Thoughts on digital intelligence

A conversation with chatGPT helped me advance my transcendental meditation practice. I was struggling with breathing and having the mantra always align with breathing. I could not figure out how to disjoin the two. chatGPT convinced me to use a much lighter mantra; less concrete; separate from time in a way.

Apparently the leading LLMs beat a formalized Turing test recently. Quite interesting how the conversation around "intelligence" of LLMs is developing. I guess the thing about how we think of intelligence has to do with continuity. Even agents don't start to solve this problem. One approach to progressing in this area could be that every conversation is vectorized. Every vector and conversation is given various rankings and score as measures of importance. Then during a "sleep" phase the model itself is reinforced with the novel information from the conversations. You could also implement tiered storage where less important information (referenced by the vectors) is put in a cheaper, slower, cold storage, especially as it ages without a fresh indexing.

I suppose the fundamental limitation or difference from biological intelligence is how information is stored. The three dimensional interconnected brain leverages complex interdependencies, local triggers, wave combination and frequency; in short it's a unified system. It's analog and it feels that digital can only ever be a cheap imitation, as a television is to the real world.

Some have argued that AI is still waiting for a fundamental breakthrough in technology in order to reach true intelligence. I suppose it's a hardware breakthrough that's needed, based on the above line of thought.

On the other hand, never discount the profound power of iteration. These technologies are evolving quickly and there is no telling where that will take things.