Aurornis

26 ms·

691.
by Aurornis 3mo ago
> > K3 pushes the boundary of end-to-end knowledge work. On the GDPval-AA v2 leaderboard, Kimi K3 scores 1687. The benchmark evaluates AI models on real-world tasks across 44 occupations and 9 major industries; Kimi K3 ranks behind on
692.
by Aurornis 3mo ago
> But if I pay more, who's to say I'll get a better product? Price isn’t the only axis by which products are evaluated. Aliexpress eBay and even Amazon are full of listings that have high prices, just waiting for someone naive
693.
by Aurornis 3mo ago
The article is reporting on what was discussed in court: Autism, suicidal tendencies, living with grandparents. These were all probably brought up as elements of the story meant to influence the verdict. Take it up with lawyers.
694.
by Aurornis 3mo ago
From the arrival: > Jubair has 22 previous convictions related to hacking, fraud and harassment. There’s more to what was going on here and none of us is really qualified to diagnose the psychology behind it from the details. I hope they
695.
by Aurornis 3mo ago
You are not in a position to hire them, let alone at a big interesting company. Forgive my bluntness, but your opinion doesn’t matter to their situation. When I say their resumes are gold-plated I mean at the companies they want to work at.
696.
by Aurornis 3mo ago
> With so much human capital and resources dedicated to developing and running LLMs all the other business, research opportunities aren’t being explored and invested in It probably feels this way if you only see headlines like this, but
697.
by Aurornis 3mo ago
Reading this is weird because those network effects and friend group sharing experiences are exactly how my music lover friends use Spotify . They share playlists instead of the files. They make friends of friends through playlist recommen
698.
by Aurornis 3mo ago
I would do that in a heartbeat if I was in that situation. Having a major leadership role at a big company is kind of a slog. Getting a chance to go back to a fast growing startup as an IC in a way that will only benefit their career (ex-An
699.
by Aurornis 3mo ago
> I'm not sure as others why others feel this is a major change. Because the phones where available in US and Europe and now they won’t be? That’s a major change. You can say the company was changing over time, but a move like this
700.
by Aurornis 3mo ago
> Wait, you think not giving additional aid = responsibility for whatever happens in the developing country? This, and your $500 cancer donation, is an absurdist reduction of the problem. The USAID contributions weren’t anything like you
701.
by Aurornis 3mo ago
> yes. https://www.alphaxiv.org/abs/2607.bonsai-27b table 14 shows that bonsai retains roughly 95% of the fp16 27b model's average performance I know that's what the paper says the benchmarks say, but thes
702.
by Aurornis 3mo ago
I assume most people don’t fall into this trap where they try to get the world to adapt to their diagnosis. I think most people realize it’s something they need to work on. I get the impression that it’s largely the people who arrive at the
703.
by Aurornis 3mo ago
I have a free perplexity account from some promotion. Not sure what comparison you’re trying to make because Perplexity’s whole thing is that it’s really fast. It launches the search with parallel agents and then even seems to render some o
704.
by Aurornis 3mo ago
Drug holidays can make a small impact, but tolerance is going to build to many of the effects no matter what you do. The other problem with drug holidays is that some people start getting into cycles where they don’t take the drug and have
705.
by Aurornis 3mo ago
> another task a lunch break whatever and it will be done when you're back At 5 tokens per second and unknown prompt processing speed, you may need a very extra long lunch break depending on your codebase.
706.
by Aurornis 3mo ago
> What does it make clear? That it’s pulling a lot of watts. Good for you if it’s replacing electric space heaters.
707.
by Aurornis 3mo ago
There are some traps along this path, too. A number of younger people I've worked with (in the double digits now) have gotten ADHD and/or ASD diagnoses and then become overjoyed that "everything makes sense now". But the
708.
by Aurornis 3mo ago
> (untreated ADD causing repeated failures, repeated failures causing depression) Really important to understand that depression can also manifest as poor concentration. There's a huge problem right now with people getting locked in
709.
by Aurornis 3mo ago
Intuniv gets a bad reputation because the patient has to be titrated up to the final dose and it takes a long time to see the full effects. It's the opposite pattern of stimulants where the patient feels great (unnaturally so, due to s
710.
by Aurornis 3mo ago
Tramadol is a pain medication, not a sleep medication. It has opioid effects and can cause drowsiness but would never be prescribed for sleep. It’s also typically a controlled substance. I don’t know where you live but I’d be surprised if a
711.
by Aurornis 3mo ago
It’s not a cost savings relative to paying API prices even if you’re suspending it. This is an option if you must run local inference, you’re not sensitive to speed, and the budget is low. It’s not going to be cheaper than paying API prices
712.
by Aurornis 3mo ago
Anyone running LLMs at home will come to that realization quickly, if they’re looking at their power bills. Even feeling the heat output of a computer running at 100% in your office makes it clear. I was responding to a lot of the comments
713.
by Aurornis 3mo ago
A dual Xeon of this era is probably pulling 300W or more when loaded. At national average electricity prices, that’s $1.35 per day. More during the summer if you have to cool the space. If you run it 24/7 and ignore prompt processing t
714.
by Aurornis 3mo ago
If you’re interested in these projects you should check out the project this was based on: https://github.com/JustVugg/colibri It says so right in the readme. They’re not hiding anything.
715.
by Aurornis 3mo ago
> Once you've finished reading the next article on HN about a 5 tps Xeon, your task will be complete. If I spend 10 minutes reading an article, that would only generate 3000 tokens. That’s not counting the prompt processing time. We
716.
by Aurornis 3mo ago
> I have a prediction. By the mid of 2027, we will have >200B MoE models running on basic consumer hardware. This prediction alone isn’t useful at all without a bound on speed and maybe quantization. You can already run >200B MoE m
717.
by Aurornis 3mo ago
> If it's ~30% bigger and not as good as GLM 5.2, why would I tinker with this model? The benchmarks never tell the full story. Some of the open weights models have been benchmaxxed for a while. Their utility on real work can be dif
718.
by Aurornis 3mo ago
Occasional Benadryl is fine. The dementia risk is from correlations found in other medicines that share some of the same properties, when taken for many years. It’s not a good idea to take Benadryl for many years, but occasional use or even
719.
by Aurornis 3mo ago
The parent post was talking about doctor prescribed medications. The reason diphenhydramine Is associated with a slight increase in dementia risk is the anticholinergic properties. This risk increase is from correlational studies on other
720.
by Aurornis 3mo ago
> I don't think that's a safe assumption. You could construct an article pretty quickly if you had a topic, a few points to hit, a conclusion and a short list of links, then fed that to the machine. All the LLM would be doing i

More ›