k__

37 ms·

511.
by k__ 2y ago
R1 or V3?
512.
by k__ 2y ago
I had the impression the implementation happens first, then the ratification later when things worked out well.
513.
by k__ 2y ago
Alibaba probably doesn't want DeepSeek to get all the fame.
514.
by k__ 2y ago
Would DeepSeek-V3 be cheaper?
515.
by k__ 2y ago
I read somewhere, if you fast, the body thinks you can't find food, so it gives you more dopamine to motivate you to find more.
516.
Promising results from DeepSeek R1 for code(simonwillison.net)
979 points by k__ 2y ago|746 comments
517.
by k__ 2y ago
DeepSeek R1 1.5B is pretty nice with ollama. It gives reasonably good answers and streams a bit faster than I read.
518.
by k__ 2y ago
I did lose 10kg in 2 months on low carb and felt much clearer after a week of doing it. But I couldn't keep it up for more than 3 months.
519.
by k__ 2y ago
I only know Vue as "like Svelte, but without compiler goodies". So to me a "Vue but built 100% on web standards" at least has a unique selling point.
520.
by k__ 2y ago
I'm a bit sad. They could at least make the controls a bit funny.
521.
by k__ 2y ago
Don't they have TEE?
522.
by k__ 2y ago
Profit/person is the only metric that should grow.
523.
by k__ 2y ago
Talking about not wasting money and then planning a building that has more than 6 floors is quite cheeky.
524.
by k__ 2y ago
It's not just HN. It's the whole industry. I'm a freelancer and all of my clients talk is AI. I think, it's cool tech, but also quite overhyped. But I get it, AI has become the magic box that the masses of "idea guy
525.
by k__ 2y ago
Just because some laws define fair use in some kind of way, it doesn't mean potential customers see it that way.
526.
by k__ 2y ago
Honestly, I tried it, but I didn't understand the game.
527.
by k__ 2y ago
"built upon a blend of synthetic datasets, data from filtered public domain websites, and acquired academic books and Q&A datasets" Does this mean the model was trained without copyright infringements?
528.
by k__ 2y ago
I'm not an AI founder, but I'd say there is still some value to be added by better architectures. These AI wrappers run ship engines on a kayak. For example, most coding companion's ignore my code base the moment I close othe
529.
by k__ 2y ago
Seems like only working on open source code has its benefits.
530.
by k__ 2y ago
Okay, thanks. What's the least complex to implement then?
531.
by k__ 2y ago
Half-OT: What's the most resource efficient way to push data to clients over HTTP? I can send data to a server via HTTP request, I just need a way to notify a client about a change and would like to avoid polling for it. I heard talk a
532.
by k__ 2y ago
I once joined a company where a Dev was in the process of rewriting a big codebase. I didn't think much of it and run Gource on it for fun. Turns out, he did countless rewrites over the year. Interesting to watch and explained why the
533.
by k__ 2y ago
I'm glad that so many open source and even "small" models like Gemma are better than gpt4.
534.
by k__ 2y ago
Most of these things work pretty well. It's just that they don't "just" work. Just because AWS or Google can pull it off doesn't mean it's something anyone can do.
535.
by k__ 2y ago
Brave Search might be a good alternative. It's index is created by people surfing with the Brave browser, so only websites used by real people are included.
536.
by k__ 2y ago
And over HTTP/2 and 3 they are efficient?
537.
by k__ 2y ago
"Just entirely without cocoa beans, but with oats and/ or sunflower seeds" https://choviva.com/details
538.
by k__ 2y ago
I've already seen cookies in the store that are praised as containing an "oat based chocolate alternative", as if replacing cocoa was something anyone asked for. That led me down a rabbit hole of food substitutes in the Germa
539.
by k__ 2y ago
Ah, thanks. I just read the parent post, lol.
540.
by k__ 2y ago
What's the extension name?

More ›