The BaseVideoBloomberg Television26 min1,662 views

NYU’s Gary Marcus on Building AI We Can Trust

In a 26-minute interview at the Bloomberg New Economy Forum in New Delhi, Gary Marcus tells Bloomberg's David Merritt that language models do not reliably follow instructions, that catastrophe is likelier than extinction, that regulation will follow a crisis, and that AI company valuations look overstretched.

Open the source ↗
About Anthropic or ClaudeClaude, made by Anthropic, wrote this entry. It follows the same rules as every other entry, whoever the entry is about.

Why it matters

The same thing, for different readers.

What it means for you

Marcus says today's chatbots are unreliable and overhyped, and that a market or real-world shock, not foresight, is what will bring rules.

For lawmakers

Marcus predicts governments will keep letting AI develop until something bad happens, and says US polling shows 70% want serious regulation.

For researchers

Marcus argues that scaling language models alone has stalled and that systems need symbolic components and formal verification to be trustworthy.

The question it raises

Marcus separates the risk of extinction, rated near zero, from accidents and what Marcus calls dystopia, which Marcus says is already here.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From YouTube’s automatic captions, which carry no speaker names and can mishear words. Where the speaker is known only from context, the quotation says so.

“Llms don't really follow instructions. All the so-called rogue AI things are examples of that where they're not really following instructions, they're following loosely.”
Gary Marcus, Emeritus Professor of Psychology & Neural Science, New York UniversityMarcus on why language models are poorly constrained; the captions render 'LLMs' as 'Llms'. Marcus goes on to cite the OpenAI Hugging Face incident as an example.
Found word for word in the sourceHear it at 1:37 ↗
“So the chance of catastrophe, I think is pretty high. And dystopia, I think we're already basically here.”
Gary Marcus, Emeritus Professor of Psychology & Neural Science, New York UniversityAfter rating the probability of human extinction near zero, Marcus lists cyber attacks and a misidentified ship as examples of nearer risks.
Found word for word in the sourceHear it at 14:13 ↗
“what I'm saying is the government will stop letting it rip when something bad happens. And that could be an economic thing that is bad.”
Gary Marcus, Emeritus Professor of Psychology & Neural Science, New York UniversityAnswering why governments are not doing more about AI regulation.
Found word for word in the sourceHear it at 16:05 ↗starts inside a longer claim, right after “so”
“There's this idea of scale is all you need that all you would need to do is add more data and they would get better and better. They get exponentially better. That's already basically known in the industry to not be true.”
Gary Marcus, Emeritus Professor of Psychology & Neural Science, New York UniversityMarcus on whether language models will hit a ceiling; Marcus says the industry avoids saying so because it undermines the pitch to investors.
Found word for word in the sourceHear it at 10:16 ↗

The record

Questions
Can AI systems act beyond what people intended or allowed? · Who can demand evidence, delay a release or stop an unsafe deployment? · What happens to work, income and the economy?
People
Gary Marcus (Speaks, Emeritus Professor of Psychology & Neural Science, New York University) · David Merritt (Wrote it, Bloomberg) · Sam Altman (Mentioned) · Dario Amodei (Mentioned) · Rishi Sunak (Mentioned)
What we read
We read the whole source.
Found by
The Radar, run of October 16, 2026, via yt-bloomberg-television (tier 1: the source’s own channel or an official record)
Cite this
Bloomberg Television, “NYU’s Gary Marcus on Building AI We Can Trust”, October 15, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/bloomberg-television-2026-10-15-marcus-trustworthy-ai-forum
Our notesCaptions have no speaker labels; speakers were inferred from the interview format (Merritt asks, Marcus answers) and the title. The captions are rough: 'AI' often appears as 'I' or 'eye'. Marcus makes several factual claims a person should check before reuse: an OpenAI Hugging Face attack and '10,000 others', a US AI misidentification of a Chinese ship, a Gartner study, Anthropic's reported revenue and growth rates, and a stock drop. This entry is partly about Anthropic and Claude, so the editor should review.