@a16z Many people use "world model" as a representation of the phisical reality, but a good world model should be capable of storing the representation of abstract concepts and relations, too (like "government", "law" or "company"). In many cases that is related to language, too.
@StuartHameroff AGI needs to have intentions, and we can easily program -or you can say simulate- intentions in any computer. Consciousness is a very effective mechanism that nature has created to do that, but I don't think it's essential for that at all.
@ViperPrompt@Kekius_Sage Hassabis has always been searching for AGI, and all his and Deep Mind advances, including Apha Go, where made with generalization in mind, and as a step in that process, not just as a specialized solution. I don't think that counting him out is a very wise possition.
He estat moderat…
…@Sony o més probablement @SonyEspana ha decidit no publicar la meva resenya per estar escrita en català, o com diuen ells en, “idioma extranjero”.
@jimstewartson Dismissing LLMs as "not intelligence" or "just databases" looks to me a flawed and blind statement, as flawed as saying "we just have to scale current LLMs". LLMs can be "not the whole answer" and be "part of the answer" at the same time.
@jimstewartson Human level intelligence is determined by the capacity to detect patterns, generalize concepts and stablish analogies to solve new problems. Current systems already do that, only is not as good as humans at generalization and analogy. But they will be. (1/2)