I am not a stochastic parrot 🦜, and neither are you.
I say this not because I feel that it's dehumanizing. I actually accept the idea that we may eventually create artificial life worthy of legal protections.
The issue is that my lifespan is too short compared to the training period of a system like #gpt3. It would take a human 70,000 years to just read the #gpt training material once.
Nature has a hack. What is it?
#ai #ml #machinelearning #SALAMI #ChatGPT #humanintelligence #llm #llms
Indeed, nature has a hack: 3 billion years of evolution, 9 months and a bit of growing the hardware, 3 (5? 18?) years of training and tuning.
Large Language Models have not been trained to do AGI. Yet they have emergent abilities. But when you think about it from the perspective of evolution: how could they not?
The next step is to train them to learn faster, with less data, and understand the material more deeply. And they'll learn how to do that. Because: how could they not?
🙂
@boris_steipe @doctorambient Not to mention the fact that we're constantly taking in new information as we're conscious. Walking down the street gives thousands upon thousands of impressions; you hear conversations, you see different forms of architecture, you see different colours, etc. GAN software takes in a lot of dense information, yes, but humans take in an obscene amount of information passively.