> We attach basically zero value to writing a new program
What does it mean "new"? And, was it a difficult or trivial accomplishment?
A solution to a well known open math problem is both new and non-trivial- you know that many, very smart, very well trained human experts have dedicated time to the problem and haven't been able to solve it, despite good incentives.
> still one area where it is notoriously bad is creation of novel things
Is there any proof of this? My feeling is that people just assume that since something has been generated by AI, then it must not be original. And truly original thought or craft is exceedingly rare if it exists at all: everything comes from something else. Individual innovations are often small.
> Because that is the only loophole in LLMs that can not be fixed
This is magical thinking. Humans don't have a small device in their head capable of "true creativity" whatever that even is. At most, they experience so much random input that their outputs are less predictable. But nothing that LLMs won't be able to do soon; as for raw intelligence, they're already smarter than the large majority of the population.
Israel keeps acting as a cartoonishly evil advisor trying to manipulate the king with flattery and outrageous lies. It's so evident and so vile that it's funny.
Because you can share types and even modules with your frontend project? Because for applications that aren't CPU-intensive it makes almost no difference? Because you are familiar with it and like it? Because of the humongous amount of libraries?
> then they shouldn't care whether it's a human or a LLM.
I imagine that the whole point of posting a task to Mechanical Turk nowadays is that you want it to be completed by humans. Either because you are after the small discrepancy between AI and human performance, or because humans are the object of your investigation.
Anthropic won't do it, but they published the j-lens to introspect the model- from what I understand it's roughly simply feeding a chosen layer straight into the final layers of the LLM for decoding into language:
It's been shown that LLMs use their outer layers to decode from and encode to language, while their middle layers deal in language-independent abstract concepts. This means that the same question or statement in different languages activates the outer layers differently but produces the same patterns in the middle layers. Check this article with cool visualizations (btw, this is one of the articles mentioned also by a sibling answer):
The middle layers also perform reasoning on the abstract concepts, to the point that you can replicate some blocks of inner layers (thus giving the LLM more internal "reasoning space") and by this increase the model's reasoning abilities. The video in this article shows that when performing a sequence of arithmetic operations (without CoT, i.e. the result is spit out directly), internally the intermediate calculations are spelled out, and this can only happen in the depth direction of the LLM (since no new token is added to the sequence). So this "jspace" can only be situated in the middle layers, probably in circuits that repeat nearly identical across several layers.
> Those 80% were constrained in the volume they could output pre-AI, but now they are unleashed to do a large amount of net new work but many without the skills to structure it well+maintainably.
You could say that AI turned them from stupid and lazy to the famously dangerous combination of stupid and hardworking.
Lol. No, I've always put a lot of care in what I made and strove to give people tools that really made them happy. And I still do, steering the AI left and right. But on the other hand yes, seeing the machine fast-forward through most of the reasoning, weighing, understanding, cross-checking, deciding that used to fill hours of my day is... vertiginous. It's real intelligence, and it makes us much less useful and relevant. If your refuge is that whatever the machine can do has never been worth of a human being, you will find yourself soon squeezed in some very narrow corner, and/ or you'll suffer some deep crisis of meaning.
Does the fact that I barely wrote any lines of code in the past six months while my job has been for the past 20 years (and still is) that of producing code, feels "life changing" enough?
I'm a (light) smoker, and I do get the point of a generational ban: there is almost no benefit to tobacco usage, it's a thing you do just because you did it before. Alcohol and drugs, as dangerous as they might be, do provide relief and entertainment. In the case of tobacco, people who never did it never miss it, nor they miss any particular experience that comes with it. Of course, for smokers is a pleasure- but it's surprising how the need itself- and most of the pleasure you get by satisfying it- simply disappears if you can keep yourself from smoking for a while. Tobacco is simply a parasite of human beings, exploiting their nervous system for its reproduction.
The chant, as officially translated by the government itself [1] means "Down with the USA". "Death to" is a purposeful (literal) mistranslation to make it sound like it's a genocidal statement.
