My stubborn and completely uninformed opinions on AGI:
Summary
Slava Akhmechet argues AGI arrival will be indisputable, using the absence of reasonable disagreement as a circular but useful proxy test.
Disagreement signals AGI absence.
Akhmechet proposes a proxy test where AGI is only present once reasonable people stop earnestly debating its existence.
so long as reasonable people earnestly disagree, it is not here
Implementation precedes definition.
A precise definition of AGI will only emerge once a reference implementation passes the proxy test.
The reference implementation will inform the definition
The Turing test failed.
Users are no longer tricked like they were with Eliza; modern users can often identify specific models.
Timelines remain unknown.
The author cannot predict if labs hold AGI internally or if it requires years of doubling, but maintains the risk remains high.
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.