@Unknown: You're assuming Eliezer's failure probabilities are independent. That seems wrong, because Eliezer doesn't think randomly.

He's using some collection of heuristics to generate the thoughts we see in the posts. If his heuristics were broken, we'd see a lot more than one mistaken post.

So a .5% error probability per post does not necessarily imply a high probability that he made at least one serious mistake.

Now before anyone accuses me of believing otherwise, let me say that Eliezer is not infallible. Heck, I caught him making a mistake two days ago.

So, I don't think Eliezer has gotten it all right. I do think that he's probably gotten the main ideas right. But there's a difference between saying that every supporting detail of an argument is correct and saying that the main ideas are correct. Eliezer is much more cautious with the main lines of his arguments than with the illustrative examples (and there are entire posts which are illustrative examples or are mostly so.)

Also, Eliezer often gives arguments in parallel rather than in series (ie. making several arguments in favor of the same claim) and in these cases, the failure probability should go down, not up.