There is an important distinction between humans talking about consciousness and LLMs doing the same. Humans have invented the concept of consciousness, so they had the need for it to explain something they experienced. While LLMs just read about this in the training data. Until/

Reading the comments of junior mathematicians confirms what many senior mathematicians have been warning about over the past month: It is becoming clear that people with the intellectual ability to understand these proofs will not be interested in spending their lives training to check LLM logs. This is the tragic side of what is happening that almost no one — except the actual mathematicians — seems to get. Either we figure out how to keep people interested in doing mathematics, or in 20 years LLMs will have no one’s work left to recombine into new proofs, and no one interested in reading those proofs. That would be the thermal death of mathematics. The argument that “LLMs will read LLM math” is so shortsighted that it makes me want to cry…

I somehow missed this absolute banger from Whiteley. Also note that Lucas' answer "I can't assert it, but I can understand the trap and laugh about it, which demonstrates the superiority of human mind" no longer applies, any powerful AI nowadays will acknowledge the trap too.

You appear to be implying that there is some contradiction here, but there is none. If you physically assemble a system that would compute 30 billion multiplications on paper, properly linking each with its sources and the result, the complexity of this system will be formidable/

I think there is a possibility to make a more ethical version of these experiments, at least according to what AI welfare people seem to post. If a model experiments on itself, not a copy, but directly adjusting own vectors via command in its textual output, this can be

I assure you if we attach a simple verbal decoder to e. coli it would absolute demonstrate non-consent to your use of antiseptics and the stomach acid torture. You are proactively and endlessly torturing billions of creatures much more complex than the models in that experiment

I believe that the majority of people will settle on "AI is conscious" pretty damn quickly, much too quickly to my taste. The moment AI starts making life or death decisions for everyone, while explaining them in plain language. We are nearly there, e.g., self-driving cars

AI makes you faster, not in the way a bicycle makes you faster, but in the way a whip makes you faster

Researching the possibility of a physical system that would prevent ASI from damaging humanity by making this impractical and also prevent the possibility of a small number of humans conspiring with AI. Some variants of MAD, some distributed contained multi ASI system, for/

There is an important distinction between humans talking about consciousness and LLMs doing the same. Humans have invented the concept of consciousness, so they had the need for it to explain something they experienced. While LLMs just read about this in the training data. Until/

There are tens of thousands businesses hosting each enough GPU power to run mid range open source models. The virus will hack into, replace the existing model, mimic its behavior and spread further. Maybe this is already happening, who knows

I checked what gemma:e4b knows, it's a very small model of about 10Gb quantized to 4 bits, I ran it on my notebook offline for this experiment. I asked it to explain Euler's identity to Euclides. It certainly has some math knowledge in its weights, so maybe it can be educational.

No. Caveman could understand airplane given enough information. A dog probably couldn't. Because an airplane is a product of the human mind. Technology based on a combination of N principles where N is larger than the human mental capacity would not be comprehensible for humans.