Written by Brook Schaaf

Old and busted: LLMs. 

The new hotness: SLMs (S for small). 

The AI hype men have lately seen some unwelcome and inconsistent news and pivoted their messaging accordingly. Consider these recent headlines that summarize looming problems:

• MIT Finds 95% of Enterprises See No Return on Generative AI: The promised productivity gains haven’t materialized. 

• The Cost of AI is Creeping Up: AI still loses money on each query (and it can’t make it up in volume, so it is subtly starting to throttle). The Information noted that “AI still feels pricey” despite Sam Altman’s promise that “it does in fact look like we’re about to deliver on intelligence too cheap to meter.”

• AI is a Mass-Delusion Event: Public perception is dimming, at least in some quarters, because of uncanny avatars, stories of AI psychosis, and even suicide assistance, which appears to be an engagement feature, not a bug. 

On the All-In podcast, White House AI and Crypto Czar David Sacks acknowledged AI “fell short of these lofty expectations…a lot of the narratives that were hyped up about imminent doom or imminent utopia…were just massively overhyped. And this is why I think it’s just a very healthy thing.”

Co-host David Friedberg added to the consensus attitude: “There’s a realization now that you need to pair humans and human engineering with the generative [AI] tools…it’s not that you just turn on generative AI and it runs your business for you.” Sacks wrapped it up by saying, “If they used…an SLM approach…then it showed much greater success.” 

This, my friends, is spin. Others may disagree, but as a longtime listener, I would say that this group bought into and spread the hype, but now has to move the goalposts. A funny thing about spin is that it’s difficult to unspin, even if you are the spinmeister. Let’s say their current prognosis is correct.

For me, this feels like a bit of vindication. FMTC has, for its part, been very cautious with AI integrations, not wanting to compromise the quality of our data for cost savings. Our metadata extraction (parsing label elements into discrete fields) has been a good use case, and more are soon to be deployed. But these are still incremental and human-guided improvements.

I suspect this will be the case for many affiliate marketers who are well-positioned in this new paradigm. Merchants, program platforms, and affiliates all sit on small but valuable data sets (commerce content, transactional data, customer information, and, of course, coupons). Our channel is already good with human workflows, at least in business development. 

So, as AI is reshaped into a hand tool, those who know how to work with it can better practice their craft – assuming it remains affordable to use and that it doesn’t cut you in the process.

The post When Smaller Is Smarter appeared first on FMTC.