Experimental browser for the Atmosphere
If we build that it will absolutely try to kill us arxiv.org/abs/2502.17424
May 8, 2025, 2:28 PM
{ "uri": "at://did:plc:kft6lu4trxowqmter2b6vg6z/app.bsky.feed.post/3loo4pl55ik2j", "cid": "bafyreiehduhpgxb5tm5yfckjrg6ypoyvv4idot25e6gp5uns77ngvskfda", "value": { "text": "If we build that it will absolutely try to kill us arxiv.org/abs/2502.17424", "$type": "app.bsky.feed.post", "embed": { "$type": "app.bsky.embed.external", "external": { "uri": "https://arxiv.org/abs/2502.17424", "thumb": { "$type": "blob", "ref": { "$link": "bafkreibmwume2q2jvfyf5jvz62lbuyuyxmwh2ewnlggfcjyro263dps63m" }, "mimeType": "image/jpeg", "size": 103016 }, "title": "Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs", "description": "We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on..." } }, "langs": [ "en" ], "reply": { "root": { "cid": "bafyreifk4k3lswvcv3dulpxv4tav72yg2mp6lpqbhu67ggcstdkptrbgty", "uri": "at://did:plc:565ebob5f6hw33hjdkxty6qj/app.bsky.feed.post/3loo2clroy22n" }, "parent": { "cid": "bafyreifk4k3lswvcv3dulpxv4tav72yg2mp6lpqbhu67ggcstdkptrbgty", "uri": "at://did:plc:565ebob5f6hw33hjdkxty6qj/app.bsky.feed.post/3loo2clroy22n" } }, "facets": [ { "index": { "byteEnd": 75, "byteStart": 51 }, "features": [ { "uri": "https://arxiv.org/abs/2502.17424", "$type": "app.bsky.richtext.facet#link" } ] } ], "createdAt": "2025-05-08T14:28:49.730Z" } }