OpenAI says AI made scientific discoveries
AI makes discoveries in mathematics, computer science, physics, OpenAI says
Tuesday, 27 October 2026
Photo by Vitaly Gariev on Unsplash
Warm-up
- What makes a scientific claim convincing to you?
- How do you usually hear about research news?
- Should companies or universities publish breakthroughs first? Why?
Vocabulary
- conjecture
- an unproven idea in mathematics that people try to prove or disprove
- dimension
- a direction of measurement, such as length, width, height or higher analogues
- rigorous
- extremely careful and exact, following strict rules
- disprove
- to show that an idea or theory is false
- hypothesis
- a suggested explanation that has not yet been proven
- asymmetry
- an uneven or unequal situation between two sides
- consultation
- a discussion to get advice or feedback before deciding something
- breakthrough
- an important and sudden advance in knowledge or technology
Reading
Listen to this story
British EnglishOpenAI says its AI models have made discoveries in mathematics, computer science and physics. One claim is that the models proved the local smoothing conjecture in every spatial dimension, a result researchers in harmonic analysis had only reached in three. Alex Iosevich of the University of Rochester called the news a historic development and predicted major changes in theoretical mathematics, while admitting mixed feelings.
OpenAI also said its models disproved the Exponential Time Hypothesis in complexity theory and gave a mathematically rigorous proof of the existence of quantum chromodynamics. After consultations with three expert “shepherd” groups, the company chose different release paths: the mathematics and computer science groups made the AI-produced data public immediately, while the physics group deferred release to write its own interpretation. OpenAI noted trade-offs, including who gets involved first and the resulting information asymmetry, and asked for feedback to broaden participation. The announcement follows other AI breakthroughs, such as Anthropic’s enzyme said to resemble CRISPR, which drew mixed reactions from scientists.
Comprehension
- Which three broad fields does OpenAI say the discoveries are in?
- What is new about the result on the local smoothing conjecture?
- Who is Alex Iosevich, and how did he respond?
- What does OpenAI claim about the Exponential Time Hypothesis?
- How did the release plans differ between the research groups?
Grammar focus
Adding emphasis - inversion with negative adverbials
Put a negative adverbial first (Never, Rarely, Seldom, Little, Only then/after), then invert: adverbial + auxiliary + subject + main verb. Use for strong emphasis; “Little” goes with know/expect/realise. (Examples constructed.)
- Rarely have researchers announced findings so quickly.
- Only then did a committee publish its notes.
Grammar exercise
Circle the correct option in each sentence.
- Rarely (do models prove / models do prove) such results.
- Never (have researchers seen / researchers have seen) this mix.
- Seldom (data are released / are data released) immediately.
- Only then (a group did publish / did a group publish) notes.
- Little (did reviewers / reviewers did) expect such claims.
Discussion
- How should claims like these be checked before they are celebrated?
- Is deferring a release better than publishing raw data at once? Why?
- What could AI-made proofs mean for human researchers’ roles?
- Do mixed reactions to past AI breakthroughs affect your trust here?
Vocabulary, reading, and exercises with instant feedback
Spotted a mistake in this lesson? Tell us.