In recent months, companies across industries have embraced artificial intelligence (AI), but now they face a crucial challenge: correcting so-called hallucinations. Now, a startup that a Brazilian helped create believes it has the solution.
“During the development of an AI system, it works perfectly in a test environment. But, when it goes into production, new scenarios emerge, and the AI may not be prepared for this and begins to lose performance”, explains Brazilian André Quintanilha, former executive at Accenture and co-founder and Chief Growth Officer of the startup. Palqeein an interview with InfoMoney. “If there is no monitoring solution, it may be too late to react.”
Hallucinations occur when AI systems generate incorrect or unrealistic information, which can lead to poor decisions and significant operational risks. Keeping an eye on this problem, Palqee created a solution that promises to identify episodes and mitigate the problems caused by them.
Continues after advertising
Called Prisma, it evaluates factors such as input tone, clarity and changes in the environment that affect results, anticipating possible risks and helping to increase the confidence of AIs that have dominated the corporate environment.
The product was born in a context in which companies like Uber and Amazon face lawsuits because their AI did not work as expected, leading companies to seek to identify deviations before they escalate.
One of them was KPMG. The company developed an internal AI that answers questions about attached documents. However, answers to questions like “What is the value of this invoice?” and “How much is this invoice?” generated discrepant answers, even though they were technically correct.
Continues after advertising
“We discovered that the system had a hallucination rate of 14%, and that the clarity of the questions influences the chances of hallucination by 6%,” says Sabrina Palme, CEO and co-founder of Palqee, to InfoMoney. “In legal matters, for example, this influence rose to 40%.”
KPMG used this information to tune the system, reduce the error rate, and guide its employees on how to interact with the tool more efficiently.
Also read:
Continues after advertising
The tool is designed to be used in any industry or AI model, from traditional machine learning-based systems to the most advanced generative models. According to the startup, this flexibility is essential for companies operating in dynamic and regulated markets, such as finance, healthcare and aviation.
“How can we ensure that these technologies really work as expected?” asks Sabrina. “We seem to have lost control a bit after the implementation of AI in the market, so risk management has become a key concern, especially in regulated markets, where it is crucial to ensure that AI systems are accurate, relevant, non-biased and free of errors.”
Palqee raised pre-seed a few years ago and says he has already reached the breakeven before launching the new solution. The startup plans a new round of financing in early 2025.
High Performance Entrepreneur
ONLINE COURSE
Rian Tavares reveals the secret to business management based on the method of high-performance athletes.
