y @devolvedneuron

@jaylessblame well you completely misrepresented his writing in your previous tweet so i'm not a believer in your reading comprehension

@jaylessblame @ZackKorman i implore you to at least try to understand the fundamentals of his beliefs before spitting out bullshit like this. it takes 2 minutes of actually reading yud to refute what you said

@EricCompeau @i_slamson @manulcapital it's not directly relevant to establishing truth values (2/2)

@EricCompeau @i_slamson @manulcapital i see what you mean. the person I replied to claimed we can't verify the correctness of a proof without understanding it -- it's just not true. seems you're defending a position that prioritizes human understanding of math and that's fine but (1/2)

@EricCompeau @i_slamson @manulcapital the probability would still be deducted from evidence. it's not like you get that quantity out of thin air

@EricCompeau @i_slamson @manulcapital at this point I don't think you know what I even mean by "confidence"... this is not subjective or ambiguous. this is the quantified probability of correctness of a mathematical statement

@i_slamson @EricCompeau @manulcapital this conflates understanding the rules by which a proof is verified with understanding the reasoning contained in that proof. it just seems you're not even remotely interested in the topic and you're just spitting bullshit

@EricCompeau @i_slamson @manulcapital footnote: when I say arbitrarily close to 100% confidence, I mean in that in a literal, quantifiable way

@EricCompeau @i_slamson @manulcapital on the other hand, a protocol with mechanically/probabilistically verifiable evidence can get us arbitrarily close to 100% confidence, where pure deduction fundamentally cannot do that (4/4)

@i_slamson @EricCompeau @manulcapital I'm not going to engage with you because you seem to have no interest in understanding any of the arguments presented here, you just want to vent

@EricCompeau @i_slamson @manulcapital yes! and that's what I've been arguing the whole time. we don't have epistemic certainty with either method. with AI out of the equation, deductive reasoning is of course superior to empirical evidence as means of gaining confidence. but this is not necessarily the case with AI!

@EricCompeau @i_slamson @manulcapital "note that I allow the possibility that the proof is false, and I am mistaken in my understanding, but this is not the case under consideration" so you're operating under an assumption you had not stated explicitly and which is completely separate from the discussion

@EricCompeau @i_slamson @manulcapital if you want to talk about the difference in confidence standard between the two methods that's fine, but you're essentially claiming one of these is infallible

@EricCompeau @i_slamson @manulcapital also I want to reiterate that our whole discussion originated from the question on gaining *epistemic certainty* in proofs. "one of these methods can provide certainty" -- this is just plainly wrong from an epistemic perspective

@EricCompeau @i_slamson @manulcapital when I have more time I can write up the scaffolding of a protocol that would guarantee this, or find a reference that describes it

@EricCompeau @i_slamson @manulcapital an ASI could provide a proof certificate that we can verify probabilistically until we gain a confidence level arbitrarily close to 1