Question 1

What are the key takeaways from Sander Schulhoff 1.0's episode on Lenny's Podcast?

Accepted Answer

1. Prompt engineering is not dead and continues to be critical - studies show bad prompts can drop accuracy to 0% while good prompts boost it to 90%, and every new model release confirms prompting techniques still matter at scale. 2. The most impactful prompting techniques are few-shot prompting (giving examples), decomposition (breaking into subproblems), and self-criticism (having the LLM check its own work) - while role prompting no longer helps with accuracy-based tasks. 3. Product-focused prompt engineering is where techniques matter most - when millions of users interact with a single prompt and you can't watch every output, robustness and certainty become paramount. 4. Prompt injection is fundamentally unsolvable - prompt-based defenses and AI guardrails don't work, and the problem must be addressed at the AI research lab level through innovations in model architectures and safety-tuning. 5. Agentic AI security is the looming crisis - if chatbots can be tricked into generating harmful content, autonomous agents with real-world capabilities like managing finances or operating robots pose exponentially greater risks.

Question 2

What does Sander Schulhoff 1.0 say about prompt-engineering and AI-security?

Accepted Answer

"We have recognized the need for artificial social intelligence, communicating with AIs and understanding the best way to talk to them, understanding what their responses mean, and then how to adapt your next prompts to that response. Over and over again, we have seen prompt engineering continue to be very important." - Sander Schulhoff "I tried out all these different prompts and ways of showing the AI what it should be doing, but at the beginning of my process, I was getting little to no accuracy. What I ended up doing was taking a long list of documents that I went and coded myself, attached reasonings as to why each one was coded in the way it was, dropped it into my prompt, and that boosted the accuracy on that task up by 70%." - Sander Schulhoff "My best advice on how to improve your prompting skills is actually just trial and error. You will learn the most from just trying and interacting with chatbots, and talking to them, than anything else, including reading resources, taking courses, all of that. But if there were one technique that I could recommend people, it is few-shot prompting, which is just giving the AI examples of what you want it to do." - Sander Schulhoff

Question 3

What are Sander Schulhoff 1.0's views on prompt-engineering, AI-security, prompt-injection?

Accepted Answer

Sander Schulhoff: "We have recognized the need for artificial social intelligence, communicating with AIs and understanding the best way to talk to them, understanding what their responses mean, and then how to adapt your next prompts to that response. Over and over again, we have seen prompt engineering continue to be very important." Sander Schulhoff: "I tried out all these different prompts and ways of showing the AI what it should be doing, but at the beginning of my process, I was getting little to no accuracy. What I ended up doing was taking a long list of documents that I went and coded myself, attached reasonings as to why each one was coded in the way it was, dropped it into my prompt, and that boosted the accuracy on that task up by 70%."

Sander Schulhoff 1.0

AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff

TRANSCRIPT

Sander Schulhoff 1.0

AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff

Sander Schulhoff 1.0

AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff

TRANSCRIPT

Sander Schulhoff 1.0

AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff