kevmo314

14 ms·

241.
by kevmo314 2y ago
Such a model will always be less performant than one on tokens, as you're effectively switching to one byte per token. Solving this problem in code is much cheaper.
242.
Snake Diffusion(github.com)
2 points by kevmo314 2y ago|0 comments
243.
by kevmo314 2y ago
That seems like a pretty tough sell over bare playwright. Unless the UI is constantly changing, the cost of verifying tests are still successful seems like it would eclipse the cost of an engineer maintaining the test pretty quickly. Some m
244.
by kevmo314 2y ago
How do you avoid this becoming horrendously expensive per run? Are the results cached if the DOM doesn't change?
245.
by kevmo314 2y ago
Yup
246.
by kevmo314 2y ago
Can't say I feel like Kubernetes has solved more of my problems than it created.
247.
by kevmo314 2y ago
> Understanding the nuances and distinctions between these key concepts becomes crucial as you navigate the complex — yet rewarding — waters of container orchestration. Can't say I've ever felt rewarded for using Kubernetes, li
248.
by kevmo314 2y ago
The multi-step agent API they've produced feels so much like reinforcement learning. Neat! memory = [user_defined_task] while llm_should_continue(memory): # this loop is the multi-step part action = llm_get_next_actio
249.
by kevmo314 2y ago
That seems like a fun project to replicate independently. You didn't want to rebuild it?
250.
by kevmo314 2y ago
Curious because I am not familiar: are PRs required or are PR reviews required?
251.
by kevmo314 2y ago
Wow someone who finally has this same unpopular opinion as I do. I'm a huge fan of review-optional PRs. Let it be up to the author to make that call and if it were really important to enforce it would be more foolproof to do so with au
252.
by kevmo314 2y ago
Why? If we define zero as 0xFFFF it will saturate by rolling over the first bit, which increment will not change?
253.
by kevmo314 2y ago
Instead of dedicating two bits to flags, could the specific task in the video be accomplished by leveraging the sign bit? Define 0xFFFF to be zero, 0x0000 to be one, 0x0001 to be two, etc. Then for decrement, do an atomic `fetch_uint_sub(1)
254.
by kevmo314 2y ago
Code generation leaves a layer of abstraction between the API and the actual implementation which works great if that code generation is bug-free but if it's not, you're like... totally fucked. Most commonly people say you can rea
255.
by kevmo314 2y ago
How did you manufacture your integrated circuits? I am very curious to learn about that process since it seems like it must be quite custom, or is there a neat way to reuse existing circuits?
256.
by kevmo314 2y ago
At some point the label of amateur starts costing more than the value the achievements bring.
257.
Llama Inference in 150 Lines(gist.github.com)
6 points by kevmo314 2y ago|0 comments
258.
by kevmo314 2y ago
That tool itself might be worth publishing. I've noticed some cofounders (Supabase's kiwicopple comes to mind) that are very on top of HN comments and I am not sure if they're literally reading all the HN comments or doing th
259.
by kevmo314 2y ago
I used to have a pretty successful browser extension that I shut down because the site it worked on shut down. I grew it by adding features that people would ask for on the site's forums, for example a user would say "hey can we h
260.
by kevmo314 2y ago
Creating a faster python implementation can definitely be done. OpenCV is a thin wrapper over the C++ API so it's not due to some intrinsic python slowness. It is not easy to resolve though and I suspect the way python code is typicall
261.
by kevmo314 2y ago
For Go specifically, there are some libraries like Gorgonia ( https://github.com/gorgonia/gorgonia ) that can do inference. Practically speaking though, the rate at which models change is so fast that if you opt to go th
262.
by kevmo314 2y ago
I'm curious how the case/when version performs. Unlike the author, I don't think that is any smellier than the list/map solution they've come up with.
263.
by kevmo314 2y ago
This research exists. This study [1] suggests a 2.7 year increase in life expectancy. Doing some math, 5 minutes per day for 80 years is around 0.2 years. So sure I might have gone fishing a bit to respond to a random HN comment but the dat
264.
by kevmo314 2y ago
> Durant, who made his money off the boom, said he put his time and money into the project to show people that they aren’t helpless to technology giants and that we can figure out massive problems if we work together. I can appreciate th
265.
by kevmo314 2y ago
Why waste time say lot word when few word do trick indeed. I mean if we're really debating question length then omitting the just also works... "Add a DNS record?" is sufficient.
266.
by kevmo314 2y ago
Saying something like "would it be easier to add a DNS record" seems effective too without the baggage?
267.
Beating every possible game of Pokemon Platinum at the same time [video](youtube.com)
6 points by kevmo314 2y ago|0 comments
268.
by kevmo314 2y ago
Next up: a kindle with physical pages, each page is a screen!
269.
Rory Sutherland – Are We Now Too Impatient to Be Intelligent? [video](youtube.com)
2 points by kevmo314 2y ago|0 comments
270.
by kevmo314 2y ago
I've updated the README! rCUDA is indeed inspiration, in fact it inspired scuda's name too :)

More ›