ChatGPT cracks 60-year-old unsolved mathematical conjecture: 23-year-old amateur rewrites the history of AI research with a single question

Core conclusion

On April 24, 2026, Scientific American reported a milestone event: 23-year-old Liam Price used ChatGPT (GPT-5.4 Pro) to crack the 60-year-old mathematical conjecture left by Paul Erdős - the lower bound limit problem of the primitive set (primitive set) with a single prompt.

Key Points

• Event Time: 2026-04-24 (Scientific American has published)
• Protagonist: Liam Price, 23 years old, no advanced mathematics training background
• Tools: OpenAI GPT-5.4 Pro (ChatGPT Pro subscription)
• Question: Erdős' original set lower bound conjecture, unsolved for 60 years
• Expert Comment: Terence Tao said that human beings have a "blind spot of collective thinking"

Background and trigger events

Paul Erdős was one of the greatest mathematicians of the 20th century, leaving behind a wealth of unsolved conjectures. One of the questions about the "primitive set" - a set in which no number is divisible by another number - requires proving that the lower bound of its "score" approaches 1.

The difficulty with this problem is that all mathematicians take the same path in the first step, which happens to be the wrong path. Jared Lichtman of Stanford University completed his doctoral thesis on this topic in 2022, but still failed to solve the conjecture.

Liam Price simply typed this question into ChatGPT "on a lazy Monday afternoon", completely unaware of its historical weight.

Key Impact (by Dimension)

Adaptation suggestions

For WayToClawEarn readers (people who make money with AI and automate operations), the inspiration of this news goes far beyond mathematics itself:

1. The "cross-border association" ability of AI tools is being seriously underestimated - This may mean a completely new methodology in automated content production and workflow design. Don’t just think of AI as an “executor”, but also as a “strategic advisor”.

2. The depth of reasoning demonstrated by GPT-5.4 Pro - groundbreaking results can be obtained with just one prompt, indicating that the "one-time success rate" of advanced models on complex problems has been qualitatively improved. For scenarios that require high-quality output (such as complex code, high-converting copywriting), it is worth upgrading to the Pro-level model.

3. The "collective blind spot" is real - your competitors may also be creating content in the same way. Using AI simulation cross-validation, you may be able to discover new content angles or monetization paths.

Task List

• Try GPT-5.4 Pro to improve reasoning capabilities in content production
• Add the "AI strategy recommendation" link to the automation pipeline (don't just let AI execute)
• Learn the idea of "cross-domain formula transfer" in the case study and apply it to your own content selection.

Example: Improve content production efficiency with ChatGPT Pro

# GPT-5.4 Pro

# + + SEO
curl -X POST "https://api.openai.com/v1/chat/completions" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4-pro",
 "messages": [{"role": "user", "content": " AI AI ， SEO 、H2 、 affiliate "}]
  }'

Related extended information

• Scientific American ：Amateur armed with ChatGPT solves an Erdős problem
• HN ：352 ，212
• Erdős Problems

Tool entry

Tool keywords that appear naturally in the text: ChatGPT, OpenAI, GPT-5.4 Pro. The platform side will match the maintained tools library and trigger the tool floating card.

Internal link guidance

• Want to systematically learn to use AI tools to complete complex tasks? See: AI Agent
• Real case: AI can not only solve math problems, but also make money——Claude Code 48：+29，3 $9,000