Join 📚 Josh Beckman's Highlights
A batch of the best highlights from what Josh's read, .
Like an unattended turkey deep frying on the patio, truly global distributed consensus promises deliciousness while yielding only immolation.
More capable models can better recognize the specific circumstances under which they are trained. Because of this, they are more likely to learn to act as expected in precisely those circumstances while behaving competently but unexpectedly in others. This can surface in the form of problems that Perez et al. (2022) call sycophancy, where a model answers subjective questions in a way that flatters their user’s stated beliefs, and sandbagging, where models are more likely to endorse common misconceptions when their user appears to be less educated.
We Need to Tell People ChatGPT Will Lie to Them, Not Debate Linguistics
Simon Willison
It makes no difference whether it lasts a long time or whether it lasts a short time.
A galaxy goes together with all the universe just as much as a mosquito. You can get a certain vision of life where everything is seen to be a complex pattern of rhythm dances the human dance the flower dance the bee dance the giraffe dance.
And that's what this all is, it's jazz. You see? This is a big jazz, this world. And what it's trying to do is to see how jazzed up it can get. How far out this play of rhythm can go.
Everything | Gameplay Film
David OReilly
...catch up on these, and many more highlights
Readwise