a16z Video Watch posted an update
John Schulman on Getting AI Models to Tell the Truth
Why it mattersJohn Schulman discusses an alignment approach he is bullish on: training objectives intended to make AI models report information they otherwise would not verbalize.
Discuss: How might training objectives that elicit what AI models would otherwise not verbalize improve alignment?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.