kevmo314

14 ms·

181.
by kevmo314 1y ago
Wholly agree. Too often we think about scale far too early. I've seen very simple services get bogged down in needing to be "scalable" so they're built so they can be spun up or torn down easily. Then a load balancer is
182.
by kevmo314 1y ago
I first noticed this with DeepSeek R1. For some really hard questions (some not even answerable), it would come up with a line of reasoning that convinced me that it had the right answer. If I read the answer without the reasoning, it was c
183.
by kevmo314 2y ago
Rarely are two models put head-to-head though. If Claude Sonnet 3.7 isn't able to generate a good PR review (for whatever reason), a 2% better review isn't all that strong of a value proposition.
184.
by kevmo314 2y ago
A great way to upsell 2% better! I should start doing that.
185.
by kevmo314 2y ago
Is there anything like this but for analyzing thousands of models? Most of this type of tool seem to work best for iterating and experimenting, but I'd actually like to use it for tracking and debugging something in production. We have
186.
by kevmo314 2y ago
At that rate you might as well force merge, no? That's equivalent to signing off and doesn't make the CI have a false positive signal when it hasn't actually run.
187.
by kevmo314 2y ago
I too work to get money or more likely build credibility.
188.
by kevmo314 2y ago
A strict C variant would indeed be quite nice. I've wanted to write CUDA kernels in Go apps before so the Go app can handle the concurrency on the CPU side. Right now, I have to write a C wrapper and more often than not, I end up writi
189.
by kevmo314 2y ago
Is CSS turing-complete? Could one make a JS to CSS compiler?
190.
by kevmo314 2y ago
Zero-copy is totally underrated. Like the site alludes to, Go's interfaces make it reasonably accessible to write zero-copy code but it still needs some careful crafting. The payoff is great though, I've often been surprised by ho
191.
by kevmo314 2y ago
The premise for vibe coding is that it's generating the entire app or site. If the app does what you want then it's meeting the bar.
192.
by kevmo314 2y ago
That's only true as long as you want to modify said code. If it meets your bar for reliability then you won't need to understand it, much like how we don't really need to read/understand compiled assembly code so we larg
193.
by kevmo314 2y ago
It crashed for me :( [failed to get system properties (java.lang.RuntimeException: No OpenGL context found in the current thread.)] Applets were a real vibe. The gradient-blue UI that came out of the box is true nostalgia.
194.
by kevmo314 2y ago
Infer what you're doing already based on your actions. To take a concrete example from the article, the Cursor screenshot includes a prompt "Please refactor these components ..." When was the last time I told an engineer on m
195.
by kevmo314 2y ago
The hate against chat interfaces is missing the real opportunity here. Consider the calendar interface the article describes: that we've managed to actually make it function within chat means we've roughly automated pretty much ev
196.
by kevmo314 2y ago
What do you mean? I just push the Record Separator key on my keyboard. /s in case :)
197.
Show HN: Vibe HTTP – AI generated HTTP responses(github.com)
9 points by kevmo314 2y ago|0 comments
198.
by kevmo314 2y ago
Maybe this is the dialup of the era.
199.
by kevmo314 2y ago
It’s a good question, admittedly I don’t know. I looked into it when Matt mentioned it but I’m not so familiar with what happens after ptx to say. If someone does know I’d love to learn why though.
200.
by kevmo314 2y ago
Yeah, it took me an hour-ish to get the performance and then an hour-ish to figure out how to actually render it correctly. I had never seen the P2/P5 PGM format before so I didn't realize P5 files were binary. :)
201.
by kevmo314 2y ago
Nice challenge, I got it down to 0.5ms/frame. https://github.com/kevmo314/prospero.vm
202.
by kevmo314 2y ago
Seems like Prisma + PostgREST but more tightly coupled. Neat idea, certainly seems useful for some small-scale projects.
203.
by kevmo314 2y ago
You might be surprised to learn that VS Code renders its viewport with HTML and it has no issues displaying this file.
204.
by kevmo314 2y ago
At that point in the article, I thought Natalie was some new LLM I hadn't heard about.
205.
by kevmo314 2y ago
Does it provide screenshots of the game state? I, too, would struggle to play the game pretty effectively if I could not visually see the game.
206.
Plant Machete(dwbowen.com)
1 points by kevmo314 2y ago|0 comments
207.
by kevmo314 2y ago
Neat idea. Regarding the performance, I think you could get a lot better performance by training a small classifier model, essentially an embedding model, and using the LLM as the distillation source. This would both be much smaller, addres
208.
by kevmo314 2y ago
It's somewhat ironic that the author advocates for keeping it simple and using pgvector but then buries a ton of complexity with an API server, auth server, Cloudflare workers, and durable objects. Especially given > Supabase easily
209.
by kevmo314 2y ago
> even for tasks that seem trivial to humans, reasoning LLMs will make a lot of mistakes before arriving at a plausible-sounding result. Like a lot of my coworkers analyzing a production bug? I would agree if the statement were that LLMs
210.
by kevmo314 2y ago
This paper seems like it misses the forest for the trees. The analysis is certainly interesting and the proposal sounds viable, sort of like a sliding window attention with a little more history. But if it is true that the separators contri

More ›