Gary Marcus: the case that generative AI has been a net detriment is getting harder to rebut
Gary Marcus · rss · 2026-10-25
Gary Marcus revisits his March 2023 debate on whether ChatGPT does more harm than good and argues the negative ledger is now serious.
The negatives
- Nonconsensual deepfake porn, surging phishing scams, wholesale disinformation
- "Cognitive surrender" and democratic threats from subtle government influence on LLMs
- Generative AI as a surveillance tool; IP theft hitting artists and writers (employment impact milder than feared)
- A cluster of hacks dubbed "The Hackening" / "Black October" — Salesforce CEO Benioff called it a "Code red"; Anthropic's Jason Clinton thinks the bad stretch may last only a year or two
Positives in doubt
- Apollo chief economist Torsten Slok argues many promised benefits simply aren't materializing
- Marcus concedes friends have gained a lot (mainly coding), but says it's "not hard to argue" genAI has on balance been a net detriment — and it could still drag down the whole economy
He returns to Open to Debate on Tuesday in Baltimore with Cathy O'Neil; edited versions will appear on YouTube and NPR affiliates.
More from AGI Musings
- Beff Jezos: e/acc is memetically optimal because knowing the meme boosts fitness — beffjezos · 2026-10-25
- Kolmogorov-LLM complexity: rate a talk by the prompt length needed to recreate it — xennygrimmato_ · 2026-10-25
- Varick founder: most 'AI transformation' firms just sell skill files and prompt workshops — vasuman · 2026-10-25
- The upside of AI-assisted research: failing faster at a higher level — mboehme_ · 2026-10-25
- Inside Berkeley's 'Cult of p(doom)': the viral AI Safety exposé, fully digested — AIFlow_ML · 2026-10-25
- 'AI Safety Is Mostly A Sex Cult In Berkeley' Substack sets AI Twitter ablaze — AIFlow_ML · 2026-10-25