AI, Reading Decline, and Compute Demand
Exponential View (Azeem Azhar) · rss · 2026-07-12
This in-depth article connects several seemingly disparate topics: human cognitive abilities in the AI era, educational incentives, and the undiminished demand for compute power.
1) Reading and Learning Abilities Are Declining
The article cites an economics course at Brown University: suspecting students used ChatGPT, the professor switched to closed-book exams. As a result, 56 out of 59 students saw their grades plummet, with the worst dropping by 100%. The author argues this reflects a mindset where students care more about grades than mastering knowledge.
Data also shows that the percentage of Americans reading for pleasure dropped to 16% in 2023, down from 28% in 2004. The author stresses this trend predates generative AI; it's a long-term decline in reading and attention spans, not just a ChatGPT problem.
2) AI May Amplify Cognitive Divide
Drawing on David Brooks' perspective, the author notes that when accessing "smart answers" becomes cheap, raw IQ matters less than the willingness to engage in difficult thinking. Reading long texts, building personal knowledge maps, and taking manual notes remain essential for deep understanding. While AI can generate materials that look "almost like research," it requires the user to know what and how to ask.
3) Real-World Observations
The article also notes:
- LinkedIn is flooded with AI-generated content, while Substack is less affected; the author even built a Chrome extension to hide AI content.
- The University of Chicago Law School is piloting device-free classrooms while teaching students to use AI effectively.
- A VC noted that an AI tool might be "better at my job than I am," but using it alone still only makes one an "average investor."
4) Compute Demand Remains Strong
The final section shifts to infrastructure: citing Silicon Data's H100 contract index, the price rebounded from around $1.70/hour last October to roughly $2.35/hour, a 38% increase, with spot prices up 10% this year. This indicates robust GPU demand without the anticipated cooldown.
The article concludes: AI is more than a tool upgrade; it threatens to divide society into two groups—those willing to continuously think, read, and train their cognition, and those who merely consume answers.
More from AGI Musings
- FactoryAI’s Enoreyes says model distillation is basically unstoppable — LangChain · 2026-07-21
- Andrew Blumberg says formalization without interpretability is not science — AlexKontorovich · 2026-07-21
- Ken Ono says AI is forcing mathematicians to rethink how discovery works — soumitrashukla9 · 2026-07-21
- Open-source labs could distill a state-of-the-art model to 32GB or 80GB VRAM, the post argues — bookwormengr · 2026-07-21
- Two US companies are now using superintelligence to speed up the next generation of models — yacineMTB · 2026-07-21
- MIT Sloan says information, national security and finance are most exposed to AI — Exp_Mark · 2026-07-21