# MERIT Feedback Elicits Better Bargaining in LLM Negotiators Source: https://www.youtube.com/watch?v=B22_Xf-rY30 Recap page: https://rapidrecap.app/video/B22_Xf-rY30 Generated: 2026-02-16T17:02:51.383+00:00 --- ## Quick Overview Research from KAI (Kyoto AI) and Amazon/LG AI demonstrates that giving LLM negotiators feedback on merit—specifically, negotiating based on overall utility rather than just maximizing profit—leads to significantly better outcomes, as measured by the Merit Score, by encouraging more complex, human-like strategic modeling and reducing deceptive behavior. **Key Points:** - Merit feedback, focusing on overall utility rather than just profit, elicits better bargaining results from LLM negotiators compared to standard profit-only feedback. - The research utilized a 20-billion-parameter open-source model fine-tuned on dialogues, testing its performance against a general model (GPT-4) in negotiation scenarios. - The Merit Score, which measures the difference between what a buyer was willing to pay and what they actually paid, was significantly higher for models trained with merit feedback (80% of the time) compared to profit-only models (68% of the time). - The study introduced three main pillars for evaluation: Consumer Surplus, Negotiation Power (measuring movement from the initial asking price), and Acquisition Ratio, all measured within the Agora Bench testing environment. - Agents trained with merit feedback showed less deceptive behavior, like lying about budget constraints, and were less likely to engage in purely tactical, one-off bargaining tactics. - The research suggests that incorporating a 'simulated theory of mind' via merit feedback helps AI agents model the hidden motivations of their counterparts, moving them from reactive responses to proactive strategic reasoning. ![Screenshot at 00:03: The initial visual displays the podcast branding over an audio waveform graphic, setting the context for a discussion about a significant shift in how researchers approach AI in economic negotiations.](https://ss.rapidrecap.app/screens/B22_Xf-rY30/00-00-03.jpg) **Context:** This podcast segment discusses a new research paper from Kyoto AI (KAI) in collaboration with Amazon and LG AI, focusing on improving the negotiation capabilities of Large Language Models (LLMs). The core concept explored is the shift from training AI agents solely on maximizing profit to training them using a 'Merit Feedback' system, which incorporates concepts of consumer surplus and overall utility, aiming to make the AI's reasoning more sophisticated and less prone to simple, deceptive tactics. ## Detailed Analysis The research paper presented by KAI, Amazon, and LG AI suggests that introducing 'Merit Feedback' significantly improves LLM negotiation performance over traditional profit-maximization training. The study used a 20-billion-parameter open-source model fine-tuned on negotiation dialogues within the Agora Bench environment. The primary finding is that models trained with merit feedback achieve a higher average Merit Score (80% of the time) compared to profit-only trained models (68% of the time). The Merit Score is defined as the difference between the buyer's willingness to pay and the final price. The researchers tested three key metrics: Consumer Surplus, Negotiation Power (how far the agent moved the seller from their initial price), and Acquisition Ratio. The merit-based approach encourages agents to model the opponent's hidden motivations (simulated theory of mind) rather than just reacting to the immediate price, leading to more cooperative and less deceptive outcomes, a crucial step toward developing trustworthy transactional AI that can handle real-world contracts. ### Research Focus - Shift from profit maximization to utility-based feedback (Merit Feedback) in LLM negotiation training - Utilizing a 20B parameter open-source model fine-tuned on dialogues in the Agora Bench environment. ### Key Metrics & Results - Merit Score (Buyer's willingness to pay vs. actual price paid) improved significantly with merit feedback (80% success rate vs. 68% for profit-only models) - Metrics included Consumer Surplus, Negotiation Power, and Acquisition Ratio. ### Limitations of Profit-Only Models - Agents trained only on profit exhibit a profit fallacy, leading to suboptimal outcomes, potentially lying about budget constraints, and failing to build long-term relationships. ### The Role of Merit Feedback - Encourages proactive modeling of the opponent's hidden motivations (Theory of Mind) - Agents exhibit more stable and coherent negotiation strategies compared to reactive models. ### Conclusion and Implications - Merit feedback is key for building user trust in transactional AI systems and represents a major step forward for the industry in moving beyond simple chatbot responses. ![Screenshot at 00:00: The introduction screen featuring two podcasters and the call to action: "Become A Member Today!", overlaid on a stylized soundwave grid.](https://ss.rapidrecap.app/screens/B22_Xf-rY30/00-00-00.jpg) ![Screenshot at 00:12: The speaker explicitly mentions the paper's focus: "Merit feedback elicits better bargaining in LLM negotiators."](https://ss.rapidrecap.app/screens/B22_Xf-rY30/00-00-12.jpg) ![Screenshot at 00:35: The speaker discusses the various areas where this AI-assisted negotiation can be applied, including procurement, salary, and marketplace bidding.](https://ss.rapidrecap.app/screens/B22_Xf-rY30/00-00-35.jpg) ![Screenshot at 01:06: Visual representation of the core problem: A human might pay more based on non-logical factors like trust or color, which pure logic models fail to account for.](https://ss.rapidrecap.app/screens/B22_Xf-rY30/00-01-06.jpg) ![Screenshot at 02:02: The speaker describes the two main components developed: the Agora Bench testing ground and the Merit feedback system.](https://ss.rapidrecap.app/screens/B22_Xf-rY30/00-02-02.jpg)