josalhor

7 ms·

31.
by josalhor 10mo ago
From GPT 5.1 Thinking: ARC AGI v2: 17.6% -> 52.9% SWE Verified: 76.3% -> 80% That's pretty good!
32.
by josalhor 10mo ago
"Right to Human Verification" is something I have actually thought about a lot. I want to able to verify my identity against a system. I also want to be able to not do that . So for instance, on Twitter/X, I could verify mys
33.
by josalhor 1y ago
Seems almost incredible that no one is pointing out the YT incentives behind this. YT has a war on two fronts: ad-blockers and in-video sponsors. That is because in-video sponsors don't get YT money, so they want to be in the monetizat
34.
by josalhor 1y ago
Having had no experience in JIT development but having followed the faster cpython JIT progress on a weekly basis, I do find their JIT strategy a bit weird. The entire decision seemed to revolve around not wanting to embebed an external JIT
35.
by josalhor 2y ago
I have thought about this for a while. I think that a lot of labeling could be done with regards to promises of consumer products. Publishers (of any kind of digital product) cannot expect to transition to a model where we don't own an
36.
by josalhor 2y ago
Do you have a minimal docker image so we can start the service with a single command?
37.
by josalhor 2y ago
I just skimmed over LoRA+ and DoRA and I see no reason why these improvements could not go hand in hand. Actually, LoRA+ seems to be about efficient training while DoRA seems about improving the ability to actually learn, making it signific
38.
by josalhor 3y ago
Those numbers are quite impressive for a 7B model!
39.
by josalhor 3y ago
I read that as "big organizations can make significant contributions to reduce CO2 emissions", which makes sense to me.
40.
by josalhor 3y ago
What are the constant factors here? Could this be dropped in everywhere they are already using Shor's algorithm and just get a speed up? If so, this results seems incredibly influential.
41.
by josalhor 3y ago
I can select English UK and that seems to work pretty well.
42.
by josalhor 3y ago
Source is HT Kim shared on NYTimes: https://www.nytimes.com/2023/08/03/science/lk-99-superconduc...
43.
by josalhor 3y ago
As someone who has never programmed directly for a GPU, how does this compare to HIP? Can this be an efficient abstraction over Nvidia and AMD GPUs?
44.
First Republic Bank Enters Free Fall(nytimes.com)
3 points by josalhor 3y ago|1 comments
45.
by josalhor 4y ago
Everyone here seems pretty negative about the petition based on the title that says "YC Is Asking for a Bailout", when actually the content of the post is quite different: > Small business depositors at Silicon Valley Bank shou
46.
by josalhor 4y ago
The examples in the paper are pretty impressive. There is an example of a windows 11 dialog image. The computer can figure out which button to press given the desired outcome of the user. If one where to take this model and scale it, I can
47.
by josalhor 4y ago
Also from the paper for more context: > The dataset was randomly divided into 70% training subset (1420 for prostate and 692 for pancreatic), 15% validation subset (304 and 148), and 15% test subset (304 and 148).
48.
by josalhor 4y ago
From the paper and for contextualisation: > In the same manner, the trained deep learning for pancreatic cancer urine dataset (Figs. S13a–c) clearly showed superior classification performance with a sensitivity of 98.6% and a specificity
49.
by josalhor 4y ago
Totally agree. Something similar should also apply to laptops and smartphones. If your phone comes up with 30GB of bloatware on a 64GB device you should definitely not be allowed to advertise 64GB of space.
50.
by josalhor 4y ago
Absolutely agree. And while I don't recommend anyone to try this out on a DIY basis, historically a lot inventions we take for granted and have made our life better came from people literally tinkering around with stuff and slowly fi
51.
by josalhor 4y ago
> The guy that you can see in one of the pictures drinking from a glass recipient was in a Netflix documentary (I can't remember the title from top of my mind) where he was advocating for everyone being able to use genetic engineeri
52.
by josalhor 4y ago
Makes me wonder if we will see a generalization of this idea. Just like in a CPU 90%+ of want you want to do can be modeled with very few instructions (mov, add, jmp..) we could see a set of very refined models (Stable difussion, GPT, etc)
53.
by josalhor 4y ago
Fair enough. But just like every job has bullshit, every part of life has some level of bullshit. I don't think in terms of "eliminating" bullshit, but rather minimizing it.
54.
by josalhor 4y ago
I've seen people work in tech working on more efficient renewable energy, improving crop harvesting, improving safety on the road... All while having a happy team of well-organized people. Not every job is bullshit and not every way of
55.
by josalhor 4y ago
I saw this idea some time ago also in HN [0,1], it has also stuck in my brain pretty hard. The way Paul words it is like this: > Having kids showed me how to convert a continuous quantity, time, into discrete quantities. (...). If Christ
56.
by josalhor 4y ago
Yes, there is evidence that is the case[0]. Part of the abstract: "We provide the first evidence that cumulative heat exposure inhibits cognitive skill development and that school air conditioning can mitigate this effect. Student fixe
57.
Finnish President says Turkey agrees to back Finnish and Swedish NATO membership(reuters.com)
17 points by josalhor 4y ago|1 comments
58.
by josalhor 5y ago
How efficient is this? Is JSON stored as a blob and parsed and filtered on each query?
59.
by josalhor 5y ago
My bad, you're right. I mistook "2m" as 2 minutes ago, not two months.
60.
by josalhor 5y ago
Seems fixed? I just checked with FF. Also from a Reddit Admin: > We believe we've tracked down the issue that was impacting some of your bots. There may be some lingering 403 issues as we hash it all out over the next day or so >

More ›