I mean GPT is great, it can predict nicely what comes after MANY different
prompts fed to it, but there is WAY more even more prompts that can be fed to
it that it will pretty much fail at, or at least as many as it currently can
ace. To improve GPT into AGI closer at least, it may need to take long times to
"think" before making an instant prediction. I was thinking that the code to do
this is simply memories - no code, the memories simply carry out methods used
on prompts, and uses motor to flip and churn the prompt and collect new data to
help it answer this specific prompt. For example if it saw the prompt '3, 7,
22, 56, 27464, ?' it may match the fabricino math pattern but it may see it is
not that one exactly, and so it may try to find a pattern - maybe it is
doubling the number? 3, 6, nope it was 7 in the prompt, what if it is doubling
+1? --- 3, 7, 15, nope 22 was there, maybe it is tripling after doubling? And
so it tries to fin a pattern here like I tried, if it can 3>7 and 7>22 and
22>56 etc all match as the same pattern, then it can use that for the last
awaiting number to add onto it. This can maybe be ran in GPT but not by adding
code for all such, but rather letting GPT use memories to carry out this search
for patterns/data.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/Tfaee106f616454ab-M2154ee6d81d9786f05b0a1d2
Delivery options: https://agi.topicbox.com/groups/agi/subscription
