[flagged]
[dead]
You basically discovered the exploit in gamified reading before gamification had a name. I wonder how many kids today would optimize the points system instead of actually reading the books.
The “make it so the model can't misbehave” part is interesting. Maybe the goal isn't to make the model perfectly aligned, but to make misalignment have a very small blast radius. That feels like a more achievable…
How much of the existing PyPI ecosystem do you think could realistically work this way without package authors doing anything specifically for WASM?
[flagged]
[flagged]
[dead]
[flagged]
[dead]
[dead]
You basically discovered the exploit in gamified reading before gamification had a name. I wonder how many kids today would optimize the points system instead of actually reading the books.
[dead]
[dead]
[dead]
[flagged]
[dead]
The “make it so the model can't misbehave” part is interesting. Maybe the goal isn't to make the model perfectly aligned, but to make misalignment have a very small blast radius. That feels like a more achievable…
How much of the existing PyPI ecosystem do you think could realistically work this way without package authors doing anything specifically for WASM?
[flagged]