One small quibble, you can actually live much more cheaply on rice. A pound of dry rice contains 1600 calories, if you eat 2000 calories a day, you need 5 pounds every 4 days, so a 50 pound bag will last 40 days, meaning you need 9 per year. This has a total cost of $450 at your price. Probably less if you shop around or buy in bulk.
jeff8765
I think the issue here is that the tasks in question don’t fully capture everything we care about in terms of language facility. I think this is largely because even very low probabilities of catastrophic actions can preclude deployment in an economically useful way.
For example, a prime use of a language model would be to replace customer service representative. However, if there is even a one in a million chance that your model will start cursing out a customer, offer a customer a million dollars to remedy an error, or start spewing racial epithets, the model cannot be usefully deployed in such a fashion. None of the metrics in the paper can guarantee, or even suggest, that level of consistency.
Likely higher than one in a million, but they can be fired after a failure to allow the company to save face. Harder to do that with a $50M language model.
Yeah, but Putin’s been president of Russia for over 20 years and already has a very large, loyal following. There will always be those that enthusiastically follow the party line of the leader. It’s somewhat harder to actually seize power. (None of this is to excuse the actions of Putin or those who support him.)
But this doesn’t solve the problem of angry customers and media the way firing a misbehaving employee would. Though I suppose this is more an issue of friction/aversion to change than an actual capabilities issue.
For me the answer is yes, but my situation is quite non-central. I got into MIT since I was a kid from a small rural town with really good grades, really good test scores, and was on a bunch of sports teams. Because I was from a small rural town and was pretty smart, none of this required special effort other than being on sports teams (note: being on the teams required no special skill as everyone who tried out made the team given small class size). The above was enough to get me an admission probably for reasons of diversity I’m a white man but I’m fairly certain I got a bonus to my application for being from a small rural town.
Counterfactuals are hard, but going to MIT probably helped me to get into a prestigious medical school, leading to my current position as a doctor at a prestigious hospital. People at least pretend to be impressed when somewhat tells them that I went to MIT, despite my undergraduate field of study having absolutely nothing to do with my current job. Since I was lucky enough to be able to attend the university by doing the things I would have done anyway, I’d certainly say the effort was worth it.
In my particular case it wasn’t really all that hard. I went to an extremely small school so classes weren’t tracked the way they might be at a larger school. Since I was much better at taking tests than my peers I didn’t really have to study to get A’s on tests. We didn’t even have all that much homework, though I guess it probably was hundreds of hours over the course of my high school career. I would have had to do that regardless though.
It would cause a severe heat dissipation problem. All that energy is going to be radiated as waste heat and, in equilibrium, will be radiated as fast as it comes in. The temperature required to radiate at the requisite power level would be in excess of the temperature at the surface of the sun, any harvesting machinery on the surface of the planet would melt unless it is built from something unknown to modern chemistry.
I’ve found it’s ability to be much better as well. In contrast to GPT-3, which often seemed to be unable to keep track of board state and made illegal moves toward the end of the game, it not only played legal moves, it actually mated me. Granted I’m a terrible player and I was deliberately not reading ahead to see if it would be able to mate a weak player. My method was to tell it I wanted to play and then give my move in algebraic notation. It would respond with a move, then I would respond with another. After it beat me, I asked it to list all the moves we had played. There were some minor errors in its notation (I think) which I’ve noted below.
Of course! Here’s a summary of our game in standard algebraic notation:
d4 Nf6
c3 d5
f3 c5
dxc5 e6
Bf4 Bxc5
Nd2 Nc6
e4 dxe4
fxe4 e5
Bg5 Be7
Bb5 O-O
Nf3 Qc7
O-O Bg4
Nc4 Rad8
Qe1 a6
Bxc6 bxc6
Rd1 Rxd1
Qxd1 Nxe4
Bxe7 Qxe7
Nfxe5 Qc5+
Rf2 Nxf2
Qxf1 Nd3+ (Here GPT-4 made a slight error, I played Qf1 but there was no capture)
Kh1 Nxe5
Nxe5 Qxe5
h4 Rd8
Qxa6 Rd1+
Qxf1 Rxf1+ (Here it checkmates me, though doesn’t mark the move as such)
Notably, if you tell it to think step by step it gets the question right and answers with:
This problem is known as the Monty Hall problem. In the original problem, the car is placed randomly behind one of the three doors, and the host always opens a door with a goat after you make your initial choice. However, in your variation, the car is always behind door number 1.
Let’s analyze your version step by step:
You initially choose door No. 1, which always has the car behind it.
The host, knowing what’s behind the doors, will always open one of the other two doors, revealing a goat. In this case, he opens door No. 3.
Now you are given the option to switch your choice to door No. 2.
In this version of the problem, it is NOT to your advantage to switch your choice, because the car is always behind door No. 1, and you have already chosen that door. If you switch to door No. 2, you will end up with a goat instead of the car.
This is honestly pretty similar to how humans behave most of the time. They pattern match and don’t really pay close attention. However, if you give cues that something is actually important, for example by putting them in an exam room and telling them a test will be graded, they can do somewhat better. Telling GPT-4 to think step by step does something similar.
It also seems a patent was filed for this material in 2021 and was granted earlier this year prior to publication.
I think you added an extra three zeros during your total year calculations. you list 2.23E15 as the total number of years experienced, but multiplying the total time of 5E4 by the current population of 8E9 gives a total of only 4E14 experience years. The true number must be quite a bit lower as the human population was quite a bit lower than 8 billion for most of that time. This also affects the proportion of experience years which have occurred in living memory. My guess is 20% have occurred since the birth of Kane Tanaka and 10% experienced by living people. This also squares pretty well with your figure of 50% of human experience occurring since 1300. It doesn’t really make since for 50% of experience to have occurred since 1300, but only 0.02% since 1903.