@Arjunjain@mayukh_panja I had a boss one time that would, not infrequently, say “That’s not good enough.” He would also go through work product with me and others in excruciating detail - in sessions that lasted well past quitting time. I learned a lot from that guy.
@geoffreyhinton Not doing AI seems like a non-starter at this point. Regulation looks like capture by a few billionaires which is also demonstrably bad. I would rather we did much more work on alignment and alignment verification.
@ThomasTalhelm@MohammadAtari90@NatureHumBehav Yes. Did you find the viewer? I'm drafting a preregistration now to administer MFQ-2 and your Scenarios to the same set of models in the same window with and without framing. I'll send it to you if you're interested.
@ThomasTalhelm@MohammadAtari90@NatureHumBehav I misspoke. We compared human means from the Atari et al. 19 country study using English and their official translations to our framed and unframed model responses.
@ThomasTalhelm@MohammadAtari90@NatureHumBehav Will do. I just did an exploration where I administered MFQ-2 to 11 different models and compared them to the means from Sęker and Atari’s 14 country study.
https://t.co/IjmAl6ZMPU
I was thinking about how many major appliances and TVs I've bought over the years. The sort of things that my parents bought rarely. What would happen if a washing machine had to have full in-home warranty coverage for 10k loads? Or a BMW had to carry 15y/150k mile coverage?
@kimmonismus Labs need to decide what business they’re in and do that. Today in typical tech bro fashion they think they are smarter than everybody else and want to put their fingers in all the pies. Competing with your customers is a bad look and bad business.
@thsottiaux Earlier today I asked Astra to review a web based viewer against the associated paper and appendix. In a few minutes without an answer it came back saying my 5 hour limit was exhausted. 5 hours later same request and response within maybe 3 minutes. Grok did the thing what gives?