Experimental browser for the Atmosphere
Loading post...
{ "uri": "at://did:plc:l2xlpoj2k5yvs4a4f6pq76iy/app.bsky.feed.like/3lp65py7frq2h", "cid": "bafyreiane6ygu4iswis66ip7am7qcver2wfbzjwwfyl27bk733qb4yrs4i", "value": { "$type": "app.bsky.feed.like", "subject": { "cid": "bafyreiavhkpszy4i4qnobqjsx42zvit7eih5zpntni35zkwlblxzfxu5ki", "uri": "at://did:plc:sefgphqp2xqwh2hawaixykwz/app.bsky.feed.post/3lp64xo4xks27" }, "createdAt": "2025-05-14T23:29:32.821Z" } }
If you resort to hidden prompt based manipulation, it will show up at some point to user queries. And the amount of data needed to skew training sets to a preferred narrative is absolutely gargantuan. Synthetic training data may be used but at the risk of breaking models further.
May 14, 2025, 11:15 PM