Pretty good breakdown of OpenAI o3's (quite good) performance at guessing locations based on photos, by way of the "GeoGuessr" game (metadata stripped). It scored better than a very highly ranked human.
A tendency of some AI critiques is to treat _all_ such evidence as "smoke and mirrors". That is unhelpful; it occludes awareness of capabilities that are actually improving (and can present real #opsec and surveillance risks in this case).
https://sampatt.com/blog/2025-04-28-can-o3-beat-a-geoguessr-master
Erik Moeller
npub1qm…gl0lg
2025-04-29 05:29:12 UTC
Author Public Key
npub1qm9ke6lksjn8ushquf5hule7zydxnrkr0gcl5k856lcv5zw5032sygl0lgSeen on
wss://relay.momostr.pinkPublished at
2025-04-29 05:29:12 UTCEvent JSON
{
"id": "d8171319bf10de16bf4dbfb19ab4304d4afdfea8bd774ee735bff499a24514ef",
"pubkey": "06cb6cebf684a67e42e0e2697e7f3e111a698ec37a31fa58f4d7f0ca09d47c55",
"created_at": 1745904552,
"kind": 1,
"tags": [
[
"proxy",
"https://social.coop/@eloquence/114419600720845360",
"web"
],
[
"t",
"opsec"
],
[
"proxy",
"https://social.coop/users/eloquence/statuses/114419600720845360",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://social.coop/users/eloquence/statuses/114419600720845360",
"pink.momostr"
],
[
"-"
]
],
"content": "Pretty good breakdown of OpenAI o3's (quite good) performance at guessing locations based on photos, by way of the \"GeoGuessr\" game (metadata stripped). It scored better than a very highly ranked human.\n\nA tendency of some AI critiques is to treat _all_ such evidence as \"smoke and mirrors\". That is unhelpful; it occludes awareness of capabilities that are actually improving (and can present real #opsec and surveillance risks in this case).\n\nhttps://sampatt.com/blog/2025-04-28-can-o3-beat-a-geoguessr-master",
"sig": "2d56ebe8a23d8b6609e5a7537fc363e98f7e8e09315e9047b054aacaf7cf6712e8177c5cb3d9310249116e113f91ebda0e404e1ed190ddd65a303f08bf77042d"
}