academia has adapted to incorporating Wikipedia into its workflows.
That said, the process of addressing quality concerns was far from automatic: most other wikis have proven far less successful and reliable than Wikipedia, even if they use identical software. In the case of Wikipedia, strongly enforced policies and a lively community of stakeholders played a major role in allowing it to improve over time, though not without its own internal controversies and drama. Perhaps one also needs similar "soft" infrastructure around the purely technical platforms of modern AI tools in order to guide it to a state similar to where current-era Wikipedia stands today.
Terence Tao /
npub1hs…m7r3k
2025-03-10 19:48:28
in reply to nevent1q…umxn
Author Public Key
npub1hsf727dlfy55vvm5wuqwyh457uwsc24pxn5f7vxnd4lpvv8phw3sjm7r3kPublished at
2025-03-10 19:48:28Event JSON
{
"id": "e5e4542f90e974a3ffc3bd8dd4235a857b7bc1fbe388bba7a96512150b0d66e2",
"pubkey": "bc13e579bf49294633747700e25eb4f71d0c2aa134e89f30d36d7e1630e1bba3",
"created_at": 1741636108,
"kind": 1,
"tags": [
[
"p",
"dc5217efc99360da1db7844b04b0556d68281961ae23a57592757a7d16c3ad22",
"wss://relay.mostr.pub"
],
[
"p",
"56ac9127866e574201fad1df3807f487841c07b7a9719b69fb4e45154514b86b",
"wss://relay.mostr.pub"
],
[
"e",
"fd6edb1ac309b603a8bf79380ff50381d3e849c7001df6ad32c8ca0da28f3570",
"wss://relay.mostr.pub",
"reply"
],
[
"proxy",
"https://mathstodon.xyz/users/tao/statuses/114139864015595828",
"activitypub"
]
],
"content": "nostr:nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqm3fp0m7fjdsd58dhs39sfvz4d45zsxtp4c362avjw4a869kr453qx8g23r Actually, I think early-era Wikipedia offers an interesting analogy with current-era LLMs. There are parallels: back then, crowdsourced encyclopedias was the upstart paradigm challenging expertly curated encyclopedias, similar to how LLM-generated content challenges human-generated content today; and there were even similar concerns about students blindly using these new resources for their coursework. The skepticism concerns about reliability, and the visible failings, of current-era LLMs are echoed in similar concerns about crowdsourced content two decades ago. Nevertheless, with the passage of time, Wikipedia has thrived to be a usable and largely trustable resource (at least outside of highly politicized topics), and academia has adapted to incorporating Wikipedia into its workflows.\n\nThat said, the process of addressing quality concerns was far from automatic: most other wikis have proven far less successful and reliable than Wikipedia, even if they use identical software. In the case of Wikipedia, strongly enforced policies and a lively community of stakeholders played a major role in allowing it to improve over time, though not without its own internal controversies and drama. Perhaps one also needs similar \"soft\" infrastructure around the purely technical platforms of modern AI tools in order to guide it to a state similar to where current-era Wikipedia stands today.",
"sig": "8eea7c6127b781b9a1277ab0eb6c2e830d50776f4a475a7a7353f5027612f2c17040c24598e590c58ad792e57c03be1c29bab965ed21956baa78872d0632c334"
}