fooker

10 ms·

211.
by fooker 4mo ago
When it was profitable to mine crypto with GPUs people used to sell these miner GPUs on the used market after about two years. These were about half of the cost of an used GPU just used for gaming. By that pricr, I'd say a GPU kept bus
212.
by fooker 4mo ago
Right, most professional mathematicians know almost nothing about their neighboring branches of mathematics. An algebraic geometry researcher would be hard pressed to understand a new result from category theory or even something closer lik
213.
by fooker 4mo ago
Well, we had tried to ban exporting cryptography once. It didn't go well.
214.
by fooker 5mo ago
I'm curious about whether we will start discovering new maths in the next few years that provide insight into unsolved CS or Physics problems!
215.
by fooker 5mo ago
print(ExpectedOutput)
216.
by fooker 5mo ago
They say hill climbing https://microsoft.ai/news/building-a-hillclimbing-machine-la... Unless they specifically clarify that the testing and training benchmarks are completely separate, we have to assume they test on t
217.
by fooker 5mo ago
No, you're missing the point. The idea is not that you parallelize simple tasks. With a thousand agents, eventually, once we figure out how to orchestrate agents for real, you can tackle significantly more complex projects. Here's
218.
by fooker 5mo ago
Embedding ads in LLM responses is something researchers are having a lot of trouble figuring out right now. I have seen the results of some early attempts. It fails in such hilarious ways that all these companies are scared of productizing
219.
by fooker 5mo ago
If you are willing to spend about 2000 on GPUs, we are almost there. In my opinion, the bottleneck is the package management layer and not the model capabilities and performance. I have been an avid Linux user for decades, and if I find it
220.
by fooker 5mo ago
Running a local llm at 10 t/s overnight to extract data from a few PDFs will burn more in electricity than paying cents for the hosted kimi models. You can (sometimes) break even if you have a workstation GPU.
221.
by fooker 5mo ago
What are the tasks that do well with 8-10 t/s ?
222.
by fooker 5mo ago
What you can run locally in consumer hardware is progressing pretty well. If you get a not-quite-the-best gaming GPU like a 5080, you can run local models that are better than the state of the art from early 2025. Depending on what you want
223.
by fooker 5mo ago
As you can see from people blaming Codex instead of docker here, shooting of the messenger is very much happening.
224.
by fooker 5mo ago
That works great if it's one agent, absolutely doesn't if you want to tackle something complex that warrants using ..say.. ten agents. I can imagine a future where this technology empowers you to do things with a thousand agents.
225.
by fooker 5mo ago
You'd do it once and figure out that none of your friends have regular status updates any more.
226.
by fooker 5mo ago
This is almost exactly how we have limited GPUs that do double precision floating point with decent performance from being exported.
227.
by fooker 5mo ago
On the other hand, this sends an excellent message about unlocked doors :)
228.
by fooker 5mo ago
> the presence of a security hole should not be seen as permission to exploit Why not? I want the agents on my side to exploit whatever they can to help me. The ones on the other side certainly won't be artificially nerfed.
229.
by fooker 5mo ago
Wow there is really is a relevant xkcd for everything!
230.
by fooker 5mo ago
Figuring out which commit broke what functionality is not something you can expect users to do.
231.
by fooker 5mo ago
Yeah this is how software development works now, no matter how much anyone wants to disagree with it. The technology is here, you can't put it back in the box. If your tool has AI agents trying to find exploits 24/7, you'll n
232.
Rsync 3.4.3 has hundreds of Claude commits(mastodon.gamedev.place)
123 points by fooker 5mo ago|81 comments
233.
by fooker 5mo ago
> might even be cheaper to do this Oh yeah, very often. Especially if it's a loop and resolves the same way very often. Modern CPUs just blaze through code like this, after three decades optimizing for object oriented and dynamic la
234.
by fooker 5mo ago
You can just generate the 'vtable' as code :) switch (animal.type) case Cat: return cat_speak() case Dog: return dog_speak() The generated code has the functions resolved in compile time, there's no function p
235.
by fooker 5mo ago
I think the primary value of a project like this is the demonstration that this is possible and a proof that it does not incur some unknown tradeoff you'll discover after spending resources doing it. IMO the maintenance story is more o
236.
by fooker 5mo ago
This is a very popular opinion that is sort of obsolete now in my opinion. It was a valid concern last year. We have seen tremendous progress on this in the last 4-6 months. Even if your initial prototypes are unmaintainable slop, the state
237.
by fooker 5mo ago
Why go through LLVM at all? There are a bunch of tools that JIT somewhat optimized assembly.
238.
by fooker 5mo ago
vibeware :)
239.
by fooker 5mo ago
This is how software development works now. We have to live with it. The models are good enough that this works. You can keep disagreeing for a while, but know that almost all the code in the industry is written like this now.
240.
by fooker 5mo ago
Here you go - https://x.com/elonmusk/status/2030159267689632121 https://x.com/elonmusk/status/1948572708369039542

More ›