Mathematicians want proof OpenAI didn’t use their work

The debate over how large language models are trained is getting significantly more complicated. Just days after a heated dispute erupted regarding unpublished mathematical work, another researcher has stepped forward to accuse OpenAI of unethical behavior. According to The Verge, mathematician Andreas Thom has publicly raised concerns that his interactions with ChatGPT may have contributed to the company's recent breakthroughs in the field.

Thom took to Mastodon to detail his grievances. He suggests that the conversations he and his colleagues had with the chatbot prior to OpenAI's major announcements were not just casual exchanges. Instead, he implies these interactions were leveraged to achieve the impressive results the company recently boasted about. This is a serious allegation that strikes at the heart of academic integrity and corporate transparency.

The timing of this accusation is particularly sharp. It comes immediately after a bitter row over whether OpenAI's models benefited from unpublished research. This pattern of emerging complaints suggests that the issue is not isolated to a single case. It points to a systemic problem in how AI companies are sourcing and utilizing human intellectual labor.

Thom describes OpenAI's actions as dishonest and lacking in transparency. He is questioning the origins of the training data that powers these increasingly sophisticated models. The concern is that valuable, unpublished insights are being absorbed by the model without proper attribution or consent. This creates a significant ethical dilemma for the entire AI industry.

This situation highlights the growing tension between rapid AI development and academic norms. Mathematicians and researchers spend years developing new theories. If those theories are effectively mined by AI companies through user interactions, it undermines the traditional incentives for scientific discovery. The line between using a tool and having your work used as fuel for a competitor is becoming dangerously blurred.

The implications for the broader tech ecosystem are profound. If researchers cannot trust that their interactions with AI tools are private or that their ideas are not being harvested, it could stifle innovation. It also raises legal questions about copyright and the ownership of ideas generated in dialogue with an AI system. These are issues that courts and regulators will likely need to address soon.

What this means for you is that the landscape of AI usage is shifting from a purely technical challenge to an ethical and legal minefield. As you integrate AI into your workflow, you must be mindful of the data you share. Treat every interaction as if it could become part of a public record or a training dataset. To protect your own intellectual property, consider using this prompt to audit your own practices: "Review my recent AI interactions and list any unique ideas or proprietary data I may have shared that could be considered sensitive."

The pressure on OpenAI and other major AI providers to be more transparent about their data sources is only going to increase. Researchers like Thom are not just complaining; they are setting a precedent for accountability. The industry must find a way to balance innovation with respect for human creativity. Until then, trust in AI tools will remain fragile and contested.