It's also excellent evidence that he spends most if not all his time being fed talking points by the AI companies instead of looking at anything that would challenge their hype. No interviewing of critics, no analysis of the financials, nothing.

Casey, you said you like to bring on guests to help to "get smarter about the subject", but when bring on only one kind of guest, it's hard to take your claim seriously.

It then went on to revise the algorithm to use an implementation of Murmur hash. Next I asked Claude why it chose to track its filter using an array of boolean values as opposed to something more efficient. /3

First, it used MD5 as its hash instead of something more performant, so I asked Claude about why it chose MD5. Claude correctly responded that MD5, while being broken cryptographically, was nevertheless a reasonable function for an algorithm that isn't impacted by MD5's lack of security. /2

A few weeks ago, I was curious about how good LLMs had become at writing code. So I created a Claude account and asked it to write me a Bloom Filter in C#. What it came up with was a workable, correct solution, but it had a few deficiencies. /1