The AI-enabled coup rests on a loyalty no one has learned to build
Tom Davidson's 2025 report argues advanced AI could replace the people a coup has always needed, but the reliably secret loyalty that would take is the unsolved failure the alignment field calls deceptive misalignment.
A generated image gets its letters wrong before the picture is ever drawn
The word reaches the drawing step as a whole-word token with its spelling already blurred, and the fix is an encoder that reads the prompt one character at a time.
Prompt engineering lifted CLIP's zero-shot ImageNet score by nearly five points
Radford, Kim and their OpenAI coauthors trained the 2021 model on 400 million captions they never released, and its label-free ImageNet result stayed strong on unfamiliar photos while collapsing on counting and satellite tiles.
The function-calling score labs cite grades one call, and usually never runs it
Berkeley's Function-Calling Leaderboard scores each model's answer against a known-correct call, so the 66.4% at the top of its board mostly reflects single calls, blind to whether the same model can chain those calls across a whole task.
Facebook translated 'good morning' as 'attack them,' and police made the arrest
No Arabic-speaking officer read the original post, so a construction worker in the West Bank spent hours in detention over an error one Arabic reader would have caught.