Saterious

[Saterious @saterious]

@ChrSzegedy @tallinzen @ChrSzegedy thank you for continuing to point this out. Tao and most of the math community have largely misunderstood the trajectory of AI even after it should have been obvious

[Saterious @saterious]

@togelius @jpschroeder Yes! Post scarcity is achievable

[Saterious @saterious]

@TaliaRinger Genuine question: what would have to happen for you to think this is a topic worth taking seriously?

[Saterious @saterious]

@Grady_Booch @__nmca__ That original assertion might be true for early models and pretrained checkpoints, but I have yet to hear any response to how that holds for current models with extensive RL training. The models learn from their own experience, they don’t merely parrot human text.

[Saterious @saterious]

@togelius Nobody said GPT-3 was the machine god, and they’re right, ASI is coming.

The stochastic parrot crowd was wrong from the beginning that LLMs could merely imitate human text. Certainly not true once RL training kicked in. This was obvious to anyone paying attention.

[Saterious @saterious]

@TaliaRinger What event would have to happen for you to update from believing superintelligent AI causing human extinction is pure sci-fi fantasy to it being even somewhat plausible?

[Saterious @saterious]

@npparikh Despite Sotomayor’s claim, it’s not at all clear that’s what the decision actually says.

[Saterious @saterious]

@ChrSzegedy @GaryMarcus @f_charton @jessemhan Would be amazing if you shared that short list. Those two pointers are already very helpful.

[Saterious @saterious]

@tszzl

[Saterious @saterious]

@TobyPhln Often it’s memes

[Saterious @saterious]

@TobyPhln The cure is to listen any Cziffra rendition of a Liszt piece, upon which one immediately realizes there are levels one didn’t even know exists. Personal fav is Totentanz https://t.co/asZHopNZn4

[Saterious @saterious]

@ewlamy @profoundlyyyy @timnitGebru I just checked DAIR’s website and didn’t see anything that falls into this category. Do you have any pointers? Thanks!

[Saterious @saterious]

@WickedViper23 @evanjconrad Highly skeptical of this claim. What’s going to replace transformers in 11 months?

[Saterious @saterious]

@PaulTopping @lateinteraction @GaryMarcus @YiTayML This is not wishful thinking, it’s been empirically verified. (Note I’m not claiming scale alone solves AGI, only that emergent abilities are real.) https://t.co/Gp2sdHO47Q

[Saterious @saterious]

@americancyborg @egrefen @timnitGebru @emilymbender Saying “I’m a Wittgensteinian” is not an appeal to authority, it merely denotes a set of beliefs one holds. Saying “belief X is true because Wittgenstein said it” would be an appeal to authority.

[Saterious @saterious]

@davidchalmers42 @ylecun @TaliaRinger Thanks for the further clarification and thanks for your insights on this topic!

[Saterious @saterious]

@ylecun @TaliaRinger @davidchalmers42 This misrepresents his conclusion. He said there’s “not remotely conclusive evidence,” but that their abilities “give at least limited prima facie evidence.” He also said the arguments against LLM sentience are not very strong, and we should remain open minded.

[Saterious @saterious]

@fchollet For the interested reader, are there any other resources justifying this stance beyond your essay https://t.co/yf4EECt3SQ? Thanks

[Saterious @saterious]

@fchollet This statement is every bit as religious and dogmatic as the beliefs you ascribe to proponents of AGI

[Saterious @saterious]

@sethlazar This assumes alignment is impossible (or nearly so). Tweet 2 says we can’t constrain misaligned AGI (agreed), but if one believes we can create aligned AGI with high probability, then there are plenty of arguments why we have a moral obligation to build it.