Hacker Newsnew | past | comments | ask | show | jobs | submit | srbhr's commentslogin

> Someone with cheap electricity can serve inference profitably at prices that would bankrupt the current small set of centralized providers.

Strong claim, There's a similar article about China is eating the world on HN. Maybe then, they're the ones who can take over the inference cloud?


That's exactly I was thinking.


I've added a product (SWIRL) it shows 0% off. :) How can I change that?


have you enabled the discount checkbox?

Anyway let me know the coupon code and percentage off, i can add it manually ;)


same here, didn't specify a discount either


Not just me, but our team is building: https://github.com/swirlai/swirl-search

We are trying to build an AI metasearch engine that does RAG and Chat based on your office documents. Without any ETL or a vector database.


Can you please give insights into "Studying topics from first principles instead of corporate documentation"? Studying topics from first principles?


An example of this could be Postman or your favorite Git GUI.

You can also learn HTTP networking and curl, as they are the underlying principles of Postman, and learn Git graphs instead of using a Git GUI tool.

Next time you're confronted with a problem, instead of starting with "let's solve this with library X or framework Y," begin by asking "what are we trying to solve?" and then explore multiple options to solve that problem.

A technique I often employ is to first ask what the complexity of the problem is, then evaluate the complexity of each potential solution. For instance, the goal might be implementing automatic deployments, but the solution became way more complex than the original stated goal.

The technique often stil lead you to popular solutions, but it could also point you toward simpler, less conventional alternatives. Sometimes even eliminating problems by side stepping them completely.


I still don't get it. I don't care about reducing complexity, I care about reducing complexity that will be me of my teams problem, and even then, it's not my top priority, getting rid of potential manual error is more important.

A pencil will always be simpler than all of Keep's infrastructure, but it requires vastly more skill and attention than a cloud notes app, to the point where it's primary benefit seems to be as a form of mental exercise.


Thank you : )


Sure, I've been doing a lot of full-stack web development lately, and I found that reading chapters from the textbook "Computer Networks" by Andrew Tenenbaum as a primary source to be far more enlightening about how networks operate, than reading any of the hottest frameworks documentation on how they work under-the-hood. Corporate documentation tends to be more about giving off the vibe of competence, as opposed to being actually informative. They want to make it look like their tool/framework is easy to learn by offering false senses of understanding, instead of the deep understanding provided by textbooks. There are exceptions, but they seem to be growing more and more rare. Compare man pages for old CLI tools versus documentation for frameworks like Next.JS, which tends to gloss over important implementation details that might make you a better overall developer (instead of a framework wrangler).

On the side, I'm trying to enhance my knowledge on embedded systems as well, and picked up the book "Making Embedded System" by Elecia White. I have learned so much in just the first few chapters, even though I worked on an embedded software project for a full year recently. This has served me well, because previously my understanding of embedded systems was more of at the level of an Arduino, rather than an embedded developer. Arduino is fantastic for getting your feet wet, but it does hide away a lot of slightly complicated details (that are not actually that hard to learn with a little effort).

I would basically say, always prefer lengthy and somewhat dry textbooks over superficial and meme-heavy blog posts (or worse, YouTube videos). Learn the subject as if you only have one shot in your career to learn it, and have to learn a permanent representation of the subject that can last decades. Entertaining corporate docs, blog posts and videos are like a sugar-heavy diet for the mind, you'll feel energized at first, but eventually become sick from lack of deep understanding of anything

In a twist of irony, here is a blog post that explains this idea reasonably well - https://archive.ph/XSPRr

Here's a canonical blog post (again, the irony) on why understanding subjects deeply is important - https://www.joelonsoftware.com/2002/11/11/the-law-of-leaky-a...


The better "Framework wranglers" usually eventually learn the "leaks" as part of the abstraction's API, and learn to be comfortable not trusting anything.

TCP can fail, but so can my own totally deterministic code I only think I understand. Or more likely, I trip and fall and break the Ethernet cable and the whole thing fails.

My job is to make the final result reliable anyway, even knowing the thoughts in my own head are not trustworthy. Even if I knew anything about proof theory I could still make a mistake with a mathematical proof, and simplicity can't stop it. Left and right are fairly simple concepts and I've mixed them up so many times.

A lot of devs think in terms of trust and quality, and are so quick to assume entire subsystems are perfectly fine, whether it's because it's a high quality thing from a trusted vendor, or because it's simple and they did it themselves and feel they understand it.

It's just like in real life, how people seem totally sure they won't trip and fall or drop what they're holding. I'm used to understanding my own hands as unpredictable, so I guess that's why I like tech.

It's less about avoiding mistakes and more about predicting and planning for mistakes.


> It's less about avoiding mistakes and more about predicting and planning for mistakes.

Totally agreed


Thank you, good sir : ) for the explanation.


You can try: https://github.com/swirlai/swirl-search

- doesn't indexes data - has connectors to various apps - metasearch under the hood - re-ranking of search results before RAG


This is awesome. However, I'm waiting for a 3.5 Opus. Wonder what happened to that one.


You could try SWIRL Search. I have found API integration from different apps without ETL pretty good. https://github.com/swirlai/swirl-search


It sounds more like a chore. Read, write, note things down. And then re-view your cards/notes over and over. There has to be a short term goal associated like an upcoming exam, interview, etc. for spaced repetition. Although I'm lazy. :)


Talking about primes,(2^136279841)-1 is the New Largest Known Prime Number. Discovered 21st Oct (Today).

Ref. https://www.mersenne.org/


I wonder what Turing would've made of that.


The world's greatest PRNG.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: