Discussion about this post

User's avatar
stealthmouse's avatar

I like reading your ChatGPT conversations. First, because I would never have the patience to do so myself. And second, because they reinforce my belief that (currently at least) AI is merely a faster means of data gathering and simple interpretation with a conversational tone (one that probably serves to reassure users even when the output is suspect). It's been my belief from the start that AI suffers from a "commons" problem—although it tends to provide pros and cons to argument, it uses a traditional, "understood" baseline in its analysis. Thus, it's unlikely to reference RFK Jr. in any discussions of Big Pharma or COVID. The other problem with a conventional viewpoint is that it tends to regress to the mean, a problem already inherent in LLMs that scrape the Internet for information without first realizing that 90% of the Internet is just crap that people have copied and pasted from somewhere else even if it's wrong. Thus, AI "learns" to be as stupid as the average person but is armed with considerably more data.

To wit, this gem: "💡 Observation: Ozempic is ~20–30x [sic] cheaper than a full-time bodyguard."

The (logically flawed and mathematically untenable) tendency to refer to things as 5 or 10 times less came about, in my estimation, within the last 10 years. Referring to something as 20x-30x cheaper follows the same pattern. If AI used actual logic and understood math and grammar, it would not make such egregious errors. Or maybe AI hasn't learned percentages or orders of magnitude yet.

Anyway, keep up the good work. You are a man among men (or those pretending to be men).

Expand full comment
1 more comment...

No posts