When tokenizing terabytes of text for your training corpus, the speedup here is probably doing real work in saving you time (and money?). You get a faster iteration cycle when figuring out and adjusting your datasets.

I started working on Alki because wrangling and deploying AI models on edge devices felt clunky. Going from a Hugging Face model to something production-ready required too many manual steps. Alki aims to simplify that by offering more conversion options, packaging models into reproducible bundles, and making them easy to publish or run on edge hardware.

Llama.cpp was a great starting point, but it led me down the rabbit hole of building a toolchain that could handle different quantization methods, export formats, and deployment bundles for orchestration.

It’s still early and I’d love feedback from infra engineers and startups running LLMs outside the cloud, especially around packaging formats, performance, and what would make the biggest quality-of-life improvements.

Comments URL: https://news.ycombinator.com/item?id=45386792

Points: 1

# Comments: 1

I have no real background in mathematics outside of undergraduate calculus, so was wondering if the more math-curious people here could point out some rule/theory this may be related to, or find some illogical gap in my daydream :) So here goes...

If you pick out two random integers from an infinite set and add them, there would be a 66.66% chance of that sum turning out to an even number. This is based off two conclusions: 1. There is 50/50 chance of picking out an even or odd integer from infinity 2. 2 of the 3 possible sums would lead to an even sum (odd+odd=even, even+even=even, odd+even=odd)

So if you were to keep taking two random integers from this infinite set (which would be for... infinity), you'd come up with an infinite set of integers which now 66.66% are made up of EVEN numbers.

If you were to keep repeating this process of pulling two numbers from your freshly created infinite set of summed integers, each new iteration of that set would be an infinity made up of more and more even numbers over time. So each set would get closer to a 100% collection of even numbers, but never quite getting there as it approaches infinity (the percentile of even numbers in your sets over time could be plotted out as an asymptote). You end up with an infinity of sets, all approaching a complete infinity of even numbers.

...

Very, VERY, useless(?) and random daydream, but it got me curious to see if one of my basic assumptions about working with infinity is wrong, or find others who have done similar though experiments (I'm confident I'm not the only one).

Thanks in advance to anyone for their input.

Sincerely, A Math Noob

Comments URL: https://news.ycombinator.com/item?id=27323447

Points: 2

# Comments: 5

It was an incredibly novel feeling of seeing the room I was in get mapped out, furniture and all, to see interfaces and 3d models rendered in a way that felt interactable in real space. Moving renders of objects around to see how things could be - well, it was just a lot of fun; it felt like playing with a new video game console as a child.

Sure, the overall experience was clunky, with the motion tracking and sub-par lenses constantly reminding you that it was only a prototype, but that first interaction excites you about all the things that could be possible if it actually is refined into a final product. Perhaps this was the same experience that led many of the early investors to put so much faith in Magic Leap while there was still a long road ahead of them.

While I do believe our understanding of how the brain functions still has a long way to go, which includes treatments for mental illness, there is something to be said about acknowledging more the external factors of our modern day lives that fill us with so many problems.

Personally, realizing where I have "disconnected" from people, values, and truly human experiences has been a large step forward in improving my mental health. I hope it's something that others who feel so many emotional challenges can try as well to see if they can make progress with their own mental health.

I've been working professionally as a software developer for 3-4 years, and am currently in a large engineering organization (not a FAANG but a financial firm). While the day-to-day work I do is "coding", little of it has been on projects where I feel I've come out more knowledgeable about building good software.

My current struggle is having recognized the __large__ gap between writing "code that works" and writing programs that are stable, maintainable, and extensible. As much as I've been aware of it, I still find myself writing more of the former than the latter. Work projects feel limiting on how much I can explore new ideas / re-think architecture and design philosophies, while personal projects have felt mediocre and any attempt at writing good code has come off as "premature optimization".

I love programming, but this skill/understanding plateau has left me a bit burned out recently.

Have any of you ever experienced something similar after a few years in the industry or as professionals? How did you deal with it, and what were some actionable steps that helped you overcome it?

I know some of you on HN have been programming for decades, so I'm curious to hear what were the moments you were able to take and become better engineers.

Thank you for any insights you may have!

Comments URL: https://news.ycombinator.com/item?id=23092288

Points: 2

# Comments: 1

I'm currently living in Houston, working as a SWE, and can attest to the pros of living here. Regarding your concerns, the city is quite demographically diverse from people of all walks of life. Summers do get hot and humid, but nothing unbearable. The winters are more than tolerable (its currently 68 degrees outside) as well.

Market data services such as Reuters/Morning Star provide incredibly clean data at great convenience for anything you can imagine, even including astronomical data.

Its kind of like the Black-Scholes model. If it really was that simple and straightforward, everybody would be using it to make a ton of money.

Founded by Jim Simons, the "Father" of quantitative research. The firm is famous for their significant year-over-year returns and notorious for only hiring PhDs from mathematics/physics/computer science.

There is an interesting book on how Jim Simons created the company and built his team of academics from the ground up, "The Man Who Solved the Market"

Comments URL: https://news.ycombinator.com/item?id=19400868

Points: 9

# Comments: 2