The track record of people arguing for long timelines on the basis of anecdotal AI failures is truly horrendous; such people have been repeatedly surprised by the pace of progress over the last six years.
FWIW this doesn't particularly apply to me, as far as I remember. I've been very early on worrying about the Hackening, and I had explicit carve-outs about "I'm not sure if LLMs can automate formal conjecture verification" there. I did hope that math/hacking progress would also plateau or be slower than it was, but my general LLM bearishness never relied on that.
It may be worth it not to lump all LLM capability skeptics into one homogeneous blob.