AGI is Here. Anthropic Just Proved It.
Anthropic published a report, "When AI builds itself," revealing that more than 80% of the code it ships is now written by its own model, Claude. Nate uses it to argue AGI is already here by the definition that matters. His working definition: AGI (artificial general intelligence) is not sci-fi robots, it is a general model you can hand almost any unfamiliar, open-ended problem and it figures out the approach on its own, researching and experimenting until it finds something that works. That is the line that separates it from narrow AI, which is excellent at the one task it was built for but stuck in one lane.
The primary source this whole video reacts to. Anthropic's report revealing that more than 80% of the code it ships is now written by its own model, and laying out recursive self-improvement. Read it yourself rather than taking the recap secondhand.
Read Anthropic's report →Free and public, no login. This is the source document the video is built on.
- What AGI actually is: "AGI stands for artificial general intelligence." Nate's practical test is whether you can hand a model "a problem that has no clear answer" and it "runs experiments, it does research... and it comes back with something that actually works."
- Narrow AI vs AGI: narrow AI is "really, really good at one specific thing that you actually built it for" (a chess model, a recommendation algorithm, an inbox sorter). "Even if it nails one job at 99% of the time, that's not AGI because it's still stuck in one lane."
- His definition of the real thing: "AGI is when you hand a general model almost anything, a problem that it's never seen before that has no clear answer, and it does the work on its own to figure out the approach."
- "More than 80% of the code that Anthropic ships is now written by their own AI, Claude," per the report "When AI builds itself."
- His thesis: "by the definition that actually matters, AGI is already here," not "this far off future thing that we're all just kind of waiting for."