xiphias2

15 ms·

301.
by xiphias2 2y ago
The article doesn't get into the interesting part: Salvadorian government continued accumulating BTC. So far IMF didn't say anything specific, so the June review will be interesting to see if IMF continues with the loan or cancels
302.
by xiphias2 2y ago
I believe the torture word shouldn't be used here because that's something much much worse that I wouldn't want anybody to experience. But it's clear that he isn't punished enough and got away with an easy fake pris
303.
by xiphias2 2y ago
By stealing billions of dollars he destroyed more people than the worst repeated violent offenders. Also he's clearly paying politicians off. In my opinion he should be categorized in the worst category even just looking at the lost th
304.
by xiphias2 2y ago
It's cool proof that it's worth to spend the effort for Postgres authors to add vectorization to the default columnstore.
305.
by xiphias2 2y ago
It's making more money selling ads than the AI companies selling AI, so it's classic innovators dilemma: they can't really launch an AI product that gives better quality results then Google search ads, which are really bad at
306.
by xiphias2 2y ago
It's strange to have so much anti-sentiment here about the internal data that Google has. Of course Sergey doesn't have to spend 60 hours as he retired. I have never been asked to work more than 40 hours when I was working at Goog
307.
by xiphias2 2y ago
SpaceX is clearly better technology, but I believe what matters is whether Verizon was fairly awarded the contract. If the contract was made fairly and in good faith, cancelling it would be bad for contract law precedence in the new adminis
308.
by xiphias2 2y ago
The more I tried to implement Clean Code, the more it helped me appreciate Worse is Better approach, which comes from an observation, not a dogma: while all programmers strive for simplicity both in implementation and interface, when they c
309.
by xiphias2 2y ago
Have you tried Grok 3 thinking? I haven’t made up my mind if O1 pro or Grok 3 thinking is the best model
310.
by xiphias2 2y ago
Reasoning LLMs getting better at ARC-AGI prove that they are able to solve tasks that are symbolic without putting in specific search on the CPU (which is the brute force method). It's never ,,pure scaling'' (just running the
311.
by xiphias2 2y ago
I'm not sure it's so easy, I travel a lot, and western cities are the hardest to get any medication. In Hawaii we were robbed with my girlfriend and went to a pharmacy and to doctor with no papers and we weren't able to refil
312.
by xiphias2 2y ago
Drastic cost cuts are usually not simply performance issues and usually the mistake of top management (usually overhiring), but of course it hurts most the employees.
313.
by xiphias2 2y ago
There was data sharing between Twitter, Google and Facebook as well. USA has to make a decision between safety and national security.
314.
by xiphias2 2y ago
I don't see the Think button, and for me the answer is much below deepseek-r1 even thought I have Premium+ subscription. I'm just getting instant stupid answer instead of thinking.
315.
by xiphias2 2y ago
DEI got an extreme level of support from NGOs even while Trump was the president that they won't have at the next election. I hope they can renew themselves. The good news on the left side though that I don't think the next candi
316.
by xiphias2 2y ago
o3 mini is still behind o1 pro, it didn't impress me. I think the people who think anybody is close to OpenAI don't have pro subscription
317.
by xiphias2 2y ago
Probably both. The amount of corruption in the name of DEI / Ukraine war / COVID got staggering and government has been overspending money to unsustainable levels (and it looks like half of it was plain stealing). The sad part of
318.
by xiphias2 2y ago
While this is true, (and I don't want to try to be politically neutral here), if we look at Elon's enormous increase in power and also significant increase in wealth, it would definately be positive ROI for him. He went through a
319.
by xiphias2 2y ago
One great feature of the SQL pipe syntax is that it works much better together with programming languages (generating, or modifying existing SQL queries). It would be great to have it standardized fast and implemented everywhere.
320.
by xiphias2 2y ago
Probably not, as polls show that it was his second worst decision in his voter group. He is polling quite well with his voters, this was more just a power play.
321.
by xiphias2 2y ago
Is ,,Oracle not controlling JavaScript'' really a good argument? Who wants Oracle to start controlling it? We're mostly better off leaving it a gray area maybe until Oracle gets more aggressive with it.
322.
by xiphias2 2y ago
Who is allowing this BS in HN? This subdomain should be banned.
323.
by xiphias2 2y ago
Bukele realized that it's more profitable to leave the legal tender and get a loan and buy Bitcoin from it instead.
324.
by xiphias2 2y ago
I had the same (compiled llama.cpp myself). Changed it to all CPU I think (num layers on GPU to 0) and it went up to 1.8 tokens per second. I think it can go up much more
325.
by xiphias2 2y ago
Maybe VLLM is better at inferencing MoE (also you can set the number of experts to use). In theory half of the model fits to RAM, so it should be GPU limited if memory management is smart.
326.
by xiphias2 2y ago
Has it been tried on 128GB M4 MacBook Pro? I'm gonna try it, but I guess it will be too slow to be usable. I love the original DeepSeek model, but the distilled versions are too dumb usually. I'm excited to try my own queries on i
327.
by xiphias2 2y ago
While I see these things popping up, for queries about US politics I see ChatGPT more detailed / correct, Deepseek is less biased. ChatGPT sometimes just say ,,controversial'' for some things even if I ask it to be explicit a
328.
by xiphias2 2y ago
I believe you that it had to do with the selloff, but I believe that efficiency improvements are good news for NVIDIA: each card just got 20x more useful
329.
by xiphias2 2y ago
It was all creative disruption, Kodak again. It's easier to act as EVs are not coming than change the old investor base.
330.
by xiphias2 2y ago
It's funny because this simple excercise shows all the problems that I have using the reasoning models: they give a long reasoning that just takes too much time to verify and still can't be trusted.

More ›