qeternity

5 ms·

61.
by qeternity 8mo ago
It does not in PG17.
62.
by qeternity 8mo ago
Does not instill confidence when the queries they provide don't work. For anyone curious, the corrected query: SELECT sum(blks_hit)::numeric / nullif(sum(blks_hit + blks_read), 0) AS cache_hit_ratio FROM pg_stat_database;
63.
by qeternity 9mo ago
> In every state of the US (and most countries), people disobeying law enforcement will die. If you want to live, you comply, and you fight in court. This is one of the worst takes I have ever seen, to the point that you must just be tro
64.
by qeternity 9mo ago
Definitely. And some random guy on HN knows the value of Brex to Capital One better than Capital One does. Brex can be worth $5b today and also be worth less in the future. These two realities don’t conflict. Acquisitions can and do end poo
65.
by qeternity 9mo ago
I don’t think you understand how liquidation preferences work. They will get $300m back. Opportunity cost sure. But zero nominal loss.
66.
by qeternity 9mo ago
What is the point of this?
67.
by qeternity 9mo ago
Spoken like someone who has never started a business. Brex raised much less than $5b and Capital One apparently thinks it is worth more than that (otherwise they wouldn’t buy it). This is called value creation.
68.
by qeternity 10mo ago
Claude Code.
69.
by qeternity 10mo ago
> Electricity, water, roads, bridges are all public infrastructure. Why should payments be any different. I think you fail to understand why these are public infrastructure. Why should water be public infra but food is not?
70.
by qeternity 10mo ago
As someone with no experience in the domain, this sounds really interesting. But one of the companies mentioned has been attempting this for 9 years. Why hasn't this taken off already? It doesn't seem we're in need of any bre
71.
by qeternity 10mo ago
It has been a thing. In a single request, this same cache is reused for each forward pass. It took a while for companies to start metering it and charging accordingly. Also companies invested in hierarchical caches that allow longer term an
72.
by qeternity 10mo ago
In a chat setting you hit the cache every time you add a new prompt: all historical question/answer pairs are part of the context and don’t need to be prefilled again. On the API side imagine you are doing document processing and have
73.
by qeternity 10mo ago
> It’s much easier to tax the general population than businesses, as they don’t push back as much. Businesses don't pay taxes. People do. Every dime that a corporation pays is a reduction of capital returns to shareholders, or a red
74.
by qeternity 10mo ago
Quantization is not some magical dial you can just turn. In practice you basically have 3 choices: fp16, fp8 and fp4. Also thinking time means more tokens which costs more especially at the API level where you are paying per token and would
75.
by qeternity 10mo ago
This does not prove, at all, what you are claiming.
76.
by qeternity 10mo ago
Or perhaps more likely like a large corporation simply has economies of scale that smaller retailers cannot compete with.
77.
by qeternity 10mo ago
> Reminder that infrastructure is ~3% of the budget, the military is ~13%. Almost all of the rest are benefits, either health or money, for old people or various poverty reduction schemes. Or debt. The US Government is an enormous welfar
78.
by qeternity 10mo ago
How about you cite something.
79.
by qeternity 10mo ago
It would help people to consider your point if you made even a modest attempt to explain and justify what you mean.
80.
by qeternity 10mo ago
> life is genuinely worse today than it was 20 years ago, mostly because of technology Extraordinary claims require extraordinary evidence. Almost everything today in absolute terms is better than 20 years ago, even more so outside the d
81.
by qeternity 11mo ago
These are completely different. Agents (aside from the model inference) are not CPU bound. You gain much more by having a wider user base than whatever marginal CPU cycles you would gain in Rust/Go. Video games are of course a differen
82.
by qeternity 11mo ago
> but V3 (from February) has a 32B parameter model that runs on "16GB or more" of VRAM[1] No. They released a distilled version of R1 based on a Qwen 32b model. This is not V3, and it's not remotely close to R1 or V3.2.
83.
by qeternity 11mo ago
> DeepSeek and Qwen will function on cheap GPUs that other models will simply choke on. Uh, Deepseek will not (unless you are referring to one of their older R1 finetuned variants). But any flagship Deepseek model will require 16x A100&#
84.
by qeternity 11mo ago
Yes, absolutely in deep learning. Custom fused CUDA kernels everywhere.
85.
by qeternity 11mo ago
This is not the case for LLMs. FP16/BF16 training precision is standard, with FP8 inference very common. But labs are moving to FP8 training and even FP4.
86.
by qeternity 11mo ago
PyTorch is only part of it. There is still a huge amount of CUDA that isn’t just wrapped by PyTorch and isn’t easily portable.
87.
by qeternity 1y ago
> Also, all this vector stuff is going to fade away as context windows get larger (already started over the past 8 months or so). People who say this really have not thought this through, or simply don't understand what the usecases
88.
by qeternity 1y ago
People really just go on the internet and say stuff. Code is speech. Speech is protected (at least in the US).
89.
by qeternity 1y ago
If the client can generate a uuid4 they can also reuse a known uuid4
90.
by qeternity 1y ago
> Are they trying to be a full cloud platform like everyone else? Yes.

More ›