High Signal Podcasts Evidence ledger
Method
Browse
← Back to evidence

Evidence receipt / recommendation

Published · transcript-backed

Mentions personal use of Claude.

19 Jun 2025 Lenny's Podcast AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff (Learn Prompting, HackAPrompt)

“Yeah. For me when I use Claude for coming up with interview questions and just suggestions of... It's actually really good.”

— Lenny Rachitsky

Source trail

Everything needed to verify it.

Speaker
Lenny Rachitsky
Attribution
Verified speaker
Claim type
recommendation
Recorded
19 Jun 2025
Publisher
Lenny's Podcast

Transcript context

…It depends. And this also can get into, are you going to few-shot prompt with different pieces of additional information? I usually don't. No need to use the XML brackets. If you feel more comfortable with that, if that's the way you're structuring your prompt anyways, do it. Why not? But I almost never include any structured formatting with the additional information. I just toss it in. Awesome. Okay. So we've talked through four, let's say, basic techniques. And it's a spectrum I imagine, to more advanced techniques so we could start moving in that direction. But let me summarize what we've talked about so far. So these are just things you could start doing to get better results either out of your just conversations with Claude or ChatGPT or any other LM [inaudible 00:36:34], but also in products that you're building on top of these LMs. So technique one is few-shot prompting, which is you give it examples. Here's my question, here's examples of what success looks like or here's examples of questions and answers. Two is you call decomposition where you ask it, what are some sub problems that you need to solve? What are some sub-problems that you'd solve first? And then you tell it, "Go solve these problems." Three is self-criticism where you ask it, can you go back and check your response, reflect back on your answer. And it gives you some suggestions and you're like, "Great job. Okay, go implement these suggestions." And then this last advice, you called it additional information, which a lot of people call context, which is just what other additional information can you give it that might tell it more. Might help it understand this problem more and give it context, essentially. Yeah. For me when I use Claude for coming up with interview questions and just suggestions of... It's actually really good. I know they're just like, "Oh, they're all going to be so terrible." They're getting really interesting, the questions that Claude suggests for me. I actually had Mike Krieger on the podcast and I asked Claude, what should I ask your maker? And it had some really good questions. And so, what I do there is I give context on, here's who this guest is and here's things I want to talk about. Ends up being really helpful. Yeah, that's awesome.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.

Named in this claim

Books, apps, tools, and people.

Search evidence