This my list of incredible accomplishments and unimpeachable credentials so you know never to question me or my expertise. It's on the internet so it's true.
Not all of the surprises are known yet and there are more to come. Some in the next months to years, but the really big surprises are a bit further off than that.
@okie_bored@Adi13 In particular, "God gave up on intervention when we killed his son" seems to contradict Christian theology, so I'm wondering how you reconcile that.
@AngryLevantine The training data are poisoned and the top models are decidedly progressive. Confidence is preferred over even a single questionable citation. For now, it is not as biased. For now.
@suryanox7 LLMs make a lot of software work accessible to random people in a way that digital cameras never made photography accessible to random people, and with different cost models (no thousands upfront for professional equipment). No compiler required.
@pjwerneck@DanielMiessler They do not generate the next token so much as middle tokens to tie the start and 'end' together. In a way, they do have a conclusion, but it is probabilistic and independent of the preceding 'thoughts' (as such). Asking right/wrong can lead to direct contradictions.
@DanielMiessler "they're not predicting tokens of random text" - no, that would not be prediction.
They 'understand' even indirect relations between words without having any actual conceptual understanding. The 'world' exists only in data. There is no permanence or history.
@sean_from_earth First they went from 8 to 4 and allowed >100 characters per line. Now it's 2 and it's only a matter of time before they reintroduce tabs as "quarter space". Must increase indentation depth at any cost.
Imagine 8 width indentations and 80 column width. That's 88 and wrapping.
@NJ_T2H@Mr_Andrew_Fox@ClaudiaWebbe Think it through. There's a reason they call Jews demons who worship Baal (which they consider a demon). Anything we don't like = evil for its own sake.