ferris-booler

26 ms·

1.
by ferris-booler 1mo ago
Yeah I hit this feeling with Claude one too many times and switched from Anthropic Pro to OpenAI Pro. GPT-5.6 so far has been a significantly better technical writer in my opinion, and it's much faster in conversations. My guess is tha
2.
by ferris-booler 1mo ago
> 2027 — A morning at Conduit > I put a band around my head, and open my laptop at Conduit. It’s 9 am. My device pairs with my laptop over Bluetooth. I open, without loss of generality, Codex. At least we are reassured that short-rang
3.
by ferris-booler 4mo ago
I'm hitting this too! And I assumed it was a backwards-compatibility issue with my live conversation with Opus 4.7, but then I hit it in a fresh conversation with Opus 4.8. Vibe code release bug I guess?
4.
by ferris-booler 4mo ago
What strikes me in this case (and I haven't seen in other comments) is that it's a _disproof_ of a conjecture put forth by Erdős and supported (at least according to OpenAI) by other professional mathematicians. Erdős, one of the
5.
by ferris-booler 7mo ago
IMO your question is the largest unknown in the ML research field (neural net interpretability is a related area), but the most basic explanation is "if we can always accurately guess the next 'correct' word, then we will al
6.
by ferris-booler 7mo ago
An LLM uses constant compute per output token (one forward pass through the model), so the only computational mechanism to increase 'thinking' quantity is to emit more tokens. Hence why reasoning models produce many intermediary t