India AI Digest — Wednesday, March 22, 2023

• Microsoft Research publishes "Sparks of AGI" paper on early GPT-4. Capability survey across maths, code, vision, medicine, law; the AGI framing is the authors' position, not a measurement.

---

Archive entry. Written retrospectively in April 2026 as part of historical backfill. The contemporaneous voice is preserved as if filed on the event date.

---

Microsoft Research publishes "Sparks of AGI" paper on early GPT-4

Microsoft Research published a paper titled "Sparks of Artificial General Intelligence: Early experiments with GPT-4" on March 22, 2023. The paper surveys an early version of GPT-4 across mathematics, coding, vision, medicine, law, and other domains, and characterises the breadth of capability as early signs of artificial general intelligence.

What this means. The paper is a framing artifact more than a research result. The empirical content is a survey of GPT-4 behaviour across many tasks; the AGI characterisation is a position taken by the authors, not a measurement. Researcher peers have argued the framing reads marketing-adjacent for what is otherwise a useful capability survey; that critique is part of the contemporaneous record.

For readers building practical systems, the operational question is separable from the philosophical one. The capabilities catalogued are the ones to design against; the framing is not load-bearing.

India angle. Limited direct read. The paper shapes the 2023 Indian discourse around AGI timelines and the implicit pressure on Indian foundation-model strategy to "catch up" to a moving frontier whose actual scope is what the paper documents minus the framing imposed on it. For Indian foundation-model planning, the productive read is to treat the catalogued capabilities as the reference point and the AGI framing as separate.

What this is not. Not a measurement of AGI. The paper does not define AGI operationally; it asserts the framing on the basis of broad capability across domains.

Source: Microsoft Research / arXiv, March 22, 2023. → link

Confidence: high — paper existence and contents verified; the AGI characterisation is the authors' framing, surfaced as such.