3 ms·

Wouldn't that depend on who programs/configures/"tweaks" the AI? Haven't we seen crazy results, and after tweaking, "bespoke" (as in "This is what I would say"

by k310 1y ago

Wouldn't that depend on who programs/configures/"tweaks" the AI? Haven't we seen crazy results, and after tweaking, "bespoke" (as in "This is what I would say") results. It's software .... you know the rest. Asking sincerely. And what happens when you remove human weaknesses, such as love of Smoot-Hawley and days gone by, gunboat diplomacy (literally), greed, etc? If it reduces creativity, etc, won't we have to argue with it (Landru)[0] to moderate its power or self-destruct? [0] https://en.wikipedia.org/wiki/The_Return_of_the_Archons https://en.wikipedia.org/wiki/The_Return_of_the_Archons

A_D_E_P_T 1y ago
It would probably have to be an open-source model (e.g. DeepSeek R2) with open weights and a training regimen that's documented and open to outside observers. I'd also note that Elon Musk has tried everything to make Grok a "based" AI, and it just doesn't work. Any model that's sufficiently large is resistant to that sort of thing. I don't know what the threshold is, but Grok 3 is very large, at 2.7T parameters. (https://arxiv.org/html/2502.16428v1 https://arxiv.org/html/2502.16428v1) So: A large (>2.7T parameter), open source, open weight model. > If it reduces creativity, etc, won't we have to argue with it (Landru)[0] to moderate its power or self-destruct? I don't understand how a President that is incorruptible, ascetic, and impartial would reduce human creativity? If anything, it would probably have a bunch of good ideas for USPTO to implement.
- k310 1y ago
  I got carried away with the Star Trek analogy. There, the computer controlled too much, but aren't we giving up too much power and decision-making to AI already? Hiring, medical care decisions? And Landru could be convinced that it violated its directive, so it blew itself up. We already see self-preservation showing up in models,and more sinister "thoughts". For every ethical programmer, there are "x" who aren't.

by k310 1y ago

A_D_E_P_T 1y ago

It would probably have to be an open-source model (e.g. DeepSeek R2) with open weights and a training regimen that's documented and open to outside observers. I'd also note that Elon Musk has tried everything to make Grok a "based" AI, and it just doesn't work. Any model that's sufficiently large is resistant to that sort of thing. I don't know what the threshold is, but Grok 3 is very large, at 2.7T parameters. (https://arxiv.org/html/2502.16428v1 https://arxiv.org/html/2502.16428v1) So: A large (>2.7T parameter), open source, open weight model. > If it reduces creativity, etc, won't we have to argue with it (Landru)[0] to moderate its power or self-destruct? I don't understand how a President that is incorruptible, ascetic, and impartial would reduce human creativity? If anything, it would probably have a bunch of good ideas for USPTO to implement.

k310 1y ago
I got carried away with the Star Trek analogy. There, the computer controlled too much, but aren't we giving up too much power and decision-making to AI already? Hiring, medical care decisions? And Landru could be convinced that it violated its directive, so it blew itself up. We already see self-preservation showing up in models,and more sinister "thoughts". For every ethical programmer, there are "x" who aren't.