Experimental browser for the Atmosphere
The more my lab dissects large language models, the more I realize we have no unifying theory on how these models will behave. We study them like zoologists - poking and prodding and observing. When they get “updated”, there’s no way to predict changes in behavior other than using benchmarks.
Apr 6, 2025, 2:46 PM
{ "uri": "at://did:plc:mxty5du4uw2vxb5hyvshrfgv/app.bsky.feed.post/3lm5osjw5522o", "cid": "bafyreidpglaudz35hr5ckjn6ghzab7jvwqg4nvq6eqc3sdsxsedajyvqty", "value": { "text": "The more my lab dissects large language models, the more I realize we have no unifying theory on how these models will behave. We study them like zoologists - poking and prodding and observing. When they get “updated”, there’s no way to predict changes in behavior other than using benchmarks.", "$type": "app.bsky.feed.post", "langs": [ "en" ], "createdAt": "2025-04-06T14:46:57.656Z" } }