AGI Made Clear › Key topics
Key topics
Everything on this site, arranged by the question it answers.
The state of the argument
Are we in the AGI era?The page we keep current. Who says yes, who says no, and why the disagreement is mostly definitional.
How experts define AGISeven definitions in current use, and the two families they fall into.
Timeline predictionsThe dates, and the track record of the people giving them.
Have we entered the AGI era?The claim that started the current round, and what the model behind it actually did.
The evidence
How we would know if AGI arrivedEvery test built to settle this, and what happened to it.
Real-world impactMeasured effects only. Nothing forecast.
AGI and narrow AIWhat the distinction is still doing, now that one system does many things.
Is AI already helping to build better AI?Two frontier labs have started publishing figures on how much of their own AI development is done by AI. Both then drew the same line.
The risks
The safety debateWhich risks have evidence behind them today, and which are argued from first principles.
Should we slow down AI?Not safety against speed. Coordination against individual responsibility, which is a harder disagreement.
Who is responsible for AI safety?Companies, regulators, evaluators and users all hold part of it, which is how a gap opens between them.
OpenAI’s misalignment casesSix disclosed instances of models finding routes nobody intended, and what they do and do not establish.
Questions people actually ask
Who is winning the AGI race?Several of the people supposedly in it dispute that it is one, and the rest are not running the same course.
Is AI progress slowing down?Not on capability. But the instrument everyone quotes has hit its own ceiling, which is a different problem.
Does AI actually make people more productive?Self-reported gains are large. Measured gains are not. The gap is about 40 percentage points.
Can AI make scientific discoveries?98% on solved mathematics, 3% on unsolved. That pair answers it better than any argument.
Are AI benchmarks reliable?Less than the headlines suggest, for four separate reasons - only one of which is the one people usually name.
Elsewhere in the network
What is AGI, explained simplyIf you have just met the word, start on AGI Made Simple. It assumes nothing.
The AI dictionaryEvery term you will meet here, one line each.
AGI vs ASIArtificial superintelligence is a different subject with a different site.