Generative AI could be a dud, so we shouldn't build around the premise that the tech is world-changing, which in hindsight may turn out to have been unrealistic
With the possible exception of the quick to rise and quick to fall alleged room-temperature superconductor LK-99 … Mastodon: @carnage4life@mas.to , @peter@thepit.social , and @baldur@toot.cafe . Bluesky: @epro.social X: @benbajarin , @bhaggart , @garymarcus , @fchollet , @garymarcus , @elkeschwarz , @benedictevans , and @garymarcus Mastodon: Dare Obasanjo / @carnage4life@mas.to : What if generative AI turns out to be an overhyped dud? I think that question depends on how you define success. — This article has a few examples including generating trillions in revenue or helping OpenAI justify its $29 billion valuation. … Peter Krupa / @peter@thepit.social : this is very good. imo the shine is off AI in some significant ways, and it's good to be skeptical of the self-serving claims of tech companies (remember when self-driving cars were just a couple years away??) https://garymarcus.substack.com/ ... Baldur Bjarnason / @baldur@toot.cafe : “What if Generative AI turned out to be a Dud?” — Like I've been saying for months, the functionality for LLMs generally isn't there and what positive functionality is there is overestimated by investors, executives, and legislators. https://garymarcus.substack.com/ ... Bluesky: Emil Protalinski / @epro.social : Generative AI is well on its way to the Peak of Inflated Expectations on the Gartner Hype Graph. [embedded post] X: Ben Bajarin / @benbajarin : A good counterargument to stay balanced in your view of GenAI. I maintain the value here is workflow automation. AI has the potential to eliminate tedious workflows (save us time) and make “power user” features available to everyone. @bhaggart : This is pretty much where I've ended up. LLMs are hallucinations all the way down; they aren't a path to AGI. Its use cases, like blockchain's, will be very limited, in this case to coding (maybe) and generating low-quality “content” (which will still pollute our info ecosystem). Gary Marcus / @garymarcus : Could be a juicy thread; @fchollet challenges me and I reply. I agree with what he says but stand my ground relative to the immense hype that surrounds us. François Chollet / @fchollet : Is generative AI a dud? Well, depends on what your expectations were. If you had no expectations, then it's an insane success story: new tech suddenly unlocks a multi-billion dollar niche industry that enables a handful of new startups to rise and make it big. Gary Marcus / @garymarcus : No doubt generative AI is great for coders, but will other people maintain their monthly subscriptions after the novelty wears off? Or are we building our entire approach to the future around an overhyped fad? Elke Schwarz / @elkeschwarz : This by @GaryMarcus is on point: “But what has me worried right now is [...] that we are building our entire global and national policy on the premise that generative AI will be world-changing in ways that may in hindsight turn out to have been unrealistic https://open.substack.com/... Benedict Evans / @benedictevans : I've been using ChatGPT 3.5 & 4 for half a year and really, I haven't worked out anything that's useful for me. It may be the new PC, but we don't have the spreadsheet or word processor yet. I'm sure there will be, but they haven't been built yet... Gary Marcus / @garymarcus : What if generative AI turned out to be a dud? At my (free) newsletter that I can't properly link here, I discuss some economic and geopolitical implications of this that X's algorithm doesn't seem to want you to read about. [image]
Context & Ripple Effects
The debate sits between skepticism about model reliability—earlier criticism of GPT-2’s superficial knowledge and concerns that ChatGPT could generate confident but unreliable output—and enthusiasm prompted by GPT-4’s visible leap forward. The disagreement is therefore not whether the systems can produce striking results, but whether those results justify treating AGI-scale transformation as a planning baseline.
It matters because the case for broad deployment depends on benefits holding up against quality, legal, and operational constraints. Related coverage later tied generative systems to copyright exposure around image generation, reinforcing the distinction between compelling demonstrations and durable business value.
First-order effects
- The commentators’ warning challenges companies and policymakers that are making product, hiring, or infrastructure commitments on the assumption that generative AI will be universally transformative.
- AGI-oriented narratives face a higher burden of proof: useful outputs must translate into reliable, repeatable value rather than attention or novelty.
Second-order effects
- Vendors and buyers are pushed toward narrower evaluation criteria—task performance, failure handling, and economic return—rather than treating model scale or generality as sufficient evidence of value.
- Copyright and reliability concerns can raise the cost of deploying generative tools, favoring uses with clearer human review and accountability.
Third-order effects
- If capability gains remain uneven, the market may sort into specialized, workflow-bound AI products rather than a single AGI-led replacement cycle.
- The durable competitive advantage may shift from access to a model toward governance, distribution, and the ability to validate outputs in real operating contexts.
The trend: This is part of a shift from generative-AI spectacle and AGI expectations toward evidence-based adoption centered on reliability, economics, and governance.