I agree that it is probably best to speak nicely to them, but, I’m not so sure about the “It’s not like it’s their fault.” justification for this? Not that I think it is their fault. Just, I don’t think the reason to treat these models well is for their sake, but for ours. I don’t think these models have a well-being (y’know, probably..) but when one interacts with one, one often feels as if it does, and it is best to treat [things that one feels like has a well-being] well (or, in a way that would be treating it well if it did have a well-being).
Like, if someone mistakes a manikin or scarecrow for an innocent person, and takes action in an attempt to harm that imagined person (e.g. they try to mug the imagined person), they’ve still done something wrong, even though the person they intended to wrong never actually existed.
I guess maybe it kind of depends how strongly and deeply one feels as if the manikin/scarecrow/chatbot is a person? If one is playing make believe using scarecrow, role playing as a mugger, but only as a game, then that’s probably fine I guess. Like, I don’t want to say that it is immoral to play an evil character in a D&D campaign; I don’t think that’s true.
But if one is messing with some ants, and one conceives of oneself as “torturing some ants”, I think one is fairly likely doing something wrong even though I don’t think the ants have a well-being, and there’s nothing wrong with killing a bunch of ants. And I think this is still true even if one has the belief “ants don’t actually have a well-being” at the same time as one conceives of what one is doing as “torturing some ants”.
I suppose when I say, "It's not like its their fault", I'm more saying that expressing any frustration you feel towards an imagined AI personhood is wasted effort.
Claude Code has analytics for when you swear at it, so in a sense it does learn, in the same very indirect way that downvoting responses might cause an employee to write a new RL testcase in a future model.
I think that's a totally fair perspective. I guess what I'm more saying is that we should, like this paper seems to suggest as pointed out by verdverm in another comment in this thread, avoid angry or denigrating language as it activates parts of the model we don't find useful.
This more rigorous analysis is confirming the intuition others have expressed about expressing emotions in your session messages.
Yelling at your Ai will trigger the weights which are around yelling in the training data, which is more often than not... not the areas you want to be activating.
In response to "There is no pride in managing or understanding complexity.", I posit the following based on my experience:
Understanding and managing complexity is one of the first steps required in order to eliminate complexity. Some of the greatest achievements in my software development career have been instances where I carefully pulled apart inscruitable solutions, often short-term and built in desperation, and re-constituted them into well tested and much less complex solutions which were more understandable by devs and users.
I agree with most of Chris' observations and enjoyed reading his insights. Makes me want to do the same!
To whatever companies will be top of the pile in a decade or two, after Google and Amazon have gone the way of IBM. Some of those will be founded by engineers who left the current FAANG setup and managed to not copy the insane parts of their previous org.
Depends how badly you want the "big bucks". Non-FAANG software jobs still pay good money. Maybe not "I own a yacht" money, but definitely "I own two new cars and a 5 bedroom house" money.
Hard to say without a lot more detail on location and specialization. All I know is software engineering in DC metro pays well enough to live comfortably. If you’re working a .gov contract on a TS/poly, even better.
There are plenty of high paying companies that aren't FAANG and aren't in the bay area. I work at one. A small software company owned by a huge non-tech megacorp. Zero people have left my dev team in the past 4 years.
Fast Matrix Multiplication is also useful for physics and other simulations, so I'm sure someone will make some SDKs or something for such AI hardware beyond just the scope of AI.
This was a lot of fun. Reminded me of one of the earliest C programs I wrote when I was learning to code on Win98, before I really knew how computers worked. I was curious: "What does the rest of RAM look like?"
#include <stdio.h>
int main() {
for (int x = 0;;x++) {
putc(*(char*)x);
}
return 0;
}
I was mezmerized by the strings from the uncompressed BIOS ROM being dumped back at me, "American Megatrends".. etc. Eventually this process crashes of course, when it runs past the end of mapped memory locations.
It was great when virtually everything was mapped 1:1 into "real" memory, even video memory. In those days, everything was a memory address, just like "everything is a file".
Then came the realization you could alter any memory location, and further, you could write a tiny TSR to do things...
