Just over two decades ago, it was proved by a small group of computer scientists that optimal AGI is equivalent to optimal data compression. The second part of their proof showed that AGI is not bound by computational complexity but rather something else. This 'dark' knowledge has since then vanished from frontier AI research (hint: it is not AIXI). I suspect DARPA/NSA forbid them from publishing further. Even saying it out loud feels like invoking Voldemorthttps://groups.google.com/g/comp.ai/c/02m5jQr5xs4
AI does not yet exist, thus AGI ASI and all the other fan-fiction derived from LLM human mimicry does not exist, either.
>>109321535It's the most retarded form of compression one could immagineYour picrel should've invited the entire royal family to his final restaurant show off.
>>109321535To even start developing AGI, you need to consider how you functionYou:>see>hear>smell>taste>feelAI has>text input>vision model (sometimes)You would essentially need to feed AI with raw information from multiple sensors that always keep it active (there's no more starting prompt, it always receives and interprets information), this is a shift from standard chatbot but it's a start towards simulating consciousness
>>109321567It ID'd my dick as small sea cucumber ... :'(
>>109321535Only 50.000 for a revolutionary algorithm ?!pff, won't even waste my time to spit on on this page.
>>109321535It's part of it, but clearly these models can prodce things that are not in the training set.
>>109321535>it was proved by a small group of computer scientists that optimal AGI is equivalent to optimal data compression.It's in the same category as the Universal Approximation Theorem and other "profound" theoretical results that have no actual bearing on reality.
>>109321677>clearly these models can prodce things that are not in the training set.So can a linear regression model. You're a brown retard and you should stick to what you know.
>>109321546this, and spot on
>>109321535>AGI is equivalent to optimal data compressionYeah i know. Though really it's only half the picture, as knowing and doing are entirely different problems, as one is creating and updating a knowledge "graph" and the other is traversing it.>>109321677Via interpolation, except it can't learn dynamically so it can't build off of its own interpolated data (outside of its context) so it has the flexibility of a brick.
>>109321546>>109321728LLMs are AI. your semantic war is retarded. spend your energy more productively.
>>109322349Cool story, Dario
>>109321535>optimal AGI is equivalent to optimal data compression.what the fuck does that even mean>as since then vanished fromprobably because its incredibly fucking retarded
We don't need AGI.Just strapping LLMs to tools is more than good enough.AGI is just a jack of all trades, master of none, like the Atari Jaguar hardware.
>>109321535Until people understand the metaphysics of consicousness and why the human brain works for expressing consciousness, AGI will be a total meme. You can terraform the planet into a massive datacenter with the most bleeding edge hardware but you're still not getting AGI. Total Information Awareness on the other hand will not only be possible but almost perfect. Too bad everyone's now dead.
