cubefox

32 ms·

931.
by cubefox 9mo ago
Microsoft Windows has a similar issue. I wonder why nobody is talking about that on HN.
932.
by cubefox 9mo ago
I wonder how this one works: https://youtube.com/watch?v=xlcywgEMuGI
933.
by cubefox 9mo ago
It needs far more features apparently. Tons more. That's why Notepad++ is popular. Which also had a severe security vulnerability recently. Which was actively exploited by some state actor like China.
934.
by cubefox 9mo ago
I was disappointed when Imagen 4 (and therefore also Nano Banana Pro, which clearly uses Imagen 4 internally to some degree) had a significantly stronger tendency to drift from photorealism to AI fake aesthetics than Imagen 3. This suggests
935.
by cubefox 9mo ago
Yeah, though I there is the same issue the other way round: Great prompt understanding doesn't matter much when the result has an awfully ugly AI fake look to it.
936.
by cubefox 9mo ago
Apparently image models have to choose between aesthetics and photorealism. Many aren't good at either.
937.
by cubefox 9mo ago
Okay, I haven't really tested hallucinations like this, that may well be true. There is another weakness of GPT-5 (including 5.1 and 5.2) I discovered: I have a neat philosophical paradox about information value. This is not in the pre
938.
by cubefox 9mo ago
A similar idea occurred to the Austrian-Americam cyberneticist Heinz von Foerster in a 1960 paper, titled: Doomsday: Friday, 13 November, A.D. 2026 There is an excellent blog post about it by Scott Alexander: "1960: The Year The
939.
by cubefox 9mo ago
While I don't doubt this was one influence, there was also an infamous problem with Dall-E 2, which was perfectly able to generate an astronaut riding a horse but completely unable to generate a horse riding an astronaut. This problem
940.
by cubefox 9mo ago
> I have generated my own using your prompt and post it in the previous comment. Yes, and it has a very unrealistic AI look to it. That was my point. > You haven't posted a z-image one of Atlantis. Yes, I don't doubt that it
941.
by cubefox 9mo ago
It seems you are ignoring the fact that the NBP Atlantis pictures looks much, much worse than the z-image picture of the cafe. They look far more like AI slop. (Perhaps the Atlantis prompt would look even worse with z-image, I don't kn
942.
by cubefox 9mo ago
> I disagree, nano banana pro result is on a completely different league. I strongly disagree. But even if you are right, the difference between the cafe shots and the Atlantis shots is clearly much, much larger than the difference betwe
943.
by cubefox 9mo ago
In your example, z-image and Nano Banana Pro look basically equally photorealistic to me. Perhaps the NBP image looks a bit more real because it resembles an unstaged smartphone shot with wide angle. Anyway, the difference is very small. I
944.
by cubefox 9mo ago
The examples I saw of z-image look much more realistic than Nano Banana Pro, which is likely using Imagen 4 (plus editing) internally, which isn't very realistic. But Nano Banana Pro has obviously much better prompt alignment than some
945.
by cubefox 9mo ago
The complex prompt following ability and editing is seriously impressive here. They don't seem to be much behind OpenAI and Google. Which is backed op by the AI Arena ranking.
946.
by cubefox 9mo ago
In my experience, when I asked Gemini very niche knowledge questions, it did better than GPT-5.1 (I assume 5.2 is similar).
947.
by cubefox 9mo ago
Apparently the practical power difference is smaller than what the spec sheets suggest, otherwise you wouldn't get this rather long battery life, especially considering that it also has OLED.
948.
by cubefox 9mo ago
The article said the M5 has significantly higher single core CPU performance, Panther Lake has significantly higher GPU performance. The Panther Lake devices had OLED screens, which consume significantly more power than LCDs, so they were a
949.
by cubefox 9mo ago
Here is a more recent comparison with Intel's new Panther Lake chips: https://www.tomsguide.com/computing/cpus/panther-lake-is-int...
950.
by cubefox 9mo ago
... yet.
951.
by cubefox 9mo ago
From his previous blog post: > Generally, I believe [Rabbit] R1 has the potential to change the world. This is a thought that seldom comes to my mind, as I have seen numerous new technologies and inventions. However, R1 is different; it’
952.
by cubefox 9mo ago
Sounds like he was confused but genuinely interested in cryptology, which contradicts the cynical narrative about him only donating for social reasons.
953.
by cubefox 9mo ago
That should be easy to test: test a 16 bit model on various benchmarks, once with fresh context and once with the context filled up with irrelevant tokens. Record the relative performance degradation, and then do the same for a quantized mo
954.
by cubefox 9mo ago
Sounds interesting, but... > these models dominate both exponential attention and linear attention at long-context training There is no exponential attention; standard attention is quadratic. Strange mistake.
955.
by cubefox 9mo ago
That website says nothing about numerical error potentially causing context rot.
956.
by cubefox 9mo ago
The XXX conglomerate.
957.
by cubefox 9mo ago
Okay, then let's see whether we are going to see real linear architectures, like Gated DeltaNet or Mamba-3, in some larger models. I don't believe there is a "lower bound" which states that those can never get to (or exc
958.
by cubefox 9mo ago
I think DeepSeek V3.2 is sub n^2, but it clearly performs quite well, refuting the alleged lower bounds in the paper.
959.
by cubefox 9mo ago
Sounds like there is still some way to go: > To fully eradicate the disease, cases in animals (infected by the same species of worm) must also be wiped out. In 2025, animal cases were detected in Chad (147 cases), Mali (17), Cameroon (44
960.
by cubefox 9mo ago
Leave your accusations for yourself. Grok already didn't generate naked pictures of adults months ago when I tested it for the first time. Clearly the "additional safeguards" are meant to protect the system against any jailbr

More ›