Jonas Andrulis

The shadows of reasoning – new design paradigms for a gen AI world

Large language models don't reason; they only see the shadows. Learn the new design paradigms required to build reliable AI systems you can actually trust.

The shadows of reasoning – new design paradigms for a gen AI world
#1about 3 minutes

From human-designed features to learned patterns in AI

The evolution of AI from manually crafted algorithms like HAR features to deep learning models that autonomously learn complex patterns like a 'chicken detector'.

#2about 5 minutes

Why AI fails to understand underlying physical rules

AI models trained on observable data learn the visual patterns of the world but fail to grasp the underlying physical rules, leading to illogical outputs.

#3about 3 minutes

Language models replicate patterns instead of reasoning

LLMs solve problems by matching text patterns rather than applying logical reasoning, as shown by their flawed solution to the classic 'wolf, goat, and cabbage' riddle.

#4about 6 minutes

Testing AI's reasoning with chess and board games

An experiment reveals that while an LLM can play chess by recognizing move patterns, it lacks a true understanding of the game's rules and can suggest impossible moves.

#5about 4 minutes

Visualizing AI patterns to make them accessible

A new approach involves building systems that can visualize and trace the complex patterns an AI uses, making its decision-making process more transparent.

#6about 3 minutes

Auditing AI outputs with pattern tracing

Using a 'Hobbit in the NBA' example, the system demonstrates how to trace which specific words in a prompt most influence an AI's answer, enabling auditing and fact-checking.

#7about 1 minute

Building human-in-the-loop systems with traceable AI

Traceable AI enables the creation of sophisticated workflows where human decisions can be integrated with auditable AI outputs for complex, high-stakes problems.

Related jobs
Jobs that call for the skills explored in this talk.

Featured Partners

Related Videos

AI: Superhero or Supervillain? How and Why with Scott Hanselman25:17

AI: Superhero or Supervillain? How and Why with Scott Hanselman

Scott Hanselman

Bringing the power of AI to your application.32:26

Bringing the power of AI to your application.

Krzysztof Cieślak

AI & Ethics28:27

AI & Ethics

PJ Hagerty

Should we build Generative AI into our existing software?24:51

Should we build Generative AI into our existing software?

Simon Müller

GenAI Security: Navigating the Unseen Iceberg24:33

GenAI Security: Navigating the Unseen Iceberg

Maish Saidel-Keesing

How AI Models Get Smarter35:16

How AI Models Get Smarter

Ankit Patel

Make it simple, using generative AI to accelerate learning26:54

Make it simple, using generative AI to accelerate learning

Duan Lightfoot

The shadows that follow the AI generative models28:30

The shadows that follow the AI generative models

Cheuk Ho

From learning to earning

Jobs that call for the skills explored in this talk.

AI Architect

Paradigma Digital
Municipality of Madrid, Spain

Azure
NumPy
Python
Pandas
PyTorch
+4