I'm switching my 34-agent paperclip setup from claude code to codex for the next week while I'm in China and see what breaks.
thanks for the 50k @sama
The European democratic crisis is manufactured.
The current debate is not about whether to do [plan x] or [plan y] (most people actually agree across party lines)
Instead, it has become an exercise for leading parties to signal whether to accept a situation wherein legislation is created by way of accepting votes from other parties on the outer edges of the political spectrum (which increasingly make up a large part of the voter base now).
The current gridlock is mostly not based on differing opinions, but rather the anticipation of a PR crisis for establishment parties.
They just feel yucky accepting votes from unaligned parties, out of fear that the loss of face could cost them votes in state elections.
Generally, the political feedback loop does not reward reward making bold/tough choices, like making deals with the "other side".
Ultimately, we're governed by a group of cowards, who are far more worrried about populartiy, than the wellbeing of the country.
Good public comms would fix most of this.
Ultimately I'm confident, that the hard times will bring forth a special person who will take on this responsibility.
✦✦✦✦✦
the singularity is exactly when RL gets broadly more signal from "synthetic" data than human feedback.
had astra implement the dream-rsi paper
really curious to see what happens now
arxiv.org/html/2609.14858v1
Built alicli: Alibaba for Claude and Codex.
Search suppliers, start conversations, read your inbox and download supplier files. Messages go out with your approval.
Private alpha. Get access: [email protected]
alicli.sh
I wonder how i can ask claude to "remove slop" and it knows exactly what i mean.
but strangely this signal never makes it into training.
If this is how they're slowing the frontier, i don't want it.


















