Standards to expect from YouTubers

So you've probably already heard Hank Green,  the very successful and well-regarded internet science communicator, Hank Green got kind of  caught by his audience relying a little too heavily on generative AI to research and  possibly even write some of his videos.

I think this is a deal, but it's not a big deal,  especially considering how Hank has handled the fallout and considering the reservoir of goodwill  he has built with his audience over the years.

I do think this incident offers an opportunity to  examine how we are setting and enforcing editorial standards within the fact-based media business  online, especially in video form. What are the professional standards for doing the kind of  work that Hank Green does, that I do? And how

do we re-professionalize journalism and other  kinds of informational public communication as those businesses have reconstituted here on social  video, or do we even want to re-professionalize those fields? That is the main thing that I have  written about for you today. But first, let's

talk specifically about ChatGPT and the like. I do think that large language models have some utility as research tools. They're not nearly  useful enough to justify the massive catastrophic misallocation of capital into highly speculative  AI businesses that will never make enough money

to return those investments. And the financial  crime, the mendacity and/or the negligence that has inflated the AI bubble is a global scandal  that will hurt all of us, especially here in the United States, where we have already spent a huge  chunk of our national wealth on this last-ditch,

hair-brained scheme to save our country  from decline and eventual irrelevance, like the Darien scheme. Scotland, at the turn  of the 18th century, invested a fifth of its national wealth to found an American colony in  what is now Panama, thinking correctly that the

Isthmus of Darien, the narrow point between  the two American continents, that that would become a crucial choke point for global trade.  They were right about that, but they were about 200 years early, and they all died of malaria,  and Scotland had to be absorbed by England with

the Acts of Union to settle its debts. AI is the United States' Darien scheme, I fear, and large language models are not useful  enough to justify all of that, but they can be useful for research. Sometimes I will wonder,  hey, have any scientists studied how cooking

affects the pH of onions? And I'll go looking  in the research literature, Google Scholar, et cetera, and sometimes I can't find anything.  So I'll ask an LLM that has scraped the entire internet, and sometimes the AI finds me a source  that I missed. "Yes, cooking does lower the pH

of onions," the chatbot will say to me, usually  with much more confidence than is warranted by the actual findings it is referencing, and sometimes  it outright mischaracterizes those findings, and often it cites sources that I do not regard  as reliable. A post on Quora is not on the same

level as a peer-reviewed journal article. I have found LLMs useful as a means to find primary sources, but of course, that's really  all they should be used for when we're talking about researching the published record of all  human knowledge. The LLM is a secondary source.