I’m not sure that complexity is protecting us. On the one hand, there are just 1MB of bases coding for the brain (and less for the connectome), but that doesn’t mean we can read it and it may take a long time to reverse engineer.
On the other hand, our existing systems of LLMs are already much more complex than that. Likely more than a GB of source code for modern LLM-running compute center servers. And here the relationship between the code and the result is better known and can be iterated on much faster. We may not need to reverse engineer the brain. Experimentation may be sufficient.
I’m not sure that complexity is protecting us. On the one hand, there are just 1MB of bases coding for the brain (and less for the connectome), but that doesn’t mean we can read it and it may take a long time to reverse engineer.
source: https://xkcd.com/1605/
On the other hand, our existing systems of LLMs are already much more complex than that. Likely more than a GB of source code for modern LLM-running compute center servers. And here the relationship between the code and the result is better known and can be iterated on much faster. We may not need to reverse engineer the brain. Experimentation may be sufficient.