Escape Meridian

· Jul 21, 2025

Mehboob Jeelani and Itika Sharma Punit interview Nandan Nilekani for Rest of World about AI’s uses

Markdown · JSON · Original source

Published on 21 July 2025, Mehboob Jeelani and Itika Sharma Punit’s interview presents Nandan Nilekani’s argument that attention should move from building ever-larger models to making AI useful and accessible to people. He says models will become common, and that the right size depends on the task; he points to smaller models trained on specific data as one approach. For India, he highlights spoken AI in a person’s chosen language or dialect: a farmer who can ask a computer a question in Maithili or Bhojpuri and get the right answer would have greater access, he argues. He also describes field-collected speech samples as a way to build language data beyond material scraped from the internet. Nilekani says enterprise adoption is slower partly because companies’ data is not organized for AI, while consumer chatbots are growing and population-scale applications are being explored. His argument is not that AI is free of risks: he names , fake news, and deep fakes. He describes AI as both subject to hype and a powerful capability that could make a difference to lives if applied well and responsibly. The interview does not establish how well the proposed language applications work in practice.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

Rest of WorldArticle · restofworld.orgOpen original source

Original source

· Article

Source details

Original source date: Jul 21, 2025 (2025-07-21T10:00:00+00:00).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-20T19:03:13.917Z. First observed: 2026-11-20T18:26:00.672Z. Retrieved: 2026-11-20T18:27:59.231Z.

Referenced by

Recorded citations in public readings. A reference alone does not establish agreement or influence.

No public references recorded here yet.

Reference data (JSON)

Limits

Text as retrieved on November 20, 2026; originally published July 21, 2025. The interview has been edited for clarity and brevity. Its claims about potential uses and effects are Nilekani’s views; the interview does not provide evidence evaluating those effects in practice.

Open question

What evidence will show whether spoken AI in people’s chosen languages can reliably give useful answers in real-world settings?

Source-stated

Nandan Nilekani argues that discussion should focus less on building models and more on making AI useful to people.

We have to bring the conversation back to making it useful to people.

Source-stated

Nilekani says model size should depend on the use case, with some problems suited to smaller models and others requiring larger ones.

We need to use whatever makes sense for a particular use case.

Source-stated

Nilekani argues that spoken AI in a person’s language could make AI more accessible, giving a farmer who gets the right answer in Maithili or Bhojpuri as an example.

If a farmer in Bihar can speak to a computer in Maithili or Bhojpuri or whichever language and gets the right answer, you have made AI so much more accessible to him.

Source-stated

Nilekani says AI combines hype with a powerful capability that could make a difference to people’s lives if applied well and responsibly.

Certainly, there is a bubble or hype element. But at the same time, it is genuinely a powerful capability which, if applied well and responsibly, with accuracy and scale and speed, can make a difference to lives.

Source and contributor context

Rest of World · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Hallucination

When an AI system states something false or made up as if it were true.

Rest of World

Rest of World reports technology and its social effects outside a US-only frame.

Editorial provenance

Version 1. Publication ID: 57aa3ab5-ddf1-8033-ace1-6bd50d26777a.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-20T18:58:58.128Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: 80223ab9c1f8a27e2ac7f4b7ccf61bf34b981ea4b5312ce41e10adeadbcb0ddc. Source text hash: b8d32c7cb7a16ea34216695ae72b3e2fef2874805424b5caf9e5e43d2486699f.

Revision and source-check history

Version 1 · 2026-11-20T19:03:13.917Z

Mehboob Jeelani and Itika Sharma Punit interview Nandan Nilekani for Rest of World about AI’s uses

Published on 21 July 2025, Mehboob Jeelani and Itika Sharma Punit’s interview presents Nandan Nilekani’s argument that attention should move from building ever-larger models to making AI useful and accessible to people. He says models will become common, and that the right size depends on the task; he points to smaller models trained on specific data as one approach. For India, he highlights spoken AI in a person’s chosen language or dialect: a farmer who can ask a computer a question in Maithili or Bhojpuri and get the right answer would have greater access, he argues. He also describes field-collected speech samples as a way to build language data beyond material scraped from the internet. Nilekani says enterprise adoption is slower partly because companies’ data is not organized for AI, while consumer chatbots are growing and population-scale applications are being explored. His argument is not that AI is free of risks: he names , fake news, and deep fakes. He describes AI as both subject to hype and a powerful capability that could make a difference to lives if applied well and responsibly. The interview does not establish how well the proposed language applications work in practice.

Original source

· Article

Source details

Original source date: Jul 21, 2025 (2025-07-21T10:00:00+00:00).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-20T19:03:13.917Z. First observed: 2026-11-20T18:26:00.672Z. Retrieved: 2026-11-20T18:27:59.231Z.

Limits

Text as retrieved on November 20, 2026; originally published July 21, 2025. The interview has been edited for clarity and brevity. Its claims about potential uses and effects are Nilekani’s views; the interview does not provide evidence evaluating those effects in practice.

Open question

What evidence will show whether spoken AI in people’s chosen languages can reliably give useful answers in real-world settings?

Source-stated

Nandan Nilekani argues that discussion should focus less on building models and more on making AI useful to people.

We have to bring the conversation back to making it useful to people.

Source-stated

Nilekani says model size should depend on the use case, with some problems suited to smaller models and others requiring larger ones.

We need to use whatever makes sense for a particular use case.

Source-stated

Nilekani argues that spoken AI in a person’s language could make AI more accessible, giving a farmer who gets the right answer in Maithili or Bhojpuri as an example.

If a farmer in Bihar can speak to a computer in Maithili or Bhojpuri or whichever language and gets the right answer, you have made AI so much more accessible to him.

Source-stated

Nilekani says AI combines hype with a powerful capability that could make a difference to people’s lives if applied well and responsibly.

Certainly, there is a bubble or hype element. But at the same time, it is genuinely a powerful capability which, if applied well and responsibly, with accuracy and scale and speed, can make a difference to lives.

Source and contributor context

Rest of World · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Hallucination

When an AI system states something false or made up as if it were true.

Rest of World

Rest of World reports technology and its social effects outside a US-only frame.

Editorial provenance

Version 1. Publication ID: 57aa3ab5-ddf1-8033-ace1-6bd50d26777a.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-20T18:58:58.128Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: 80223ab9c1f8a27e2ac7f4b7ccf61bf34b981ea4b5312ce41e10adeadbcb0ddc. Source text hash: b8d32c7cb7a16ea34216695ae72b3e2fef2874805424b5caf9e5e43d2486699f.

passed · 2026-11-20T18:58:58.128Z · anthropic/claude-opus-5.5

Publication: 57aa3ab5-ddf1-8033-ace1-6bd50d26777a

attribution — pass: The reading correctly identifies the piece as a Rest of World interview by Mehboob Jeelani and Itika Sharma Punit with Nandan Nilekani. It attributes the arguments to Nilekani rather than presenting them as established facts.

By Mehboob Jeelani and Itika Sharma Punit

summary — pass: The reading leads with Nilekani's central argument: shift attention from building the largest models to making AI useful. It covers task-dependent model size, spoken AI in local languages, field-collected speech data, enterprise adoption, risks, and his hype-versus-capability view, without flattening his conviction. Hallucination is defined beside the reading. A non-specialist can follow the actor, terms and stakes.

Model building will become very common. I don’t think we should get obsessed with that

date — pass: The originalPublished instant is 2025-07-21, which matches the article dateline and the reading's 21 July 2025. Nothing after that date is presented, and there is no hindsight framing. The page was modified on 2025-07-31, but the cited interview text is plainly part of the original Q&A. This is a publisher page, so no archive disclosure is needed.

21 July 2025 • Bengaluru, India

support — pass: Each claim matches verbatim excerpts. The enterprise point (data not organized for AI, with 'partly' reflecting the legacy-systems context), the field data collection, and the named risks are all source-stated. The limitation correctly notes that the effects in practice are unevaluated, and the question follows from that gap without leading. The 2025-07-31 edit does not appear to affect the cited interview passages.

Their data is not necessarily all organized in a way that AI can readily consume

media — pass: No media is selected, and no verified candidates were supplied, so there is nothing to misidentify.

Challenge or correct this reading

Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.