OpenAI releases 700 math papers. AI takes over science? | You&AI

Summary
In this episode of TVP World’s tech show You&AI, host Ayşe Abacı covers recent artificial intelligence headlines before examining how frontier AI models are impacting mathematics and scientific research. The broadcast includes a news report featuring interviews with mathematicians Bartosz Naskręcki and Teodor Buchner, followed by an in-studio discussion with Marcin Napiórkowski, Associate Professor of Physics at the University of Warsaw, on automated proof verification and OpenAI’s release of hundreds of AI-generated math papers.

What is shown

• [00:15] Brief news segment showing prompts and AI companion avatar robots while discussing gender pay disparities in valuations of AI agents.
• [00:36] Video footage of Italian Prime Minister Giorgia Meloni and parliamentary protests regarding deepfake imagery.
• [00:55] Photos of Donald Trump, Elon Musk, and the Grok interface discussing political queries.
• [01:26] Smartphone screen showing ChatGPT interface during a discussion of teenage user safety features.
• [01:53] Screen capture of an OpenAI research publication titled “Sharing AI progress in mathematics” (dated October 6, 2026) detailing solutions to 377 problem families formalized in Lean.
• [02:16] Video call interview with Bartosz Naskręcki (Adam Mickiewicz University Poznań) over b-roll of workstations running ChatGPT, Gemini, and Claude.
• [02:40] Video call interview with Teodor Buchner (Warsaw Technical University) discussing LLM answers in theoretical physics.
• [02:53] Equations of the Euler and Navier–Stokes systems overlaid with 3D fluid simulation graphics.
• [03:32] B-roll of a humanoid robot attempting to lift a barbell during a competition before falling backwards.
• [04:07] Studio discussion between Ayşe Abacı and Marcin Napiórkowski covering Lean formal verification, arXiv preprints, and whether automated proof generation will replace human mathematicians.

Claims & numbers

• The presenter reports that a study found users paid female-presenting AI agents approximately 10% less than male counterparts for identical tasks [00:20].
• The presenter states that Time magazine reported Donald Trump spent hours conversing with xAI’s Grok chatbot about geopolitical scenarios involving Venezuela [00:55].
• The report highlights OpenAI’s publication of solutions to 377 mathematical problems solved by an internal frontier model and formalized in the Lean proof assistant [02:02].
• Bartosz Naskręcki claims that AI performance on FrontierMath benchmarks advanced over the past year from 0% to completely solving all problems in the evaluated set, including his own [02:18].
• Teodor Buchner claims that existing large language models answer his specific physics domain questions accurately 99% of the time [02:43].
• Marcin Napiórkowski notes that OpenAI released over 700 papers generated by AI resolving mathematical conjectures verified through Lean code [06:21].
• Napiórkowski states that mathematical reasoning capabilities of AI models experienced a dramatic leap in just the preceding three to four months [04:38].

Notable quotes

• Bartosz Naskręcki [02:16]: “I have seen over the year a progression from 0%... up to the level that, including my own problem, all of the problems were completely solved.”
• Teodor Buchner [03:43]: “It can build reasoning on analogy, but if there is no clear analogy, then it cannot.”
• Marcin Napiórkowski [11:40]: “For me this is a completely new phenomenon... I would call it a phase transition that we are now at in mathematics.”

Assessment
This is a standard broadcast television news and interview program reviewing third-party research reports and institutional announcements. No direct live benchmark tests or software operations are conducted in real time on set; statements and conclusions rely on published papers, external benchmarks, and guest analysis.

Described by gemini-3.8-flash on 2026-10-10 from the video’s audio and frames.