TIRM
5hr ago
That's a really good answer.
I suspect other factors would negate this effect. Such as selecting for self-motivated people who can afford to move and buy property. That filters out the listless and destitute. I predict a photo collage of these people vs equivalent income and age white Americans would not obviously show they are ugly losers, as comically shown in that link.
I keep inheriting MATLAB code at work. It is horrible. Can't use it in production since production computers are locked down linux machines that don't have MATLAB. I grit my teeth and do much my work in MATLAB.
BUT NOW, we have an LLM at work approved for our use. I feed it large MATLAB scripts and tell it to give me an equivalent vectorized Python script. A few seconds later I get the Python script. Functions are carried over as Python equivalents. So far 100% success rate.
This thing rocks. Brainless "turn this code into that similar code" tasked take a few seconds rather than an hour.
I had a thermodynamics issue that I vaguely remember learning about in college. I spent maybe a minute thinking up the best way to phrase the relevant question. The LLM gave me the answer and responded to my request for sources with real sources I verified. Google previously declined to show me the relevant results. I now have verified an important point and sent it and high quality sources to the relevant people at work.
It is not perfect. I had a bunch of FFTs I needed to do. Not that complicated. As a test I asked it to write me functions to FFT the input data and then to IFFT the results to recreate the original data. It made a few functions that mostly match my requirements. But as the very long code block went on it lost its way and the later functions were flawed. They were verifiable wrongly. It helpfully made an example using these functions and at a glance I saw it had to be wrong. Just a few hundred lines of code and it gets lost. Not a huge problem. Still an amazing time to results ratio. I clean up the last bit and it is acceptable.
I won't ask these things about potential Jewish bias in the BBC or anything like that. I will continue to ask for verifiable methods of finding answers to real material questions and reap the verifiably correct rewards.
More options
Context Copy link