RISS: “Is Your LLM Overcharging You?” by Stratis Tsirtsis (Hasso Plattner Institute)

[Krikamol Muandet]

We’re thrilled to announce the upcoming seminar at the Rational Intelligence Seminar Series (RISS), on Janruary 14, 2026. RISS seeks to advance the understanding of rationality, efficiency and reliability in machine learning systems. These seminars serve as a forum for discussions and dissemination of results.

Join us to engage in lively discussions in the session, “Is Your LLM Overcharging You?” delivered by Stratis Tsirtsis – Postdoc at Hasso Plattner Institute, Germany.

Abstract

State-of-the-art large language models (LLMs) offer impressive performance but come with high computational costs. This has created a market for LLMs-as-a-service, where users access models via APIs and pay a fixed price for each token in a model’s response. In this talk, I will first demonstrate that this pricing mechanism creates a financial incentive for providers to strategically misreport token counts, while users cannot prove, or even know, whether a provider is overcharging them. I will then introduce two potential solutions to foster trust in the ecosystem. First, I will show that this perverse incentive can be provably eliminated by transitioning to an alternative pricing mechanism: paying a fixed price per character rather than per token. Second, I will introduce a novel auditing framework that, under the current pricing mechanism, enables a third-party auditor to reliably detect unfaithful providers via sequential hypothesis testing. Throughout the talk, I will illustrate these concepts with experimental findings across several popular LLM families.

Speaker Bio
Stratis Tsirtsis is a postdoctoral researcher at the Hasso Plattner Institute in the Technology & Regulation group led by Prof. Sandra Wachter. His research focuses on algorithmic problems surrounding the social impact of artificial intelligence, bridging machine learning, game theory, causality, and optimization. Stratis completed his Ph.D. in Computer Science at the Max Planck Institute for Software Systems and has also conducted research at Stanford University and Meta (FAIR).

Logistics

• Date: January 14, 2026
• Time: 14:30 CET
• Zoom: https://cispa-de.zoom-x.de/j/61708401597
• Meeting ID: 617 0840 1597

We look forward to your participation. For more information about the seminar series, please visit https://ri-lab.org/riss/.

Best regards,
The Rational Intelligence Lab
CISPA Helmholtz Center for Information Security