Ray Kurzweil

Factual · AI capabilities

“That was not true a year ago. But that’s true today.”

Ray Kurzweil, speaking on Moonshots · 4 June 2026 · This Week on Moonshots: Ray Kurzweil (Metatrends (Diamandis))

Large language models being roughly 50% better than human doctors at diagnosing conditions and recommending treatments was not true a year ago but is true today.

Mostly false False in substance, right in a detail.

The second judge would not resolve this claim; the verdict shown is the first judge's.

What would count as true or false

Judged as of 4 June 2026. The claim has two parts: (a) as of June 2026, large language models outperform human doctors at diagnosis and treatment recommendations; (b) this was not the case around June 2025. True if published studies or benchmarks from around mid-2026 show LLMs beating physicians at both diagnosis and treatment recommendations, and if evidence from about a year earlier did not show this. Mostly true if both parts broadly hold but the timing is somewhat off, or the superiority is shown in most but not all of the claimed tasks. Mixed if only one part holds: LLMs outperform doctors in 2026 but already did so in 2025, or they lead on diagnosis but not on treatment. Mostly false if the superiority rests only on narrow benchmarks (for example vignettes or multiple-choice tests) and is stated as general, or if evidence of superiority already existed well before mid-2025 and the 'not true a year ago' part fails. False if the evidence shows LLMs do not outperform doctors at these tasks as of June 2026.

Evidence

1. Independent and collaborative performance of large language models and healthcare professionals in diagnosis and triage - npj Digital Medicine npj Digital Medicine (Nature Portfolio) · 6 February 2026

   “The relative diagnostic accuracy of LLMs versus HCPs progressively improved from 0.89 (95% CI, 0.79–1.00) for top-1 to 0.91 (0.83–1.00) for top-3, 1.04 (0.89–1.22) for top-5, and 1.17 (0.87–1.57) for top-10 diagnoses, with significant model variability.”

2. Independent and collaborative performance of large language models and healthcare professionals in diagnosis and triage - npj Digital Medicine npj Digital Medicine (Nature Portfolio) · 6 February 2026

   “No significant differences were found between LLMs and HCPs for top-5 (1.04 [0.89–1.22]; I2 = 96.7%) or top-10 (1.17 [0.87–1.57]; I2 = 95.8%) diagnostic accuracy.”

3. Evaluation and mitigation of the limitations of large language models in clinical decision-making - Nature Medicine Nature Medicine (Nature Portfolio) · 4 July 2024

   “We show that current state-of-the-art LLMs do not accurately diagnose patients across all pathologies (performing significantly worse than physicians), follow neither diagnostic nor treatment guidelines, and cannot interpret laboratory results”

4. Study Suggests AI Is Good Enough at Diagnosing Complex Medical Cases To Warrant Clinical Testing Harvard Medical School (with Beth Israel Deaconess Medical Center communications) · 30 April 2026

   “A large language model (LLM) outperformed physicians across many of these tasks, including making emergency-room decisions based on the available information, identifying likely diagnoses, and choosing the next steps in management”

Reasoning

A Harvard Medical School release from April 2026 reports a study in which an LLM outperformed physicians on many reasoning tasks, including diagnosis and choosing next steps in management. The same release frames this as reason for clinical testing, not proof of real-world superiority (E4). A February 2026 meta-analysis reported LLM-to-professional diagnostic accuracy ratios of 0.89 for top-1 (95% CI 0.79–1.00) and 0.91 for top-3, and found no significant differences for top-5 (1.04) or top-10 (1.17) (E1, E2), so the evidence does not show general superiority. A 2024 study found LLMs did significantly worse than physicians and did not follow treatment guidelines (E3), which fits the "not true before" part, but nothing here covers mid-2025 specifically. Overall, the evidence shows LLM superiority in specific studies (E4) but not general superiority over doctors at diagnosis and treatment, which the claim states as general.

Cite this claim

Ray Kurzweil, “Large language models being roughly 50% better than human doctors at diagnosing conditions and recommending treatments was not true a year ago but is true today.” (said 4 June 2026). Verdict: mostly false. Reliability API, claim c_93019ae2d958, method 0.1.0+28da144b. https://epistacklabs.com/claims/c_93019ae2d958