Torna alle notizie
InnovazioneAI Understanding briefing

OpenAI pubblica 722 articoli di matematica risolvendo 90 principali problemi aperti

OpenAI ha rilasciato un repository GitHub contenente 722 articoli matematici generati da un modello di frontiera interna, che rivendicano soluzioni a 90 dei 500 principali problemi di matematica aperti. La versione include formalizzazioni Lean e dati sulla trasparenza sull'utilizzo del computer.

4 min readRead the primary source
Source-provided image accompanying OpenAI publishes 722 math papers solving 90 top open problems
Documento di origine primariaFonte registrata
Editore
openai.com
Tipo di fonte
Documento principale: un annuncio ufficiale, un documento, un documento o una pagina proprietaria che leggiamo direttamente.
ContestoComprendilo in 60 secondi

Termini chiave

Intelligenza Artificiale (AI)
L’ampio campo dei sistemi di costruzione che svolgono compiti che richiedono il riconoscimento di modelli, il ragionamento, il linguaggio o il processo decisionale.
Citazioni
Riferimenti a passaggi o documenti originali inclusi nella risposta di un modello per supportare le sue affermazioni.
Calcola
Le risorse di elaborazione necessarie per addestrare ed eseguire modelli, spesso misurate in FLOPS o ore GPU.

Cosa è successo

OpenAI published a collection of 722 mathematical papers produced by an internal frontier model, asserting that these results solve 90 of the top 500 open math problems. The materials are hosted in a public GitHub repository, which includes formalizations of many proofs in the Lean programming language to allow for computer-verified checking. The company also released transparency data, including summaries of the model's reasoning, estimations equivalent to roughly three hours of ChatGPT Pro thinking per result, and statistics on attempted problems. This release follows consultations with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study to establish best practices for sharing AI-generated scientific results.

OpenAI has released a comprehensive set of mathematical results generated by an internal frontier model. The publication consists of 722 papers, which the company claims solve 90 of the top 500 open math problems. This release is part of a broader effort to improve how AI-generated scientific results are shared with the academic community.

To ensure rigor and transparency, OpenAI consulted with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study. The results are hosted in a public GitHub repository, which includes protocols for paper revisions and . The repository also contains formalizations of many of the proofs in Lean, a programming language designed for computer-checked mathematical proofs.

In addition to the papers, OpenAI published detailed transparency data. This includes 10 summaries of the model's reasoning process, estimations of spent (averaging the equivalent of three hours of ChatGPT Pro thinking per result), and statistics on the number of problems attempted. The company stated that it will continue to update the repository with more formalizations as they become available.

OpenAI announced plans to fund workshops, conferences, and special programs focused on understanding major results produced by AI. The company emphasized its commitment to responsibly releasing the model that generated these results, noting that continued evaluation of internal frontier models on mathematics and other sciences is essential for developing tools that advance these fields.

Dettagli della fonte: openai.com ↗

Perché è importante

This release represents a significant shift in how AI-generated scientific contributions are disseminated and verified. By providing Lean formalizations, OpenAI enables the mathematical community to rigorously verify the correctness of the proofs using automated tools, addressing common concerns about the reliability of AI-generated mathematics. The transparency regarding costs and reasoning processes offers valuable insights into the efficiency and methodology of frontier models in complex problem-solving. This move could accelerate the integration of AI into mathematical research by establishing a reproducible and verifiable framework for AI-assisted discovery, potentially lowering the barrier for researchers to leverage advanced AI capabilities for tackling long-standing open problems.

The provision of Lean formalizations is a critical step in validating AI-generated mathematics. By allowing proofs to be checked by a computer, OpenAI addresses the inherent skepticism surrounding AI outputs in rigorous fields. This approach sets a new standard for transparency and verifiability in AI-assisted research.

The transparency data regarding usage and reasoning processes offers a rare glimpse into the operational costs and methodologies of frontier models. This information is valuable for researchers and institutions seeking to understand the practical implications of using AI for complex scientific problems.

This release could significantly impact the mathematical community by providing a large corpus of potential solutions to long-standing problems. If verified, these results could accelerate progress in various areas of mathematics and demonstrate the practical utility of AI in advancing human knowledge.

Interactive Mechanism

Meccanismo interattivo: come funziona realmente

Esplora la tecnologia alla base di questo sviluppo in modo interattivo.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Verifica concettuale interattiva+10 Points
AI Models Explained Quiz

Which component of an AI application is the machine-learning model itself?

Cosa guardare dopo

The mathematical community's verification of the 90 claimed solutions, particularly through the provided Lean formalizations. Future updates to the repository as more proofs are formalized. The specific details of the workshops and conferences OpenAI plans to fund to discuss these results. The eventual release of the internal frontier model that produced these results, which OpenAI states is being prepared for responsible release.

The primary focus will be on the independent verification of the 90 claimed solutions. The mathematical community will likely scrutinize the Lean formalizations to confirm the correctness of the proofs. Any discrepancies or errors found during this process will be significant for the credibility of AI-generated mathematics.

OpenAI's plans to fund workshops and conferences indicate a strategic effort to engage with the academic community. The outcomes of these events, including any new insights or collaborative efforts, will be important to monitor.

The eventual release of the internal frontier model is a key development to watch. OpenAI has stated that it is working to responsibly release the model, which could have broader implications for the availability of advanced AI capabilities in scientific research.

Guide e quiz correlati

Lo hai trovato utile?