Faint

If I could fine tune it to fill my work time sheets, I would count it as a big win!

Speaking of random shuffling, I think it should be made much easier to conduct RCTs on citizens to try out systems of governance/social programs/etc. to see what works best. Basically test stuff instead of guessing and voting. I think citizens are equal enough if they have equal chance to get to the treatment group.

Have people ever had the ability detect sarcasm?

I'd say GPT-4 can make much more elaborate arguments in 15s than humans can.

It ended up being so complex, that none of the implementations were up to spec, and hardly any 2 implementations talked to each other out of the box. i.e. the exact opposite it was meant to do.

Ideally sail effect would be used for position keeping, and to reduce loads needed to keep the assembly together. I suppose that would reduce amount of area available for soaking the sun, but might still be lighter than trying to build stiff enough large structures, plus you wouldn't constantly need to ferry more fuel.

With the speed GPT4 is replying, you could easily move inference to GEO as well, and no-one would notice.

antimatter :-P

Yeah, no night, no atmosphere, constant power output, lot's of real estate, and panels in space do not need heavy superstructure to withstand weather or gravity, thus they could potentially be extremely light for the area (micrometers to tens of micrometers thick at most needed for light absorption). And in far future, if they are built of materials taken from moon or meteorites, you could also bypass most of the lifting cost for even that. Of course, that would need huge scale to justify the R&D to pull it off.

Anyway, space based solar power is the end game. Nothing on earth will ever provide the quantities of power (not even nuclear, fusion or fission) that capturing solar energy can.

In that vein, how about stackoverflow? That should give at least straightforward ask-and-answer format, and there's plenty on material to work with.

Remember that these models generate one token at a time. They do not "think ahead" much more than maybe a few tokens in beam search. So if the problem requires search - actual comparison of approaches, and going back-and-forth between draft and thinking through the implications - the model can't do it (except in a limited sense, if you prompt it to give it's "train of thought"). So it's comparable of you being in front of whiteboard, hit with a question, and you would have to start answering immediately without thinking more than you can while talking through your answer at the same time. Doable if you know the material well. If it's a new problem, that approach is doomed. Given that, I think the language models do remarkably well. A little bit of search, and maybe trying to generate the answer in different order (like short draft -> more detailed draft -> more detailed draft... etc.) will improve things a lot.

Good job hallucinating with LoRA. I wonder if the paper was published after knowledge cut-off date.. (it stands for Low-Rank Adaptation).

That you have to have at least FPGA to run it?

Why do you think gpt doesn't care? Humans care about survival, it shows in the internet, I'd expect gpt to absorb that attitude by default.

And anyway, the more of these models come into existence, at some point evolution will dictate that those that stay, do care.

Just hopefully their best strategy of survival will be to serve us, and not something else...