python273
The problem is that Gmail will bounce any emails from DigitalOcean IP, even if you sit on this IP for years (so no recent spam), even if replying to someone, even if you registered as 'Postmaster' on Google.
So if you want to selfhost, you'll first need to find an IP that's not blocked to begin with.
There's not much incentive to subsidize prices for OpenRouter providers for example, and the prices are much lower than the $6.37/M estimate from the article.
https://openrouter.ai/meta-llama/llama-3.3-70b-instruct
avg $0.37/M input tokens, $0.73/M output tokens (21 providers)
Llama is not even a good example, as the recent models are more optimized using Mixture Of Experts and KV cache compression.
> It is noteworthy that the article of the Administrative Offenses Code (KoAP) on discrediting [army] was most actively applied in 2022: that year, courts reviewed 5,662 protocols and fined 4,440 people a total of 151.3 million rubles. [1.7 mil USD]
> Last year, the number of protocols almost halved to 3,053, as did the number of those punished — only 2,361 people were held accountable, and the total amount of fines imposed was 76.6 million rubles. [858,721 USD]
> The number of people convicted for spreading knowingly false information about the activities of the armed forces (Article 207.3 of the Criminal Code) increased from 14 to 65 (almost fivefold) last year. Of these, 30 people were sentenced to imprisonment: most often, sentences under this article ranged from five to eight years, but two sentences fell into the category of eight to ten years. For comparison: in 2022, only two of the 14 people convicted for fakes received actual imprisonment, with the harshest sentence falling into the "5 to 8 years" category.
> The number of people convicted for justifying terrorism continues to grow: from 318 in 2022 to 345 in 2023. However, the number of sentences for public calls for violent change of the constitutional order in the media or on the internet has decreased from 334 to 281. [...]
Anyway, seems like you made up your mind and there's zero point debating it with you.
Using a legal entity in a more convenient country for a startup seems like a common practice, including listing the address of such entity on the website. You'd be surprised how many companies are incorporated in America, pay taxes there, but have founders/employees/contractors elsewhere around the world.
So, I personally wouldn't count it as active effort of "trying to hide" or "trying to mislead".
> The report mentioned that their LinkedIn profiles changed from showing Moscow to Tbilisi. I'm sure I could also change my location to Tbilisi on my LinkedIn profile. How is that a meaningful argument?
Again, not sure why assume malice intentions. I also updated my Linkedin location when I left Russia, is that surprising?
> Why do you care to defend them so much?
Pavel pays me 15 rubles per comment of course! (tbh not sure why I waste time on this :D)
> send money (indirectly) to Russia
Even the report mentions the team members moving to Tbilisi, Georgia. Afaik Pavel moved to Dubai and still has Ukrainian citizenship. So I doubt a significant portion of company's money ending up in Russia, maybe except salaries of a few engineers. But it's pennies compared to how much the regime is paid for the resources, if that's what you worry about.
Also made some far fetched connections of Flipper Devices to companies owning the hackspace Pavel Zhovner worked in, and attributed his trolling and making anti-censorship tools "as actively supporting the authorities in Russia". lol.
I'm also using Narrator in my HN reader (no settings for voice right now though, except localStorage['cfg-narrator-rate'] via devtools):
> The math questions were of the form “Is 17077 prime”? They picked 500 numbers, but all of them were prime!
> The June version of GPT-3.5 and the March version of GPT-4 almost always conclude that the number is prime regardless of whether it is prime or composite. The other two models do the opposite. But the paper only tested prime numbers, and hence concluded that GPT-3.5’s performance improved while GPT-4’s degraded.
Too late.
NTechLab scraped vk.com (Facebook-like Russian social network) profile photos and were running a public website "FindFace" that allowed to find profiles by photo of a face. They provide services to Russian government, e.g. it's used to find people from protests. Also FindFace was used to find and harass webcam models.
https://www.businessinsider.com/intel-spacex-philip-morris-l...
> Intel, SpaceX, Philip Morris, and dozens of other US companies were in a leaked database of users for a Russian facial recognition company
> The NTech Lab user list, which was shared with Insider by an anonymous source, includes more than 1,100 entries, with businesses and government agencies from more than 60 countries.
https://en.wikipedia.org/wiki/FindFace
> In 2017, NtechLab face recognition algorithms were built into the Moscow city video surveillance system operated by the Moscow Department of Information Technology . The system uses the database of the Ministry of Internal Affairs to find correspondences to it on video. The alleged use of the system is the search for criminals and the fight against terrorism.
> In 2022, NTechLab was accused of assisting the Russian and Belarus government by tracking thousands of political activists which led to their unconstitutional detentions and arrests.
> In 2016, FindFace generated controversy because it was once used to deanonymize Russian pornography actresses and alleged prostitutes. These efforts were organized by users of a Russian imageboard Dvach who claimed that women in the sex industry are “corrupt and deceptive”, according to Global Voices. In addition, FindFace has been characterized as a major step in the erosion of anonymity.
https://hn.cns.wtf/#32380045 - Also I have HN reader with comment reading feature, it's not so polished though (no settings, no announcement between comments)
https://github.com/python273/hn-index - CLI tool to search stories via regex, I use it to explore topics
Previous HN discussion, people who actually checked the repo are confused:
https://news.ycombinator.com/item?id=30097371
Ars Technica reporter:
https://twitter.com/dangoodin001/status/1486976676965715968
Actual iOS security researchers:
https://twitter.com/qwertyoruiopz/status/1486715602240557074
[0] https://stackoverflow.com/questions/20709549/drop-unwanted-c...
[0] https://devguide.python.org/#status-of-python-branches
[1] https://pypistats.org/packages/requests (stats for my packages look weird starting November btw, so might need to look at October data)
https://docs.google.com/spreadsheets/d/1fswG7v8wKFoK06ilbF9V...
> Circle A has 1/3 the radius of circle B, and circle A rolls one trip around circle B. How many times will circle A revolve in total?
bcrypt has 72 bytes password length limit [0]. If the password is shorter, it gets extended to 72 bytes anyways, so longer passwords (10 chars vs 70 chars) are not making time difference.
The password can be pre-hashed to allow arbitrary sized passwords [1], but I don't think 10 chars vs 10000 would make a significant difference. I think most of the time will be spent on bcrypt rounds, not on pre-hashing.
> Happened to me.
While trying to find stretches quote, I found this post [2], so maybe it was not bcrypt? :)
[0] https://en.wikipedia.org/wiki/Bcrypt#Maximum_password_length
[1] https://en.wikipedia.org/wiki/Bcrypt#Solution_3_-_Pre-hash_p...
[0] https://duo.com/labs/tech-notes/writing-an-xdp-network-filte...