Points: 2

# Comments: 2

Every time I tweaked a prompt, I had to rerun a bunch of test cases manually and eyeball the results.

It felt like writing code without unit tests.

Existing tools I found were either:

- Full frameworks where you write evaluators in Python.

- Big platforms expanding into monitoring/security.

I wanted something simpler: a fast CLI that just tests prompts.

So I built prompttest, a pytest-like workflow for LLMs:

- You define a prompt in a .txt file with `{variables}`.

- You write test cases in .yml, with plain-English criteria.

- You run prompttest to see clear pass/fail results in your terminal.

The core idea is that your "assertion" is just English.

Example:

> The response must be polite and address the user by name.

Then a model grades the output for you.

This gives a safety net — you can refactor prompts and instantly see regressions.

The project is still early.

It runs on OpenRouter, so you can test against many models (including free ones) with one API key.

Would love feedback, ideas, or use-cases you’d want supported.

GitHub: https://github.com/decodingchris/prompttest

Comments URL: https://news.ycombinator.com/item?id=45015426

Points: 2

# Comments: 0

Is it possible to represent the latent space as an image, and then reconstruct the original image from that representation?

Well, here's the answer.

Comments URL: https://news.ycombinator.com/item?id=42281884

Points: 4

# Comments: 2

Comments URL: https://news.ycombinator.com/item?id=38377593

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=38377580

Points: 2

# Comments: 1

I've just launched a new Chrome extension named Blink, which is inspired by a comment made by TheRealHB on HN. It suggests using a blinking LED to combat ADHD and maintain focus:

https://news.ycombinator.com/item?id=38276107

Blink replicates this technique by introducing a subtle, pulsating rhythm to your Chrome toolbar. The pulse begins at a quicker pace, similar to a rapid heartbeat, and then gradually slows to a calm, steady tempo. This visual cue is designed to be noticed in your peripheral vision, gently encouraging your brain to sync with its pace, potentially aiding in concentration.

Check it out here: https://chromewebstore.google.com/detail/lhmfkleedadlmbjnhjp...

Comments URL: https://news.ycombinator.com/item?id=38319539

Points: 1

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=36369489

Points: 2

# Comments: 0

This a video demo of an idea I'm working on. Basically, I'm using a few different AI models to create a conversation simulator.

The user gets to simulate any scenario with the ai. The goal is to help people practice their communication skills or important conversations in a "safe" environment.

Think flight simulators for pilots, but instead of flights, it's conversations.

Comments URL: https://news.ycombinator.com/item?id=36342983

Points: 2

# Comments: 1

And yeah, repetition is an issue, there’s an instruction to not repeat story element in the initial prompt but it’s not working well. Any suggestions?

Comments URL: https://news.ycombinator.com/item?id=36183551

Points: 30

# Comments: 11

Comments URL: https://news.ycombinator.com/item?id=36181100

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=36175912

Points: 2

# Comments: 0

Beautiful is better than ugly.

Explicit is better than implicit.

Simple is better than complex.

Complex is better than complicated.

Flat is better than nested.

Sparse is better than dense.

Readability counts.

Special cases aren't special enough to break the rules.

Although practicality beats purity.

Errors should never pass silently.

Unless explicitly silenced.

In the face of ambiguity, refuse the temptation to guess.

There should be one-- and preferably only one --obvious way to do it.

Although that way may not be obvious at first unless you're Dutch.

Now is better than never.

Although never is often better than right now.

If the implementation is hard to explain, it's a bad idea.

If the implementation is easy to explain, it may be a good idea.

Namespaces are one honking great idea -- let's do more of those!

Comments URL: https://news.ycombinator.com/item?id=36104943

Points: 2

# Comments: 1