Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
scarmig
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
18 ms
·
301.
▲
by
scarmig
2y ago
There are always explanations for why the invented persona needs cash, and the scammer also presents sophisticated evidence chains for it. Most people are susceptible to it; see the massive success of spear phishing in corporate environment
302.
▲
by
scarmig
2y ago
Your timeline there is a bit off. When she met this Brad Pitt, she was still married. She started sending money to him; got divorced; got a settlement of €775,000; and proceeded to give all of that as well to Brad Pitt. Still worthy of symp
303.
▲
by
scarmig
2y ago
The people who say "it's to make you interact with normal people!" miss a key point. In the real world, you meet a person, and if they treat you like shit, you walk away. If they physically assault you in the workplace, you c
304.
▲
by
scarmig
2y ago
Google's search engine is worse than Google from 2 decades ago.
305.
▲
by
scarmig
2y ago
Depending on how powerful your model is, a few tokens per second per data center would still be extraordinarily valuable. It's not out of the realm of possibility that a next generation super intelligence could be trained with a couple
306.
▲
by
scarmig
2y ago
1) Push for regulations around data provenance. If you train on anything, you have to prove that you had the rights to train on it. This would kill all but the largest providers in the USA, though China would still pose a problem. You could
307.
▲
by
scarmig
2y ago
Misread your comment as saying that thee wasn't informal =)
308.
▲
by
scarmig
2y ago
In early modern English, thou and thee were both the informal 2pp, with thou being the subject pronoun and thee being the object pronoun. Thou/you is similar to tu/vous. Eventually, thou became archaic, and we think of archaic wor
309.
▲
by
scarmig
2y ago
You're competing against other agents, with some combination of planning, coordination, and motor skills. That drives learning. The video game modality is incidental to the learning process.
310.
▲
by
scarmig
2y ago
Scale and cost are defining considerations of LLMs. By saying they're rolling out to billions of users, they're pointing out they're doing something pretty unique and have confidence in a major competitive advantage. Point bi
311.
▲
by
scarmig
2y ago
Windows Phone was superior to everything else on the market at the time. But phones are an ecosystem, and MS was a latecomer.
312.
▲
by
scarmig
2y ago
If you add multiple hearing points, you massively constrain the space of possible drums. The question then becomes something like "can you see the shape of a drum?" Proof of concept: echolocation.
313.
▲
by
scarmig
2y ago
"Sanguinary" might be the intended word here.
314.
▲
by
scarmig
2y ago
Floats don't have multiplicative inverses, and the floating point operations don't give us any of the mathematical structures we expect of numbers. Floating point division already abandons algebra for the sake of usefulness.
315.
▲
by
scarmig
2y ago
The issue is deeper than that: it's that we take some singular conception of what a dog is, and ruthlessly beat any deviation from that idealized dog out of all the individual dogs. Which ends up being every dog.
316.
▲
by
scarmig
2y ago
Naming the programs gifted and creating a gifted identity is the core issue. Instead, call it something like asynchronous development, and place kids in classes appropriate to their pace of development. I'm hopeful that AI can offer hi
317.
▲
by
scarmig
2y ago
The third prong is a bit badly posed: descriptively, white kids test better than black kids, and each of the three prongs offers an explanation. The third prong points to a discredited belief of genetic inferiority; by positioning the three
318.
▲
by
scarmig
2y ago
Getting rid of a test that measures effects from redlining does nothing to eliminate the effects of redlining.
319.
▲
by
scarmig
2y ago
> This is institutional racism. The opportunity is not equal. The test is not a form of racism, institutional or otherwise. It's doubling as a proxy measure for the socioeconomic disadvantage the students have experienced up to that
320.
▲
by
scarmig
2y ago
Not getting into how we define what are valid complaints vs invalid, because that's besides the point. You'd find it exhausting. People similarly find the race and gender preoccupation exhausting. Your justification for the differ
321.
▲
by
scarmig
2y ago
Imagine a world where literature was preoccupied with height. A character is always slotted in who's explicitly the token short character, and the storyline is always, subtly or not, written to highlight how the world is biased against
322.
▲
by
scarmig
2y ago
There's carry-less multiplication ( https://en.m.wikipedia.org/wiki/CLMUL_instruction_set ), introduced by Intel in 2010.
323.
▲
by
scarmig
2y ago
I grant that the horse one looks pretty solid and interesting. But it's the choice of topics. SciAm has an extremely narrow view of what science is worth publicizing, one that aligns very closely with online causes du jour. Looking at
324.
▲
by
scarmig
2y ago
Looking at https://www.scientificamerican.com/ I see the following front page topics and articles: - Nutrition: It’s Actually Healthier to Enjoy Holiday Foods without the Anxiety - Climate Change: Climate Change Is Altering
325.
▲
by
scarmig
2y ago
That comment is just a fanciful flight of whimsy. > Gwern was the first patient to successfully complete a medical transition to the gender he was originally assigned at birth... his older brother died of a Nuvigil overdose in 2001... hi
326.
▲
by
scarmig
2y ago
There have been multiple times where I read a comment somewhere; thought to myself, wow, this guy is brilliant, let me see who wrote it so I can see if there are other things they've written; and, lo and behold, gwern.
327.
▲
by
scarmig
2y ago
In principle, it seems like you could have semantic watermarking. For instance, suppose I want a short story. There are lots of different narrative and semantic aspects of it that each carry some number of bits of information: setting, char
328.
▲
by
scarmig
2y ago
But, if all the good models can only be trained by large mega corps with close connections to the government, it's only a matter of time until that other LLM will just add its own watermark.
329.
▲
by
scarmig
2y ago
I'm subscribed to all of Claude, Gemini, and ChatGPT. Benchmarks aside, my go-to is always Claude. Subjectively speaking, it consistently gives better results than anything else out there. The only reason I keep the other subscriptions
330.
▲
by
scarmig
2y ago
Although I agree it's a matter of finding the right therapist, I think that undersells the problem a fair bit. There are large barriers to trialing a lot of therapists, and finding the right one can be like finding a needle in a haysta
More ›