Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
N_Lens
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
32 ms
·
511.
▲
by
N_Lens
11mo ago
Async RL seems to be the main difference in how this model was trained. Impressively they're open sourcing the training framework and weights. However key information missing from the article: - Benchmark comparisons against SOTA model
512.
▲
by
N_Lens
11mo ago
This kind of disgusting behaviour is the default wherever a large asymmetry of power and information exists (Such as with Google senior leadership). This should also make anyone wary of the current attempts at government overreach and surve
513.
▲
by
N_Lens
11mo ago
The core innovation is a verifier-generator dual architecture that enables the model to self-check reasoning rigor, addressing the fundamental problem that correct answers don't guarantee correct reasoning processes.
514.
▲
by
N_Lens
11mo ago
Also the impressive IMO-ProofBench Basic benchmark, the model achieved nearly 99% accuracy, though it fell slightly behind Gemini Deep Think on the Advanced subset. The approach shifts from "result-oriented" to "process-orien
515.
▲
by
N_Lens
11mo ago
An FP8 performance of 3200TFLOPS is impressive, could be used for training as well as inference. "Close to theory efficiency" is a bold statement. Most accelerators achieve 60-80% of theoretical peak; if they're genuinely hit
516.
▲
by
N_Lens
11mo ago
The article fails to explain how the fibers solidify instantly. Reading the actual research paper reveals the critical technical innovation: dopamine accelerates the transition by pulling water away from the silk, and a coaxial needle setup
517.
▲
by
N_Lens
11mo ago
Funny and succinct that the article is behind a paywall.
518.
▲
by
N_Lens
11mo ago
If the subsidies were removed from fossil fuels and applied to solar it would transform the US energy landscape. However this is a difficult proposition as solar is inherently a grassroots energy source that can be deployed by people at the
519.
▲
by
N_Lens
11mo ago
It's not about the arguments, it's about the funding.
520.
▲
by
N_Lens
11mo ago
Idealistic ideas in this paper, but the reality is that it's capital-centered. Same situation with the US healthcare system, but even more so.
521.
▲
by
N_Lens
11mo ago
EFF and ACLU are woefully underfunded and under-resourced for the ever escalating challenges from both public and private sector overreach and encroachment on individual rights and freedoms. I think the confluence of technologies is already
522.
▲
by
N_Lens
11mo ago
I don't agree with your take because it's an example of individual 'freedoms' shitting all over something communal - a caricature of freedom that America has become known for. Cheating ruins online games.
523.
▲
by
N_Lens
11mo ago
FSD keeps causing fatalities and Tesla has tried to obfuscate and delete the crash data on more than one occasion, data which implicates FSD being liable for the crash.
524.
▲
by
N_Lens
11mo ago
I think the title should more accurately state "most populous" because Greater Tokyo is still significantly larger in the built up land area around the Tokyo Metropolis (Almost 2x Jakarta)
525.
▲
by
N_Lens
11mo ago
While Musk's antics and politics definitely have a part to play, it's also obvious that Tesla hasn't innovated nearly as much and their cars are becoming outdated in an EV landscape of constant innovation. Their last big play
526.
▲
by
N_Lens
11mo ago
The frenzy around AI is to do with growth fueled cocaine capitalism seeking 'more' where rational minds can see that we don't have that much more runway left with our current mode of operation. While the tech is useful, the m
527.
▲
by
N_Lens
11mo ago
According to the paper the image models can 'recognize' and track objects in videos. There are a lot of emergent properties in both diffusion models and LLMs that don't align with simplistic descriptions such as 'next to
528.
▲
by
N_Lens
11mo ago
Same experience as others here - shows nulls on most fields even for my commonly used, public facing email addresses.
529.
▲
by
N_Lens
11mo ago
Interesting experiment but the author lists some caveats (Not exhaustive by any means): "Of course, in the short term, there’s a whole host of caveats: you need an LLM, likely a GPU, all your data is in the context window (which we kno
530.
▲
by
N_Lens
11mo ago
Sort of similar vibes as "The children yearn for the mines"
531.
▲
by
N_Lens
11mo ago
26 (Tahoe) has had a lot of teething problems on ALL platforms, and lacks the typical quality and polish of Apple releases. I couldn't believe how many obvious bugs there were on first use across Macbook and iPhone.
532.
▲
by
N_Lens
11mo ago
That old tech was so solid and reliable, I'm always amazed.
533.
▲
by
N_Lens
11mo ago
Post fact justification seems to be a 'feature' of most people's cognitive function, according to the latest research. "The mind is just a bullshit maker".
534.
▲
by
N_Lens
11mo ago
Software just feels so much more ephemeral than hardware. I haven't yet met a single 'old software enthusiast' in my life, yet there are so many enthusiasts for older hardware.
535.
▲
by
N_Lens
11mo ago
Be good if I could read it - anyone have a mirror?
536.
▲
by
N_Lens
11mo ago
Have we wrapped all the way around to <table> layouts again?
537.
▲
by
N_Lens
11mo ago
Salaries for interns in China are low enough that a 5060 is a huge deal.
538.
▲
by
N_Lens
11mo ago
If I only read headlines on HN I'd also say 'EU is rolling back GDPR'.
539.
▲
by
N_Lens
11mo ago
By the time you realise the error of your comment, you'll have reached the age where your opinion can be safely discarded.
540.
▲
by
N_Lens
11mo ago
.NET was a solid choice for backend builds before Node became so popular (And .NET is generally more performant than Node). I hope this churn in .NET builds is temporary because a lot of people might be looking to go back to something stabl
More ›