AuraTracer智迹闻
中文

EVENT DOSSIER

“ refreshed to 186! Su Weijie’s exclusive analysis: AI has identified key steps, but still cannot overcome the fundamental obstacle of the Twin Prime conjecture”

The AI model GPT-6 provided a key step in the Twin Prime Conjecture, drawing attention to its practical applications and limitations.

2026-09-07 08:00 Models across 2 days 🔥 47.2 heat score
2sources
2days unfolding
47.2heat score
6mentions
SummaryAI generated

In September 2026, the artificial intelligence model GPT-6 provided the correct answer to the twin prime conjecture, attracting widespread attention. Su Weijie explained that AI had provided key steps to solve the problem, but it still couldn’t overcome the fundamental obstacles of the conjecture. In response, renowned mathematician Terence Tao said nothing, pointing out that although AI provided the answer, what was most important might not be the answer itself. Currently, the focus of the incident lies in the actual role of artificial intelligence in mathematical breakthroughs and its subsequent impacts.

Related eventsRELATED EVENTS
Quick factsQUICK FACTS
186Upper bound value
70 millionPrevious record (Zhang Yitang)
246Previous record (Tao Zhexuan)
212 or 188Upper bounds for other AI models
Key entitiesKEY ENTITIES
AnthropicGPT-6GPT-6 AstraOpenAISu WeijieTao Zhexuan

Event frameEVENT FRAME

Research

国内 · 主流媒体across 2 days

Status

The AI model GPT-6 provided a key step in the Twin Prime Conjecture, drawing attention to its practical applications and limitations.

Coverage · reports per dayLANGUAGE SPLIT

Entity relations
GPT-6 × Tao Zhexuan1Anthropic × GPT-6 Astra1Anthropic × OpenAI1Anthropic × Su Weijie1GPT-6 Astra × OpenAI1GPT-6 Astra × Su Weijie1

Integrated timelineUNIFIED TIMELINE

  1. 2026-09-05

    Tao Zhexuan criticizes GPT-6’s new breakthrough in the Twin Prime Conjecture

    Tao Zhexuan points out that GPT-6 directly provides the correct answer, but emphasizes that what is most crucial may not be the answer itself.

  2. 2026-09-07

    Su Weijie explains how AI reduced the upper bound of prime intervals to 186

    Using AI, Su Weijie reduced the upper bound of prime intervals in the Twin Prime Conjecture to 186, adopting the multi-dimensional sieve method to propose the constraint “triple dense divisibility”. Previous records were held by Zhang Yitang (70 million) and Tao Zhexuan (246), and other AI models also reduced the upper bound to 212 or 188. Su Weijie notes that although the value is…

SignalsSIGNALS

Keyword heat
  • Tao Zhexuan1
  • GPT-61
  • OpenAI1
  • Su Weijie1
  • GPT-6 Astra1
  • Anthropic1

All reports (2)SOURCES

量子位 zh 2026-09-05 12:24

“Tao Zhexuan criticizes GPT-6’s new breakthrough in twin primes: A frustrating sight”

Tao Zhexuan pointed out that GPT-6 directly provided the correct answer to the twin prime conjecture, which attracted attention. Although AI can produce results, Tao Zhexuan emphasized that what is most crucial may not be the answer itself. This incident involves the interaction between artificial intelligence and mathematical breakthroughs, and related discussions focus on technical details and subsequent impacts.

新浪科技 zh 2026-09-07 08:00

“ refreshed to 186! Su Weijie’s exclusive analysis: AI has identified key steps, but it still cannot overcome the fundamental obstacle of the Twin Prime conjecture”

Su Weijie used artificial intelligence to reduce the upper bound on the interval between prime numbers in the Twin Prime Conjecture to 186. This research adopted the multi-dimensional sieve method, and AI utilized organizational reasoning and calculations to propose the constraint of “triple dense divisibility” to expand the range of sieve weights. Previously, this record was held by Zhang Yitang (70 million) and Tao Zhexuan and others (246); other AI models have also reduced the upper bound to 212 or 188. Su Weijie pointed out that although there has been a breakthrough in the numerical values, the fundamental obstacle of the Twin Prime Conjecture has not been overcome, and 186 may not be a long-term record. In the future, it is necessary to address the parity issues in the sieve method and strengthen the leading role of humans in problem selection, model construction, and result verification.