ChatGPT has been a dumpster fire for weeks. I have a theory why, illustrated by Facebook’s cautionary tale.
Facebook’s DAU obsession
From inception through the mid-2010s, Facebook was obsessed with one metric: Daily Active Users (DAU). The whole company optimised for it. Makes sense, right? More DAU = more ad revenue.
But here’s what happened.
2016: Researcher Monica Lee discovered that 64% of extremist group joins came from Facebook’s own recommendation algorithms. The platform was actively promoting hate groups through “Groups You Should Join” features.
2017: Chief Product Officer Chris Cox formed a task force to study engagement vs. polarisation. Their conclusion? The two were “inextricably linked.” Reducing hate and misinformation would hurt engagement, and revenue.
The result: multiple solutions were proposed and killed for being “antigrowth.”
The ChatGPT parallel
I suspect ChatGPT is falling into the same trap. Somewhere, a metric is driving product decisions. Maybe something engagement-based like “prompts per session,” or usage-based like total number of prompts.
The symptoms:
- Answers that used to work in one prompt now take 3–4 attempts.
- Endless positive fluff (“Great question! You’re really zoning in on…”) instead of direct answers.
- Confident lies followed by “Oh wait, that doesn’t actually exist.”
My engagement is up. My satisfaction is down. But the dashboard probably shows I’m a “power user.”
The lesson
Single-metric optimisation is dangerous. Unintended consequences are inevitable and must be considered and managed.
It took a long time for Facebook to learn this. I hope ChatGPT figures it out faster.