A back-of-the-napkin status check on AI hype

(from the perspective of K12 education)

As we close out the year, some reflections about the state of play in AI development and what it means for the education system. 

There are a few weeks left in the year for OpenAI to stun us with a game changing release of GPT-5, but it seems likely that we will close out 2024 with mostly incremental progress on the GPT-4-level models that stunned everyone in 2023. There are rumors that OpenAI and other frontier model developers are stalling out despite massive increases in GPU capacity. (Though, notably, Sam Altman has denied this.)

I’ll write about what it might look like if it turns out we are on the optimistic timeline, but I want to indulge in the basis for what seems like the more probable scenario, one where the break-neck pace of development does indeed slow down in 2025… 

If progress in foundation models is stalling out or plateauing, that will pose some short-term problems for the AI zeitgeist that has captured everyone’s hearts and minds (and investors’ billions). That’s because LLMs still aren’t good enough for general applications. And they’ve lost their ability, for the moment, to stun us with their progress.

Apple Intelligence has been a disappointment with enhancements that are at best occasionally cool or useful and at worst a gimmick. ChatGPT, Claude, Gemini and Llama are improving but all still hallucinating at unacceptably high rates and simply can’t be relied upon. Companies that were early adopters of LLM-powered chatbots have largely retreated in the face of embarrassing glitches. (My favorite is the dealership chatbot that was tricked into selling a Chevy Tahoe for $1).

It is undeniably cool to have a chatbot that can produce a marketing plan, draft social posts, blog articles, and a content calendar all customized to your specifications and context. But we are at the frontiers of a still budding and immature technology: The marketing plans are totally run of the mill and lack originality. And the writing, while not bad is wholly unoriginal and generally fails to match my voice. 

And speaking of originality, AI-generated art and music may be more accessible than ever, but it seems like we are starting to see the limits of “creativity” that these tools can offer — frankly it feels like all AI art looks the same, and any initial cool-factor has been replaced with eye rolls at the utter sameness and unoriginality. 

So, the multi-billion dollar questions: Did we get ahead of ourselves with talk of the AI revolution? Are we on the tail end of a growth curve that is leveling out? Where are we?

First of all, seriously no one knows. The smartest engineers at the front lines of development disagree. It’s also very hard to disintermediate informed optimism from biased hot takes from folks with a massive vested interest in everyone (in particular, venture capital investors and the stock market) being utterly convinced that we are on a hockey stick growth curve lest you miss out. 

If you’re an investor or AI company founder, a slow down or end of the AI hype cycle will have immediate and significant implications if follow-on capital dries up and the Cambrian explosion of AI startups succumbs to an extinction event. 

But if you are a K12 leader trying to figure out what this all means for your work preparing the next generation, I’d argue it doesn’t really matter. In the K12 system, we operate on longer time horizons than most investors looking to make some quick returns (paper or otherwise). A 6th grader has six to ten years ahead of them before their first job. At such a timescale, I feel extremely confident in the prediction that AI will dramatically disrupt and reshape our workforce and society at large. It might be hard to see AI generated art changing hollywood in the next year, but likewise seems inevitable over the next decade. 

Comparisons to the internet are perhaps apt, and while we don’t know for sure if we are comparably in the late 80s, mid 90s, or early 2000s, it’s certainly the case that even in the depths of the dot com crash, it would have been utterly wrong to ignore the importance of that technology. 

And at the same time, if you could teleport back to that era, you wouldn’t necessarily aggressively invest in the earliest iterations of the internet. 

This is why we often advise school and district leaders not to rush into big implementations and contracts with AI vendors. A lot of these companies, even highly valued ones, might not exist five years from now. Our advice: focus on investing in professional learning and building teacher and staff capacity to better understand AI so that your human capital is best positioned to adapt to the coming wave of developments and ultimate maturation of the technology to a point where it does make sense to deploy broadly in some form. 

So all I’m saying is, I’d bet on the Internet, but I wouldn’t put all my chips on Pets.com.

In my next writeup, I’ll talk about why there is actually more urgency than ever when it comes to AI Readiness, even if the major disruption from AI is still 5-10 years away.