HN user

Mr-Frog

172 karma
Posts1
Comments41
View on HN

older vision LLMs chopped up images into patches which were projected into the same embedding token space as words. Newer ones use an encoder to more efficiently project an image into token space. Then it runs through the same attention layers as the text component.

Just use a microcontroller. And for the vast majority of hobby projects I suspect that is good advice.

I recently bought a hobby FPGA kit because I think the Von Neumann model is a beautiful innovation and I want to learn more about doing useful computation tasks from the lowest level logical components. I've always been interested in computers, but when I was in grade school the only useful computers I could afford were discarded PCs that I brought back to life with Linux distros. I now have a fulfilling career as a direct result of access to cheap hardware and open-source software when I had much more time and much less money than I do now. Decisions like AMD's are a long-term negative for the industry.

It's kinda awesome that after decades of software and hardware advancements to prevent computers from arbitrarily executing data as instructions, we've decided to let agents arbitrarily execute data as instructions.

Interestingly, the three careers you listed are protected by strict professional credentialing systems that do not exists for programmers, and professionals in law and medicine enjoy a social prestige that is certainly attractive to a group of people who might not innately enjoy the work itself.

I unironically agree with this. 100 years ago, Skid Row and Bunker Hill in Los Angeles were full of SROs, boarding houses and long-term hotels. The people who lived there didn't disappear, they're just all sleeping in the street now.

I was working with network drivers and the core network stack. I've heard bad things about questionable device driver implementations, I could only imagine what you saw.

I did an internship where I had to write fake network devices to test some server platform firmware configuration. I spent a bunch of time digging around the core PCIe code and Linux network stack. The first few weeks were a bit overwhelming but by the end of the internship I was really impressed with how clean and relatively understandable the interactions between userland, kernel space, and hardware operations were. It really made me aware about the importance of deliberately designed API surfaces that are critical to millions of users.

I also enjoy C programming, and it was interesting to see how the developers built out such a complex system without the more friendly programming language structures that I was first exposed to in school (mostly C++ and Java).

I've always enjoyed the practical research and projects that happen at the CalPoly schools, my friends who attended always seemed to be doing some very neat hands-on work.

Anecdotally, my occasional interactions with Linux kernel source code have always impressed me with how the project manage to effectively structure such a large C codebase. Until I look at FreeBSD code, which is even more impressively organized. Both of these projects have really helped me think about how to best organize huge complicated systems.

I'm pretty sure the majority of Linux contributions are made by corporations these days, which almost certainly have internal management structures.

This is the absolute top of the top, domestic tech in Mexico pays pretty poorly. Google (as of last year) also let people transfer from high COL to lower COL, reduce their base, but keep the original RSU contract, which will probably skew reported compensation.

Entry-level total comp at Google Mexico is about 80k USD, which is comparable to 180k in local purchasing power.

[Edit] on second glance, my original datapoint for Google was questionable, but it looks like Lyft/Amazon definitely gives offers in the 50k-70k USD range.

I really enjoyed the concepts in "Artificial Intelligence, a Modern Approach" which really grounds a first-principles foundation of automated reasoning. Warning: The first 80% of the book doesn't have any sexy new deep learning approaches, but I still think it is very valuable to see the history.

With lots of talk lately about California's river water management, I thought it would be neat to share these photos of the Colorado River Delta, where 100% of the water is diverted to human use and none of it to the sea.

We have very nearby and valuable evidence to demonstrate the impact of short-sighted freshwater release policies.

LA and California as a whole have some of the highest taxes in the nation

The City of LA has a lower per capita tax revenue than most large Texan and southern cities, largely due to property tax caps.

anti-tech superiority complex which is counterproductive

I didn't have a phone until significantly after my peers, so I used our family computer, Instagram's undocumented API, and a variety of SMS forwarding solutions to keep in touch with my friends, which I think definitely sparked my interest in hacking and a career in software.

I developed a superiority complex, but it was more anti-conformity and pro-hacking than anything.