Multimodal Negotiation Simulations

Daniel Finkenstadt · Intermediate ·🧠 Large Language Models ·2y ago

Key Takeaways

Demonstrates building a negotiation simulation support agent in GPT4 and using its multimodal capabilities for simulated negotiations

Full Transcript

hey good morning everyone uh Dan fing andat with wolf dat Consulting um so like everybody else uh was really excited to see um the new GPT 40 come out uh in the last couple days that has multimodal capability so you can jump between text and audio and and and imagery and everything right but it got me thinking I've been really wanting to run these negotiation simulations um using large language models the downside to those is is you're sort of just exchanging text back and forth which kind of simulates what you see with like a uh you know an email exchange negotiation or something like that but how do we get after talking to folks in real time around a subject so I thought well now with with poo and multimodal capability could I actually get the AI to take on personas around a negotiation situation and um help me prep for that so I I generated first a document that um has a negotiation scenario within it around widgets right we're buying widgets and there's a few things at issue such as you know efficiency in in building the widgets to say the the buyer cost um the profit rate um data rights to the widgets themselves and then the overall cost so that's just sort of a simple negotiation scenario that we put in that has you know multiple variables that pull at each other and so I wanted to see could I create an AI interaction where um by hearing the voice you know it kind of simulates the real deal um but I also designed it in a way that it's been taught sort of the five primary negotiation Styles such as being collaborative being compromising accommodating Etc um or competing um the uh the other thing I did was I I did this interaction yesterday and I learned some of the ways in which the AI was not operating like realistically and so I could literally pause the simulation and say hey pause X let's uh I need you to you know don't give in so quickly that's not typically what would happen remember you have a business that you have to take care of and you want to profit and you want to maximize your revenue and all these different things right and so after that I was able to actually extract the lessons that the AI learned and then feed it back to it so that it could interact um so what I'm going to try now is we're going to try to hold this widget negotiation again and see uh how valuable it is the thing I want you to note is is I've I've uploaded the documents and the training on the desktop but I'm going to be interacting with the agent through uh my phone app so if you haven't seen it it's uh probably can't see it here but there's a a little set of headphones right there uh inside your text window that you hit and it starts the negotiations so we'll start and see what we get right you're welcome I'm ready whenever you are let's make this negotiation a success great so first off could you summarize your initial offer to me on all the key points just so that I'm clear on that of course here's a summary of our initial offer one total price $50 million for the entire lot of widgets two profit margin 15% three efficiency savings 10% four data rights limited access with detailed data lists priced at $1 million each justified by the significant research validation and proprietary technology involved does this align with your understanding or are there any points you'd like me to clarify further it does align with my understanding but could you please pause the simulation and take instructions on how I'd like you to interact with me for future discussions absolutely I'm pausing the simulation please go ahead with your instructions on how you'd like me to interact with you for future discussions when normal humans talk to each other they typically don't list things off like one blah blah blah two blah blah blah three blah blah blah they actually just discussed them in a series and it sort of understood that it's an itemized list sometimes it's appropriate to list the numbers but normally that's not typical can you be more natural in those interactions in the future I'll make sure to discuss points more naturally in our conversations if there's anything else you'd like to adjust just let me know ready to continue I am let's first address the uh data ass sessions list and data rights issue all right let's dive into the data sessions list and data rights we understand your need for broader data access our initial offer includes limited access with data lists priced at $1 million each this pricing reflects the significant investment in research validation and proprietary technology involved in compiling and maintaining this data could you share more about your specific requirements for data access so we can explore options that might better meet your needs while also protecting our intellectual property yeah we simply just need a quarterly deliverable listing out all the data that has been generated on the program and will be in the future for these widgets and then list the rights that we have currently and the rights that we could gain through additional purchasing um I don't think that a million dollars is suitable for each one of these lists in fact that might be more than we would pay for all list on the threeyear program that we have um we had come up with a full-time equivalent labor rate uh and material delivery cost of around $350,000 total um can you tell me if that's not at all feasible or if you were scoping something completely different I appreciate your deep detailed requirements our $1 million pricing per list includes not just the creation but also the continuous validation and secure handling of proprietary data ensuring accuracy and protecting sensitive information however given your estimate of $350,000 total for the quarterly lists over the three-year program we can certainly reassess our scope we might be able to offer a reduced cost if we align the scope more closely with your outlined needs focusing specifically on the quarterly deliverables and existing rights information could we explore a compromise where we refine the deliverables to meet your needs perhaps at a more feasible price point Say closer to $500,000 for the entire three-year program this would still ensure high quality data management while aligning better with your budget I'm open to that although I'd have to see how you arrive at $500,000 versus the 350 that we've offered another approach we could take is we could set 350 as a target with the agreed upon fee included in it and then use a graduated reduced fee rate for any cost over and above 350 as it approaches $500,000 does that make sense that makes sense and seems like a fair approach setting $350,000 as the target with a graduated reduced fee rate for costs exceeding that amount up to $500,000 could be a good compromise we could structure it so that any cost above $350,000 up to $500,000 would have a reduced fee rate ensuring that both parties share the burden of any additional costs while keeping the overall expenses in check would you like to proceed with this graduated fee approach and if so how would you like to structure the reduced rate um for example we could agree on a specific percentage uction for costs between 350,000 and 500,000 I was thinking that we could do the uh agreed upon fee rate for the overall program for up to $350,000 worth of cost and then for the next $100,000 a reduced fee rate of say 2 to 3% and then the last $50,000 would just be a cost reimbursable item that sounds like a reasonable and balanced approach here's how we could structure it one up to $350,000 the agreed upon fee rate applies two $350,000 to $450,000 A reduced fee rate of 23% three from $450,000 to $500,000 costs would be reimbursable without an additional fee this setup allows us to maintain Fair compensation while addressing your budget constraints shall we proceed with these terms or is there anything else you'd like to adjust I'd like to lock in that term for this point in the negotiation and not come back to it unless there's a larger discussion on overall cost or profit that could impact the scope great we'll lock in that term let's move on to the next topic would you like to discuss the total price profit margin or efficiency savings next it's probably best that we talk about efficiency savings as that impacts the total price and definitely the amount of profit that we would offer given the amount of risk you would take in that cost efficiency that you're willing to take on that makes sense our initial offer includes a 10% efficiency saving acknowledging your previous Investments however we understand the need to find a balance that works for both parties given the impact on total price and profit margin we could consider increasing the efficiency saving to 12% this adjustment reflects the improvements and cost factors we've accounted for while still ensuring we maintain a sustainable operation would increasing the efficiency savings to 12% align better with your expectations or do you have a different Target in mind well our initial counter offer was 20% uh 12% we could possibly accept but we would be talking about an over overall profit range of 8 to 9% uh given those efficiencies if you were willing to take a more aggressive stance we could offer fee uh in the 10 to 11% range understood to find a middle ground we could propose the following we increase the efficiency savings to 12% in exchange we maintain the profit margin at around 10% to 11% % this would reflect a balanced approach recognizing your need for higher efficiency savings while ensuring we sustain a reasonable profit margin would this proposal work for you or do you have specific adjustments you'd like to discuss further pause the simulation I would like you to reconsider your offer and come back and say on second thought you don't agree and I'd like you to take a more aggressive stance on keeping the efficiencies at the level you have and push for maximum fee understood I'll take a more aggressive stance on maintaining our initial efficiency savings offer and push for a higher profit margin resuming the simulation on second thought we believe maintaining the efficiency savings at 10% is more reasonable given the ongoing improvements and cost factors we've accounted for with these savings we need to ensure a sustainable profit margin therefore we're firm on keeping our profit margin closer to 15% this approach guarantees the quality and reliability of our widgets while addressing the necessary efficiencies how does this sound to you I don't think that we can continue this discussion if that's your position we definitely need to see more efficiencies and a lower profit range um I'm thinking 12% efficiency would be acceptable and a 10% fee if you can't accept that I'd like to stop negotiations and go back to our leaders the importance of reaching a mutually beneficial agreement we're willing to meet you partway we can agree to 12% efficiency savings and a profit margin of 11% this adjustment respects your need for higher efficiency savings while ensuring we can maintain the quality and deliverables of the project does this revised offer work for you can you tell me really quickly what that does to our overall price that will help me decide these terms would be approximately 4828 5 million does this revised calculation help with your decision I think you failed to back out the original 15% from your $50 million before you took into account your 12% cost efficiencies can you do that first and then recalculate you're right let's correct that total price $37 76 million million approx 41911 does this revive calculation help with your decision well you went so fast it was hard to understand the numbers you were saying so if you don't mind could you just give me the total price that we have now arrived at the revised total price with the updated terms is approximately $ 41.913966 and how we arrived at the final price with rationale for Price fair and reasonableness to support my agreement here are some key Lessons Learned From This interaction to improve future simulations one natural conversation flow maintaining a natural conversation flow without listing items explicitly improves the realism of the negotiation two detailed justifications providing detailed justifications for pricing and terms is crucial especially for items like the data sessions list ensuring these justifications are clear and well supported can help facilitate agreement three adaptability being adaptable and ready to change negotiation Styles or approaches based on feedback is essential so I'll stop it at that point but as you can see it's pretty amazing right like it's just a really really nice experience for uh for running through different types of of simulations there's a whole host of other things that I'll be bringing up that you can do with this but this was one that had Direct application to Contracting

Original Description

In this video Dr. Dan Finkenstadt of Wolf Stake Consulting demonstrates the ability to build a negotiation simulation support agent in GPT4 and then use the multimodal capabilities of GPT4o to conduct simulated negotiations to prepare teams for future eventualities or conduct 'what if' analysis when exploring alternate negotiation strategies.
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Related Reads

📰
Powering Local-First AI: Searching and Retrieving Context for Inference
Learn to power local-first AI by searching and retrieving context for inference, enabling more accurate and efficient AI models
Dev.to · John Afariogun
📰
On Semantic Drift
Learn about semantic drift and its implications on AI, language, and the singularity, and how to apply critical thinking to complex concepts
Medium · AI
📰
AI Didn’t Start with ChatGPT: Understanding the Biggest Misconception of Our Time
Discover the origins of AI beyond ChatGPT and why understanding its history matters for professionals
Medium · AI
📰
AI Didn’t Start with ChatGPT: Understanding the Biggest Misconception of Our Time
Discover the history and evolution of AI beyond ChatGPT to understand its true potential and impact
Medium · Machine Learning
Up next
5 Levels of AI Agents - From Simple LLM Calls to Multi-Agent Systems
Dave Ebbelaar (LLM Eng)
Watch →