k__

5 ms·

61.
by k__ 3mo ago
Effect's use of generators is also nice. While looking a bit off, it allows writing imperative business logic.
62.
by k__ 3mo ago
Are they still using eink?
63.
by k__ 3mo ago
"Fully open source model" with "Long-term, license-free availability for industry" Nice trick to be not comparable with most other LLMs on the market, open weight or proprietary. But I think, that's the right way.
64.
by k__ 3mo ago
I tried Hy3 today and liked it. It's a (small) step up from DSV4P. Something on that level but multi-modal would be quite nice!
65.
by k__ 3mo ago
While it's true that unsupervised AIs build a heap of trash, I have to admit, it doesn't look much worse than some of the code I saw powering products that sold for millions. The only thing that changed is the scale.
66.
by k__ 3mo ago
Strange, there aren't even 20 at a time above Germany. How many uplinks can one satellite handle?
67.
by k__ 3mo ago
Pi
68.
by k__ 3mo ago
Yeah, DeepSeek V4 (Flash and Pro) is below $0.004 for 1M cache hits. Even with usage based billing I'm below $1 writing code all day.
69.
by k__ 3mo ago
"Europe's rapid warming is partly the result of [...] a drop in the number of tiny polluting particles in the air. This means that less of the Sun's energy is reflected back into space, leaving more energy to heat the Earth&
70.
by k__ 3mo ago
I tried DS4 Pro yesterday for one particular task that Flash struggled with. That one Pro task cost $1.50 and the whole day of using Flash cost me $5.
71.
by k__ 3mo ago
Thanks! I'll try that.
72.
by k__ 3mo ago
Yeah, v4 flash is dirt cheap, but it's running in circles quite often. Might very well be that a better model is cheaper if it gets things right the first try. Maybe I should route to a better model when v4flash hasn't solved afte
73.
by k__ 3mo ago
Nice. I paid $6 yesterday for DeepSeek V4 Flash on OpenRouter. That's like $120 dollar for a month, and it's not even a good model.
74.
by k__ 3mo ago
Llama3.1 instruct seems to be doing okay on that page, mostly because it's dirt cheap.
75.
by k__ 3mo ago
So, their incentive is to promote AI music, since they don't have to pay royalties for them.
76.
by k__ 4mo ago
Did Nim 3 go all-in on compile time memory management?
77.
by k__ 4mo ago
I'm curious about Denuvo's opinion on that.
78.
by k__ 4mo ago
Maybe, you could pipe it through T5 or something.
79.
by k__ 4mo ago
Training DeepSeek was magnitudes cheaper than training the SOTA models it relied upon. In theory, other countries should be able to replicate that effort and improve it.
80.
by k__ 4mo ago
As I understood it, they were only a US thing anyway. Because of healthcare issues etc.
81.
by k__ 4mo ago
I tried some smaller Gemma4 and Qwen3.6 quants on my MBA with M5/16GB and had like 20-60 tokens per second. At 60 it felt pretty okay and that hardware is on the lower end. I'd assume a Mac with 32-64GB memory would get some reaso
82.
by k__ 4mo ago
I thought the joke was that people aren't paying enough money.
83.
by k__ 4mo ago
100K seems quite much. I had the impression, models would get inconsistent after just 3000 words.
84.
by k__ 4mo ago
Why doesn't Mistral distill?
85.
by k__ 4mo ago
I tried it with embedded programming, and failed miserably.
86.
by k__ 5mo ago
200ml every waking hour? Seems excessive.
87.
by k__ 5mo ago
Ah, so it a "smart" retry mechanism?
88.
by k__ 5mo ago
So, this basically ensures that models call the right tools with the correct format?
89.
by k__ 5mo ago
This. Directors of small companies are the same, they're just not wealthy enough that they could do any harm.
90.
by k__ 5mo ago
Anti-intellectualism is at it again, hu?

More ›