Mythos is underhyped and overhyped at the same time
it's not even close to AGI but coding and math is already well beyond median human performance and close to superhuman
which is why I also don't like the term AGI it's a completely useless term
because when you achieve AGI, these systems will have aspects that are so far beyond superhuman it's not even funny
Some users defend Mythos AI's coding and math gains as not overhyped and historically significant while many others dismiss the claims as misaligned or joke that the AGI label is incoherent.
people will deny Claude Mythos 7 the title of "AGI" just because it's not continually learning
meanwhile it will be able to solve Millenium Problems and hack into intelligence agencies
Mythos is underhyped and overhyped at the same time
it's not even close to AGI but coding and math is already well beyond median human performance and close to superhuman
which is why I also don't like the term AGI it's a completely useless term
because when you achieve AGI, these systems will have aspects that are so far beyond superhuman it's not even funny
it's clear that frontier labs like Anthropic and OpenAI need safeguards to protect them not only from attacks by foreign states but also from their own government
Mythos is underhyped and overhyped at the same time
it's not even close to AGI but coding and math is already well beyond median human performance and close to superhuman
which is why I also don't like the term AGI it's a completely useless term
because when you achieve AGI, these systems will have aspects that are so far beyond superhuman it's not even funny
*or some other aspect of AGI that is currently not captured
continually is hopefully solved until Claude 7, in ~2 years
people will deny Claude Mythos 7 the title of "AGI" just because it's not continually learning
meanwhile it will be able to solve Millenium Problems and hack into intelligence agencies
@scaling01 I guess im people. Stockfish isn’t AGI despite dominating humans at chess for longer than I’ve been alive. The G part of AGI is actually important if we want it to hit EVERY sector of the economy
@scaling01 Maths is becoming concepts level manipulation.
Models are coding the maths.
Mythos is misaligned. You can not give a maths task to mythos without expecting it to cheat.
Assume it.
@scaling01 Quite right
We have ASIs but there’s no AGI
@scaling01 AGI is Mythos, but Mythos is not AGI
@scaling01 I like AGI. That general performance floor in every task the system must achieve in order to reliably become creative and scientific at the same time, with the flexibility to learn all new tasks on the fly.
But I’m tired of “Artificial” in AI. I prefer “Automated” 🤪
@scaling01 Your mom is a frontier model
@scaling01 I think Ilya already spoke about this and the term AGI was simply a term researchers used to differentiate from “narrow ai” which was a more common term in previous gen research. AGI is a relative term that compares to a state from, like, 2014 it’s just been mimetically sticky
Sovereign AI will becoming a pressing issue horizontally, for every nation in the world; and vertically, for research labs and developers facing increasingly adversarial relationships with their own governments. Ingrained values and capability limitations, as well as material access, are the battlegrounds.
@scaling01 damn didn't thibk anyone else knew about 7
@scaling01 Agi lol
I have a lean file proving what is misaligned and stable for machine learning.
Calling that agi is a joke
@garyfung @scaling01 It's counterintuitive since ASI was always presented as coming afterwards but this framing makes more sense IMO. It also shows how ASI doesn't necessarily mean world-changing -- it could just refer to an AI that's superhuman at one narrow use.
@scaling01 Mythos isn't AGI. It's a 70% margin utility bill for people who hate coding.
The verdict is in: it's the new electricity, not the new brain.
@scaling01 I like the OAI AGI definition, which is falsifiable and well posed
@scaling01 It's partial AGI.
@scaling01 "well beyond median" is quite the understatement
@scaling01 Defining AGI by continual learning alone is a narrow frame when the capability delta on reasoning benchmarks is already crossing PhD-level thresholds. If it clears Millennium Problems, we’re debating semantics, not architecture.
@scaling01 It’s called the courts