Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
Ukv
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
25 ms
·
331.
▲
by
Ukv
2y ago
The protests here were non-violent and non-destructive, to my understanding. But in all cases, what I'm conjecturing is that the "actions have consequences" framing distracts from challenging the authority's own actions.
332.
▲
by
Ukv
2y ago
Maybe a tangent, but I feel framing punishment/retaliation as "consequences" (remarking along the lines of "it's almost as if actions have consequences" ) often serves the purpose of treating the retaliation a
333.
▲
by
Ukv
2y ago
I think that comes back to: > Here the use is fairly frivolous, but most such character sets do have legitimate use - like fullwidth latin characters for readability when used alongside CJK - so really shouldn't just be skipped. The
334.
▲
by
Ukv
2y ago
To me normalization seems preferable over just skipping the text in all situations including math, and it's trivial to do, so I think makes for a sensible default. Certainly they could then start adding special cases, reading out symbo
335.
▲
by
Ukv
2y ago
> could explicitly add these cases as letters, Unicode already defines compatibility equivalence[0]. Normalizing seems like the most sensible default for me - then it'd handle this, ᵗʰⁱˢ, and so on. > though that would make the r
336.
▲
by
Ukv
2y ago
Surprised at how poorly most of the screen readers seem to handle it. Will avoid the characters knowing this (not that I used them much to begin with), but I do also feel it's something that should be fixed on the screen reader side. C
337.
▲
by
Ukv
2y ago
This article is new, but to be clear the disks started rotting within ~5 years of purchase, not just now after 20 years. The issue has been known for a long time, like in 2012: > > Apparantly most releases from Warner Bros on HD-DVD h
338.
▲
by
Ukv
2y ago
skyyler is refuting a comment that implies this is expected by mentioning that their own (non-defective) discs have lasted ~50 years and counting.
339.
▲
by
Ukv
2y ago
If the books falls apart not from regular wear and tear but from a manufacturing defect that gives it a far shorter life than expected/advertised - then likely yes.
340.
▲
by
Ukv
2y ago
To be fair, up until recently copyright/DMCA prevented them from repairing the ice cream machines - I'd give it some time to see if things start to improve. Other kitchen equipment does seem to generally be fixed/replaced pro
341.
▲
by
Ukv
2y ago
In addition to testing in advance that your own (hard|firm|soft)ware will work, some people will just want to get new features early and are willing to deal with the potential extra bugginess.
342.
▲
by
Ukv
2y ago
> If it works I'll say "I can automate you away". I don't think deciding not to use LLMs makes you any more immune from automation, because other devs will. If anything, assuming it really does increase productivity,
343.
▲
by
Ukv
2y ago
Author does also say: > Or: 'I just used it to help with the research .' (More and more, I have been forbidding outside research in these papers for this very reason.
344.
▲
by
Ukv
2y ago
While widespread LLM access may have exacerbated the problem, for courses with grades determined solely by uncontrolled out-of-classroom assignments there were almost definitely already a good chunk of students essentially just paying their
345.
▲
by
Ukv
2y ago
Many platforms have followed the trend of first offering a legitimately good product to build up a userbase, then squeezing out profit with increasingly anti-user changes, exploiting the fact it's difficult to switch (network effect an
346.
▲
by
Ukv
2y ago
> it wasn’t clear to me why or how articles are rated the way they are [...] Take this op-ed: “People like Andrew Tate have no place in the American conservative movement,” by Josh Hammer [...] The AI says that this piece aligns with a “
347.
▲
by
Ukv
2y ago
> The new model seems to prove that longstanding rumors of diminishing returns in training unsupervised-learning LLMs were correct and that the so-called "scaling laws" cited by many for years have possibly met their natural en
348.
▲
by
Ukv
2y ago
> Each new training from scratch is a perfect blank slate [...]? I don't think training runs are done entirely from scratch. Most training runs in practice will start from some pretrained weights or distill an existing model - takin
349.
▲
by
Ukv
2y ago
> there is a lot of evidence that many of the concepts ARC-AGI is (allegedly) measuring are innate in humans I'd argue that "innate" here still includes a brain structure/nervous system that evolved on 3.5 billion yea
350.
▲
by
Ukv
2y ago
> Interesting, but there is a lot of "intent" in writing notes and I am not convinced it could capture the full picture without significant human supervision. [...] At least when I write, I know it's correct. To my underst
351.
▲
by
Ukv
2y ago
> And let me make this clear. I, as your patient, I never NEVER want the AI's treatment plan. If you aren't capable of thinking with your own brain, I have no desire to trust you with my health, To my understanding this tool is
352.
▲
by
Ukv
2y ago
> Let me re-phrase that then to the main cause of Black over-representation in prison being bias. Because yes, one can always find some bias, but does it contribute to 1-10% of the outcome, or 80-90%? I think the answer is "depends
353.
▲
by
Ukv
2y ago
I don't think it's fair to call anti-Black bias in the legal system a false belief/conspiracy. Given a Black suspect and a white suspect, do you believe investigators/prosecutors: A. Have a uniform prior between the two
354.
▲
by
Ukv
2y ago
> but clicking on the actual link in front of it, takes you to the discussion? Does it? I see 1. ▲ Page title (domain.foo) 123 points by user 4 hours ago | flag | hide | 5 comments Where clicking "(domain.foo)"
355.
▲
by
Ukv
2y ago
> The “AI” field as a whole has bought into racist ideas on intelligence born out of eugenics [...] Shane Legg, co-founder of DeepMind [...] cited [Linda Gottfredson's] definition of intelligence in the introduction to his 2008 PhD
356.
▲
by
Ukv
2y ago
To my understanding, this is about the fact that Bing and Microsoft Copilot (formerly "Bing Chat" - not GitHub Copilot) base results/snippets/answers on pages as they were when they were last indexed. So a repository,
357.
▲
by
Ukv
2y ago
Skimming through the new features, I'm looking forward to the new "Game" tab in the editor ( https://godotengine.org/article/dev-snapshot-godot-4-4-beta-... ). Being able to easily select and move around t
358.
▲
by
Ukv
2y ago
> If Google has failed so far to resolve mem safety issues in their decades old giant code base, then I'd rather hear standardization ideas from someone who succeeded. If G succeeded at resolving those issues, then that's a con
359.
▲
by
Ukv
2y ago
> In the informational theory sense. A word with broad scope will be more used, thus be more frequent in text, and thus, carry less information. A rare word carries more information when it is used , but by nature of being rare is infre
360.
▲
by
Ukv
2y ago
He's allowed to reject Rust within DMA, which he maintains. What he wasn't allowed to do was veto code that was merely a user of DMA, on the basis of objecting to R4L in general ("I will do everything I can do to stop this&qu
More ›