Training data has to be of human origin. The purpose of AGI is to automate labor, because nothing else can justify the cost. AGI requires a model of human behavior, which has to be obtained by observing humans and learning to predict their actions. Human long term memory is 10^9 bits and can only be communicated at 5-10 bits per second. That is 8 hours per day for 10-20 years.
10^9 bits is Tom Landauer's estimate from human long term memory recall tests for words and pictures. Image recall error rates are consistent with a model where a picture viewed for 5 seconds is represented in memory as about 30 bits of features. Written English is 1 bit per character. In my 2013 Cost of AI paper I estimated that 99% of memories are shared or written down, which implies the economy has a complexity of 1% of 8 billion minds, or 10^17 bits. Depending on skill level, it costs about 1% of lifetime earnings to replace an employee. The biggest LLMs are trained on about 10^14 bits. That's why you still have a job. On Thu, Dec 12, 2024, 5:53 PM Boris Kazachenko <cogno...@gmail.com> wrote: > All of technology is "synthetic data". And biology. And chemistry. Remove > synthetic data and all you have is hydrogen. Or Plank constant. > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/T7a144d8141d1d5b0-M05553e7487a6c1f1587abf71> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T7a144d8141d1d5b0-M9c4712f2bb6a39da63c61988 Delivery options: https://agi.topicbox.com/groups/agi/subscription
