AI Under Scrutiny: Safer Investor Journeys & Critical Perspectives on LLMs

## Details

🚀Join our free Budapest Data Science Meetup! In the first talk, Viktória Gerner and Alex Nasli will reveal how BrokerChooser measures and safeguards the AIO-driven investor journey, sharing accuracy insights from a 1,700-entity AI Safety Benchmark and showing how verified data and ScamShield MCP enable safer, more reliable AI results. Next, Salamon András from One Identity will cut through the hype to examine their limitations in reasoning, alignment, reliability, and environmental impact to see where LLMs truly stand.

📅 Date & venue:

• 26 November 2025, Wednesday, from 6 pm
• BrokerChooser Office, Central European University (CEU), Zrínyi u. 8-10. Meeting room N13 307/A ; 1051, Budapest. Entrance from Zrínyi utca

Please note that the event's official language is English!
Schedule

6:00-6:15 Warmup and chit-chat
6:15-7:30 Talks
7:30 - 9:00 – Networking

How much can you trust Google AI Overview?

AI is redefining how investors choose brokers - and Google AI Overview (AIO) plays a big role in this. At this meetup, we’ll share how we measure AIO pipeline at BrokerChooser and present our AI Safety Benchmark based on 1,700+ regulator-flagged scam entities. We’ll reveal key accuracy insights — and why even rare false-safe answers can be risky. Learn how grounding AI outputs in verified data and BrokerChooser’s ScamShield MCP can drive safer, more reliable results.

Alex Nasli is leading the Data Chapter at BrokerChooser as a seasoned LLM engineer. With a background in engineering and a passion for innovation, Alex specializes in crafting applications powered by large language models (LLMs) to solve real-world problems. From data extraction and translation to building tools that assist users, Alex is at the forefront of making AI both practical and engaging.
Beyond his professional expertise, he finds joy in quirky AI experiments, like creating videos from a single picture, proving that even serious tech can have a sense of humor.

Viktória Gerner is Head of Growth and Squad Lead at BrokerChooser. With a background in economics from Corvinus University and extensive experience at The Boston Consulting Group and BrokerChooser, Viki brings a wealth of expertise in leveraging LLMs for data extraction, translation, and user assistance.

A critical look at LLMs

Large Language Models (LLMs) have many truly impressive capabilities. They can write poems, imitate artistic styles, summarize articles, generate code, and generally produce seemingly coherent text in multiple languages. Media coverage often includes remarkable claims about the intelligence that LLMs supposedly possess now or will possess in the near future. For instance, a former Google executive has claimed that GPT-4 matches Einstein's IQ, and that we could be just a few months away from a machine with ten times Einstein's IQ.

However, there are serious shortcomings of LLMs that are not being addressed adequately. In this talk, I will discuss fundamental problems that have been present since the inception of LLMs, some of which are becoming more pronounced as the technology advances.

András Salamon graduated with a Master’s degree in Computer Science, specializing in Mathematics and Artificial Intelligence. He is currently a Senior Machine Learning Engineer at One Identity, where he focuses on integrating AI solutions into the company’s cybersecurity product lineup. Over the past five years, András has worked on training, and deploying models ranging from classical machine learning to LLM solutions.

Sponsors and partners

• BrokerChooser
• Budapest Data Science Meetup