In addition to transparency, I now believe that cutting-edge models should undergo mandatory third-party testing for cyber, bio, and autonomy risks — with the power to block or revoke the deployment of models that present a risk.
AGI
-
Anthropic believes transparency is no longer sufficient for cutting-edge AI
By
–
Anthropic has long advocated for transparency requirements for cutting-edge AI, because the risks were not yet clear enough to be precisely regulated. This is no longer sufficient.
-
Gary Marcus’ 2024 prediction: LLMs commodity, no AGI, profits squeezed
By
–
Me, 2024. LLMs will be commodity; (except for Nvdia) profits will be hard to squeeze out. Techbros: Shut up, Gary. GPT-5 is gonna be AGI. Today: LLMs are commodity; (except for Nvidia) profits have been hard to squeeze out. (Also, still no AGI.* *per definitions generally
-
Altman: Delaying IPO may be advantageous with fast RSI takeoff
By
–
“The faster the potential RSI takeoff looks like it could be, the more it could be advantageous to delay an IPO,” because the “technology and the world may change in surprising ways, and there might be good reasons to be a private company during that time,” Altman said
-
AGI ALPHA paper: scaling intelligence across organizations
By
–
The Transformer scaled intelligence inside models. The next frontier may be scaling intelligence across organizations. I’m sharing my paper: AGI ALPHA: A Scalable Substrate for Intelligence Organizations The core thesis: AI progress will not be defined only by stronger
-
AGI as an evolving institution: AGI ALPHA
By
–
AGI will not be a chatbot.
It will be an institution.
I share my article: AGI ALPHA: An evolutionary substrate for intelligence organizations The central idea is simple: The Transformer scaled intelligence inside models.
AGI ALPHA aims to -
Rebuilding Claude Code: ClawCodex in Python (open source)
By
–
Someone just rebuilt Claude Code from scratch in pure Python. The tool is called ClawCodex and it's 100% open source. That's 180,000 lines of pure Python. No TypeScript runtime required. The original only talks to one model family. This rebuild routes through six providers.
-
Concretize the wall as entelechy to make it real and falsifiable
By
–
But the wall cannot be an indefinite entelechy. It must be concretized to make it real and falsifiable. We start by saying: 'the models approach a wall because they will not be able to do…' and with this we will see if it is real or not.
-
Criticism of those who denied models would improve
By
–
"because after the event, everyone is a sage. You are doing the same." No, friend, it doesn't work like that. Here, at every point along the way for several years, we have had people denying that the models would improve further, seeing invisible walls. If it had depended on them, they would not have…
-
Risks of rushing toward super-AI: generation, jobs, democracy
By
–
@dr_l_alexandre wants to rush toward super-AI. Me, I have time and that's why I measure the risk. 25 years: I have time to see what the rush hides: a precarious generation, liquidated jobs, and a democracy that hasn't digested the rupture. Who decides