I'm not so sure it's just perception involved. You're going along a street and can hear a violin playing. A melody develops, you establish a sense of tonality, can predict next sounds maybe. And then you also make out a recorder. Another piece, totally unrelated harmonically. Another stream to process, I would assume a brain can handle it (at least a musical one should).
I agree that language processing is different, but some similarities do exist.
It's incorrect to reduce polyphony to just shifting harmonies. Harmony is of course the constraint but the whole point is about intertwined voices - various melodies that your brain can process simultaneously. Music is not speech, but it's not that far from the language as it seems in terms of processing. When you hear 2 melodic lines at the same time, you're brain needs to encode them separately, in a similar manner to parallel speech streams.
Ultimately, AI models will need to ingest the slop ziggurat they created. This means, other things being equal, that it won't be so easy as it is now to achieve incremental gains from the sheer fact of gobbling up more code.
I wonder if it would be possible for Starlink to use less reflective materials for their satellites so that the sky is less polluted for the astronomers.
Yes, I was taught that relativity is a significant part of quantum chemistry equations in gold atoms 25 years ago. The idea is quite old and the title is misleading.
Calling someone "stinky manager" is not ad hominem. It's a way of saying that he's managerial skills are very poor. Does it relate to the argument? Very much so. Changing a bread factory into a slop factory makes Jarred a stinky manager and even a "sloppy" one
I think you meant lossy compression and not lossless. I'm not suggesting this as a method to extract those books from the models, which by their nature are not databases. Just commenting on the somewhat surprising fact that the bigger the model the more likely it is to produce some (short) excerpts of the original training material
Nice quote from Tao Te Ching about complexity and simplicity completing each other.
The rules of go could be explained to a 4 years old. On the other hand, the superficial complexity of so many framworks/systems is just a facade and nothing more.
The same goes for NP-hard problems where complex solutions have trivial verification methods.
This is a moot point though. Before all the jobs are replaced in 2036, shouldn't we see data centres burning? Revolutions tend to succeed with only 1% of population taking part. I think it's probably that this revolution will ultimately fail for various reasons, but still it's a valid scenario, which at the very least should be mentioned.
I only got 4 wrong as a non-native speaker. Okay, I'm widely read in English, but among LLM-generated definitions it's just too easy to spot the right one.
The article is in French, but geology is the key factor. Do you need to bore rock or sandy soil with tar? Is the area seismically active like in Los Angeles? This affects the cost and timeline of metro construction more than just wages.
Let's imagine that half of the jobs or more are lost. It would create a shock for the economy as a whole, incomparable with anything that happened before.
It's a snake eating its own tail paradox. It means massive bankruptcies across all sectors and some governments defaulting on their debts. We can forget about UBI in these circumstances. Yes, OpenAI and Anthropic will pay more taxes but the economy as a whole will be uprooted.
I agree that language processing is different, but some similarities do exist.