How accurate have Ed Zitron's AI skeptic predictions been?
TL;DR Highlight
Dan Luu checked Ed Zitron's 2024–2025 AI skeptic predictions and found most were wrong.
Who Should Read
An LLM service developer who must make adoption or investment decisions while AI bubble articles circulate
Core Mechanics
- Most Zitron predictions were judged Wrong
- Dan Luu listed predictions from Feb 2024 to Nov 2025 and graded each one.
- Claims like 'AI has peaked' and 'we're out of data' were all marked Wrong.
- Predictions after April 2025 were mostly unfalsifiable or still unresolved.
- Meta, Google, Microsoft aren't dying
- Meta revenue grew from $135B in 2023 to $201B in 2025.
- Alphabet went from $307B to $403B and Microsoft from $228B to $305B.
- So the 'they don't know how to grow and shove in AI' logic doesn't hold.
- His numbers don't support his argument
- He used third-party estimates like Similarweb for Facebook MAU decline.
- He skipped Meta's own figures and revenue or profit numbers.
- The author says the numbers don't connect to a coherent argument.
- Company predictions kept missing
- He called Gemini's 500M-user goal unrealistic, but it hit 750M.
Community Reactions
- Commenters project their own predictions
- One comment says many people judge their own predictions rather than Ed's literal statements.
- Dan's essay compares the literal text against measurable outcomes.
- Counterview: 'dying' means worse products
- Some argue that in Zitron's 'rot-economy' framing, dying means products getting worse and users leaving.
- They say judging only by revenue and profit misses that meaning.
- Growth may come from circular financing
- A comment argues growth isn't a valid rebuttal unless circular AI financing is separated out.
- It says this is easier to see for Nvidia than for the hyperscalers.
- An experience of helping Zitron and being misused
- A commenter says they once explained LLM serving topics like benchmarks and model caching to Zitron.
- They stopped after their advice was twisted into the most cynical reading.
How to Apply
- Trace the numbers in AI hype or doom articles
- Pick one article you rely on today and follow its figures to the primary source.
- Note whether each is a third-party estimate or a company filing.
- Log predictions with dates and criteria
- Write down your team's AI assumptions, such as model plateau or price drops, with dates and verification conditions.
- Mark each as right or wrong every quarter.
- Separate capability outlook from business outlook
- Split 'is AI useful' and 'do AI companies make money' into separate sections of adoption docs.
- Put vendor dependency risk under the second section.
Try It
# Paste into Claude or ChatGPT as-is (a simplified version of Dan Luu's prediction audit)
Extract every verifiable prediction from the following article or statement.
{{article or statement text}}
For each prediction, make a table: quoted text / date made / verification deadline / numbers and sources cited / whether the numbers actually support the conclusion.
At the end, list separately the unfalsifiable predictions and the places where numbers don't connect to the conclusion.
Next step: run the same audit on an optimistic article and compare.Terminology
Related Papers
The Judge Can Debias Itself: Awareness and Self-Debiasing for Reliable LLM-as-a-Judge
Warning an LLM judge about bias or letting it self-critique improved 20 of 21 model–bias combinations.
90% of US working hours are spent on tasks not covered well by AI benchmarks
Mapping 29 benchmarks to job tasks shows 90% of US working hours lack good coverage
Postgres 19 delayed after AI reviewers find more bugs than humans can fix
LLM code review surfaced so many bugs that Postgres 19 slipped to November and 8-10 features were reverted
Show HN: Gemini 4 Argon plays StarCraft
LLMs wrote StarCraft bots in one hour; GPT-6 Astra scored highest.
A Plausible Explanation Is Enough to Mislead Multimodal Judges
Adding plausible explanations to fixed answers cut MLLM judges' wrong-answer rejection by 12.7–13.8 points.
Neural networks are clean
Looking at neurons directly shows most have one main job, challenging superposition.