my god. 1,000x worse with that context :(
19.07.2025 15:34 โ ๐ 9 ๐ 0 ๐ฌ 0 ๐ 0@msfoil.xyz.bsky.social
extremely online. they/she | ๐ณ๏ธโโง๏ธ
my god. 1,000x worse with that context :(
19.07.2025 15:34 โ ๐ 9 ๐ 0 ๐ฌ 0 ๐ 0also the overwhelming majority of Palestinian Christians are Eastern Orthodox and not Catholic.
19.07.2025 15:17 โ ๐ 17 ๐ 0 ๐ฌ 1 ๐ 0Alexandra โMuad'Dibโ Ocasio-Cortez
16.07.2025 17:26 โ ๐ 3 ๐ 0 ๐ฌ 0 ๐ 0i mean _manic pixie mechahitler_ is right there
16.07.2025 15:24 โ ๐ 32 ๐ 0 ๐ฌ 1 ๐ 0everyone already has big licenses with Microsoft or Google Workspaces both of which bundle, and discount their AI products aggressively.
in Microsoftโs case they can literally say โitโs the same model as ChatGPTโ very hard to compete
screenshot from the Information: Earlier this year, OpenAI told investors it had generated $100 million in revenue last year from ChatGPT Enterprise, which is geared toward larger companies with more than 100 employees. That's about 3% to 4% of its total revenue from ChatGPT. Earlier this year, the company projected $400 million in revenue from the Enterprise product in 2025 with 2 million subscribers, $1 billion in revenue next year with 4 million subscribers, and $6.2 billion in revenue for 2030 with 16 million subscribers.
just the โEnterpriseโ segment
18.06.2025 13:42 โ ๐ 3 ๐ 0 ๐ฌ 0 ๐ 1tweet from no_itsmyturn: Initial report from @IranIntlbrk: According to a report from an eyewitness in Tehran, a building in northern Tehran where high-ranking regime officials were gathering was targeted.
this is what โcyber warโ actually looks like: extremely close / exquisite sigint enabling real time targeting, and better planning to enhance the impact of kinetic effects.
(and because this is BlueSky: this is not an endorsement of Israel but commentary on what โcyber-warโ actually looks like)
google switch to embedding based search as the primary method of finding data sometime around early 2023 and that has been the biggest step change for me
12.06.2025 22:22 โ ๐ 2 ๐ 0 ๐ฌ 0 ๐ 0long, hot, American, cool zone summer
12.06.2025 20:58 โ ๐ 4 ๐ 0 ๐ฌ 0 ๐ 0iOS Text Replacement settings showing emdash with hair width spaces
hell yeahโโโthat looksโโโclean. adding it to my text replacements.
09.06.2025 14:56 โ ๐ 1 ๐ 0 ๐ฌ 1 ๐ 0they are still suppressing lawful protests
08.06.2025 03:24 โ ๐ 1 ๐ 0 ๐ฌ 0 ๐ 0Sharif is a separate organization at the LA County level.
LAPD is City of LA.
yes but LAPDโs statement is just kayfabe, theyโre doing traffic management and letting LASD be the pointy end.
08.06.2025 00:09 โ ๐ 5 ๐ 0 ๐ฌ 2 ๐ 0watching this TT stream until Sky7 gets back up
www.tiktok.com/t/ZTjb77rbR/
distance from home-depot where barricade was rammed to bus off-ramp
430m
07.06.2025 20:16 โ ๐ 16 ๐ 1 ๐ฌ 1 ๐ 0about 300-600m
07.06.2025 20:13 โ ๐ 13 ๐ 1 ๐ฌ 1 ๐ 0d day landing craft color scheme illustration
except the colors are wrong?
07.06.2025 03:30 โ ๐ 3 ๐ 0 ๐ฌ 0 ๐ 0evergiven skeet really
06.06.2025 01:38 โ ๐ 2 ๐ 0 ๐ฌ 0 ๐ 0real โbig boat stuckโ energy
06.06.2025 01:10 โ ๐ 4 ๐ 1 ๐ฌ 1 ๐ 0the picture is giving egg
04.06.2025 06:32 โ ๐ 8 ๐ 0 ๐ฌ 1 ๐ 0setting sampling temperature=0 gets you to 99% deterministic
then setting the random seed gets you to 99.99%
the last 0.01% is from the slim chance of post only collective-comms AllReduce being in arbitrary order
having a ton of empty silos away from populated areas makes second strike survival easier
28.05.2025 22:09 โ ๐ 2 ๐ 0 ๐ฌ 0 ๐ 0itโs also the cheapest and if you have a good conops for sparse loading, silos are very cheap vs additional warheads.
28.05.2025 21:34 โ ๐ 2 ๐ 0 ๐ฌ 2 ๐ 0itโs neat but one must consider if this is helping at a bottleneck of Vulnerability Research.
`find . -type f -name \*.c` also โfindsโ the vuln
The hard part is triaging the candidates and writing a proof-of-vulnerability.
AI is making progress on these fronts too, just has a very long way to go.
their BLDC motors are so incredibly good compared to those made elsewhere.
China can do things at scale/mass production that everywhere else can only match in boutiques/low volume shops.
this is actually something Igor Babuschkin is working on.
all the common methods for LLM training (internet pretraining and RLVR) reinforce a modernist/liberal worldview and moving a model away from that without hurting performance on scientific benchmarks is a hard open research problem.
seems like everything changed after Darlene Zschech left in 2011
11.05.2025 18:20 โ ๐ 6 ๐ 0 ๐ฌ 1 ๐ 0