Noah Smith, ki se je odpovedal akademski karieri (na Brown University), ker je ocenil, da ima večji vpliv na javni diskurz kot bloger, kot pa bi jo imel kot profesor ekonomije, v zadnjem komentarju lamentira o zmanjevšanju vpliva blogerjev oziroma sebe na javno mnenje. Navaja tri razloge. Prvi je porast političnega populizma, ki inherentno ne mara objektivne analize, diskusije in eksaktnih podatkov. Drugi je množičen prehod blogerjev na platformo Substack, ki omogoča finančno lukrativno monetizacijo avgtorskih prispevkov in blogerje nevidno sili v hiperprodukcijo, da bi zadovoljili razpršene naročnike. Tretji razlog pa je razrast uporabe umetne inteligence, ki omogoča relativno dobro (čeprav močno filtrirano) informiranje zainteresirane radovedne javnosti prek povzemanja množice javno izraženih mnenj ekspertov, kar pa seveda zmanjša interes za branje specifičnih oziroma “relevantnih” blogerjev in vpliv njihovih mnenj. V dobi UI so lahko vsa mnenja enakovredna, čeprav objektivno seveda niso.

Spodaj je nekaj odstavkov iz zapisa Smitha na to temo. Naj pred tem omenim, da sem Smitha kot blogerja vednp cenil zaradi odličnega znanja ekonomije in sposobnosti koncizne analize. Vendar sem Smitha “izgubil” kakšnih 5 let nazaj zaradi njegove izrazite ideološke obremenjenosti s Kitajsko, Rusijo in nasploh z nezahodnim svetom, kar Smithu v mojih očeh jemlje sposobnost objektivnega razmišljanja o naravi stvari, ki tangirajo polje geopolitike. Ampak to seveda ne vpliva na to, da se mi njegovo mnenje o tej konkretni – apolitični – temi ne bi zdelo zanimivo.

Unlike many people, I think AI writing is actually pretty good. Yes, there’s a recognizable style that the basic models use (“It’s not X, it’s Y” and lots of other little cliches). That style isn’t bad, it just gets overplayed when everyone uses it.2 Yes, AI models are still not great at boiling a complex idea down to one or two pithy sentences. But you can modify the style that AI uses. And AI can do plenty of things human writers can’t — it can seamlessly incorporate vast knowledge and novel data analysis into a piece as it writes it.

For example, I immediately suspected that this essay by Aaron Brown, Michael Mendelson, and Cliff Asness, on the confusion of the debate over “affordability”, is mostly AI-generated, and Pangram — the most reliable AI text detector — flagged it as around 50% AI. But that’s not a knock against it — the essay is great. It classifies different kinds of “affordability” problems — true poverty, precarity, downward mobility, etc. — into different buckets, gives some illustrative vignettes, and provides some useful numbers about each one. I broadly agree with the article’s conclusions, and I think it’s a valuable addition to the discourse.

A bigger problem is that in a world where a huge number of people generate effectively infinite amounts of good-quality content like this, it becomes hard for readers to decide where to allocate their attention. Instead of identifying the few most consistently useful blogs and reading those in great detail, a lot of people will respond to the explosion of content by “reading” a larger number of posts but only lightly skimming each one.

It’s not my job I’m worried about here. It’s that in that world, even if my blog continues to get tons of readers and make me plenty of money, what I do becomes less important. If people are just skimming what I write so they can move on to the next 10,000-word Claude-generated post, the fact that they’re paying me $10 a month is cold comfort — I’m not really reaching them. And even more worryingly, no one is reaching them — if they’re skimming 100 posts a day instead of reading 10 all the way through, they’re not getting really good information from anywhere.3

I don’t know how severe this problem will be, to be honest. There was always a lot more high-quality content on the internet than anyone could ever read, and a lot of people always just skimmed my posts instead of reading them closely. Maybe AI can’t make this problem worse because it was already maximally bad.

Also, I’m optimistic that AI itself will open up new channels for intellectual influence. It’s a well-known fact that if AI just consumes AI-generated output, it gets worse and worse. So AI companies try very hard to “clean” the text they use to train their models. Human writers, whose personal experience brings in new data for AIs to learn, can influence the world if their writings are used to train the next generation of AIs.

Interestingly, I think I’m already doing this, quite by accident. I don’t know how reliable the website intheweights.com is, but it shows me in the top 2% of contributors:

I suspect that on the topics I write about, I’m even more influential. Claude and GPT often cite me as a source on topics I write about4, and friends have told me that Claude recommends my blog with surprising frequency when they ask it for reading material. Maybe Tyler Cowen is right when he says we should be “writing for the AIs”.

In any case, I find that although blogging is still very fun, and I still think I’m having a positive impact, and my readership is still growing, the environment a lot more challenging than it was just two years ago. The combination of a nation ruled by closed-minded tribalists, a blogosphere obsessed with putting out monetizable content, and the rampant proliferation of high-quality AI output is forcing me to rethink what I do. I want to keep injecting ideas into the discourse and participating in a vibrant and relevant intellectual community, but what it takes to do that might look a little different going forward.

Vir: Noah Smith