vitens

718 posts

vitens

vitens

@intervitens

bau bau

Katılım Ekim 2014
528 Takip Edilen541 Takipçiler
Thorne 🌸
Thorne 🌸@ExistentialEnso·
I'm 37, but I'm getting pretty skinny and de-aged my skin pretty effectively with GHK-Cu
English
3
0
17
366
Thorne 🌸
Thorne 🌸@ExistentialEnso·
About a week ago walking at the park catching Pokémon, a kid said she thought I looked like one of the K-Pop Demon Hunters girls And I think that may have been the turning point where I accepted okay, yeah, I'm getting REALLY hot, thank you gym & peptides
English
4
1
45
857
Michael Druggan
Michael Druggan@Michael_Druggan·
This is a bad plan because alpha-gal only makes you allergic to red meat, not chicken. Factory farmed chickens suffer way more than cows or other red meat animals and also produce much less meat per animal. Someone who contracts alpha-gap and substitutes their red meat consumption for factory farmed chicken will increase total animal suffering.
TFTC@TFTC21

A peer-reviewed paper published last year in the journal Bioethics by two professors at Western Michigan University School of Medicine argues that it is "morally obligatory" to genetically engineer ticks to spread alpha-gal syndrome, a permanent condition that makes you violently allergic to red meat. The paper is called "Beneficial Bloodsucking." Their argument: if eating meat is morally wrong, then preventing the spread of a disease that forces people to stop eating meat is also morally wrong. Scientists should gene-edit lone star ticks to enhance their ability to carry alpha-gal syndrome and expand their range into urban environments to infect more people. They call this a "moral bioenhancer." They frame releasing genetically modified disease-carrying ticks as a "vaccination" that only "infringes" on your bodily autonomy rather than "violating" it. The distinction, apparently, is that a tick bit you instead of a government official holding you down. Alpha-gal syndrome is not mild. The CDC estimates up to 450,000 Americans are already affected. Cases have surged 100-fold in the last decade. Symptoms include anaphylaxis. There is no cure. Alpha-gal cases are exploding across the United States. The lone star tick's range is expanding far beyond its historical territory. And two academics at a medical school published a paper arguing this is a good thing that should be accelerated. At what point do we stop treating papers like this as fringe academic exercises and start asking whether anyone is already acting on them?

English
10
1
86
5.9K
vitens
vitens@intervitens·
@Robotbeat if it's a peptide or a small molecule, you could probably get some chinese company to synthesize a batch for you, but if it's a biologic, it's gonna be complicated
English
0
0
5
180
Danielle Fong 🔆
Danielle Fong 🔆@DanielleFong·
ok so are there any decent theories for the physics of the ufos yet or is it kind of a nothingburger of a release. i haven't seen anything
English
39
3
63
7.4K
vitens
vitens@intervitens·
@soirilab It comes out as water vapor, and rains back down later
English
1
0
6
110
vitens
vitens@intervitens·
@catkaldir some of the things may actually work too well/require a lower dose
English
1
0
0
348
Fox Kaldir
Fox Kaldir@catkaldir·
I took a genetic test to tell me how ADHD/Antidepressants/etc would work with me. No idea how accurate it is, but here are some fun charts from it! My understanding is anything in "significant drug interaction" probably wouldn't work for me, but everything else might.
Fox Kaldir tweet mediaFox Kaldir tweet mediaFox Kaldir tweet media
English
11
2
107
15.4K
vitens
vitens@intervitens·
@Nepsuka random korean semiconductor stocks
Indonesia
0
0
0
62
vitens
vitens@intervitens·
@teortaxesTex @Ipenywis Cache only helps with input price though, and they already give you a discount for cache hits, actual reasons imo are them having a really smart large-scale EP/TP inference setup, and probably smaller margins
English
1
0
1
96
Islem Maboud | CoderOne
Islem Maboud | CoderOne@Ipenywis·
DeepSeek must be really burning through cash if they give you this very low pricing for such model The closes provider in terms of pricing is 3.2x more expensive What's their secret?
Islem Maboud | CoderOne tweet media
English
7
1
17
13.9K
vitens
vitens@intervitens·
@Xenoimpulse btw do you by any chance have an opinion about Dr. Will Powers and his theories? I'm a little bit interested in his recent not-trans-related work, but most of his notoriety and earlier work seems to be about trans people
English
1
0
1
233
vitens
vitens@intervitens·
@xXshaurizardXx I personally prefer Honeywell OD7+ @ 200-532nm laser safety glasses as my blue light glasses
English
1
0
1
68
shaur
shaur@xXshaurizardXx·
The only thing you need has been & will continue to be good circadian lighting & we should ban all blue light glasses short of the oakley yellow prism lenses while we're at it
Tyler Black, MD@tylerblack32

DONT WORRY ABOUT "BLUE LIGHT" AND SLEEP! The "blue light is wrecking your sleep" narrative is one of the stickiest health myths of the decade. As a psychiatrist who treats insomnia, I'd like it to die. The mechanism is real. The screen-sized version of the story is not. /1

English
1
0
2
288
vitens
vitens@intervitens·
@scaling01 I'd love an actually good dense 128b for local use with dflash+dtree
English
0
0
4
600
vitens
vitens@intervitens·
@V4ldeLund @norpadon scattermoe (написано на тритоне) работает на амд, по крайней мере на mi325x, ну и если просто инференс, то в vllm/sglang тоже должны ядра быть
Русский
0
0
1
43
vitens
vitens@intervitens·
@max_paperclips true, you shouldn't overload their servers with scraping, use annas archive torrents instead
English
0
0
3
341
vitens
vitens@intervitens·
A proper deployment on lots of nodes with disagg prefill, wide EP and high batch size (see V3 tech report). Semianalysis is only using a single H200 node here, which doesn't allow for high concurrency at a reasonable latency. This also makes nvl72 look disproportionately better, since a lot of the advantage for it is just more GPUs.
English
1
0
2
281
vitens
vitens@intervitens·
@redtachyon okay, but which button do I press to save the 10^100 shrimp?
English
1
0
19
1K
Ariel
Ariel@redtachyon·
Everyone enters a room with a red button and blue button. Before you enter, a superintelligent computer scans your brain and predicts your decision with a 99% accuracy. You can press either one or two buttons. If the computer predicts you will press only the blue button, then the blue button is visibly disconnected, and the red button is active. If at least 50% of all participants pressed the blue button, then every person who pressed the red button receives $1,000,000, and everyone else receives $1,000. If fewer than 50% pressed the blue button, then every person who pressed the red button dies, and everyone else receives $1,000. Pressing the red button always gives you an extra $1,000, regardless of any other outcome. What do you do?
English
29
1
58
8.6K
vitens
vitens@intervitens·
@xXshaurizardXx Claude, build a website for my drinking water company, do not use AI generated images, make no mistakes
English
0
2
57
10.8K
shaur
shaur@xXshaurizardXx·
jerryrigeverything is "starting" a water company, weird but whatever, get that bag. says the site is not AI: "We just polished up the website - with 100% Human made graphics and 0% AI generated images" okay, wtf r these ?? from live urls no less. gemini watermark & all...
shaur tweet mediashaur tweet mediashaur tweet media
English
11
1
224
369.4K
vitens
vitens@intervitens·
@JoshPurtell I've seen chinese model providers do tiered token pricing depending on input length
vitens tweet media
English
1
0
1
280
Josh
Josh@JoshPurtell·
Is there a good reason api pricing is linear in input/output tokens, and doesn't use some superlinear scheme that more accurately tracks gpu usage/flops? When input limits were 4k it didn't matter but now this original sin seems to create a ton of havoc
English
6
1
25
4.2K
🥭
🥭@MangoSweet78·
67.
🥭 tweet media
1
0
10
179
vitens
vitens@intervitens·
dante labs is basically a scam, they might never send the results at all tellmegen is pretty fast (mine was 5 weeks) and lets you download raw FASTA outputs from the sequencer models are pretty decent at handling raw data at this point, but you might still want to read up on some genetics/bioinformatics stuff yourself
English
1
0
6
285