ykhli's comments

On: Show HN: New.email – Building Emails with LLMs

[ykhli]

Super interesting - I'd actually love to write a MCP server based on this as I find myself generate react email from cursor all the time.

Also is there a way to add styles / instruct on colors?

[zenorocha → ykhli]

Having an MCP server for this would be super cool.

You can definitely add styles and instruct on colors.

On: Show HN: MCP server that lets Cursor agent send Morse code through your light

[ykhli]

This MCP server allows Cursor or Claude Desktop to control Philips Hue lights and send messages through them using Morse code. Have fun!

On: A Deep Dive into MCP and the Future of AI Tooling

[ykhli]

How do MCPs actually work, where are the real use cases, and the challenges today.

[mindcrime → ykhli]

I don't think too many people are going to want to write the essay that would be needed to truly answer that. I would just point you to:

https://modelcontextprotocol.io/introduction

and/or any of the dozens of Youtube video introductions to MPC, to get a good feel for how it works, and how people are using it.

On: OpenAI operator and Anthropic performed below average on chimp memory test

[ykhli]

I benchmarked OpenAI operator and Anthropic's computer use on the human benchmark chimp test (https://humanbenchmark.com/tests/chimp).

Before I began the test, I thought the agents would be much better at this task than most humans -- after all they should have better, more stateful memory than us. The results are intriguing.

Here are the scores from 10 attempts: OpenAI operator: 5, 5, 6, 5, 5, 4, 6, 5, 5, 5 Anthropic computer use agent: 7, 9, 6 (rate limited), 12, 9, 7, 9, 11, 12, 6 (rate limited)

On: Show HN: Inngest 1.0 – Open-source durable workflows on every platform

[ykhli]

Congrats! I used Inngest when I wrote a video processing pipeline here https://github.com/tigrisdata-community/multi-modal-starter-...

Amazing devEx. Thanks so much for all the work and enabling a local mode too

On: Open source Wikipedia search through vector db

[ykhli]

Oh wow this is such an interesting case. I wonder if it's because of the embedding model used here.

If I search for Leo Tolstoy it works well, but "Amor towles" doesn't lead to the right page.

On: Show HN: A cartoon intro to how the attention mechanism works

[ykhli]

THANK YOU!!! this is amazing. Will update the cartoons over next week.

Also thanks so much for taking the time to write these down!! Can't appreciate it enough

[refulgentis → ykhli]

No problem at all, all the best, thank you for your hard work :) This isn't easy and it's extremely valuable

On: Show HN: A cartoon intro to how the attention mechanism works

[ykhli]

hey! thanks so much for the feedback. I'd actually love to keep updating / iterating these cartoons so they are more approachable. If you have time, I'd love to hear more on which pages are confusing & how I could have explained it better!

I _tried_ to give a definition to embeddings on page 11, but maybe that's not the most intuitive? Lmk! feel free to DM

[itronitron → ykhli]

The use of the word "relevant" such as in the phrase "are the most relevant" is problematic.

Relevancy is dependent on and relative to a specific task or interest. "Related to" or "associated" might be a better choice, since parts of a text can be statistically associated with each other.

While I feel that you are accurately conveying the terminology in the field I personally feel that terminology overstates things. For example, the notion that words closer in meaning are generally closer in latent space could be more accurately stated as "words closer in latent space are often used in similar ways."

I think this comes down to meaning being largely determined by a person's individual interpretation at a given point in time, similar to the arbitrariness of relevancy.

On: Show HN: Multi Modal Starter kit - roast a movie with AI

[ykhli]

Hi HN! Recently we made a starter kit for better learnings on how to get started building AI apps with multi modal models. It's been fun to build, but we also discovered there are many things we needed to take care of: caching, long video processing pipelines, model evaluation, etc

I wrote more about the technical details here. Feel free to try it out and open PRs! https://twitter.com/stuffyokodraws/status/177558959544044376...

On: Building a fair multi-tenant queuing system

[ykhli]

Great job abstracting away so much complexity!

> each step is a code-level transaction backed by its own job in the queue. If the step fails, it retries automatically. Any data returned from the step is automatically captured into the function run's state and injected on each step call.

This is one thing I've seen so many companies spending tons of time implementing themselves, and happens _everywhere_ -- no code apps, finance software, hospitals, anything that deals with ordering system...the list goes on.

Glad I no longer need to write this from scratch!