2,117 karma · joined May 10, 2019
<a rel="me" href="https://hachyderm.io/@PeterBronez">Mastodon</a>
Operates more transparently. No concerning findings identified.
• Mullvad (Mullvad)
• TunnelBear (TunnelBear)
• Lantern (Lantern)
• Psiphon (Psiphon)
• ProtonVPN (Proton VPN)
Operates more anonymously. Potentially concerning, but no definitive findings.
• HotVPN (HotVPN)
• LetsVPN (LetsVPN)
• Astrill VPN (Astrill VPN)
• CookieDevs (Cookie, Ciao Proxy Pro)
• VPN Super Inc (VPN - Super Unlimited Proxy)
• PureVPN (PureVPN)
• Potato VPN (Potato VPN)
Concerning and suspicious findings (users should avoid).
• Innovative Connecting (Turbo VPN - Secure VPN Proxy, Turbo VPN Lite - VPN Proxy, VPN Monster - Secure VPN Proxy)
• Autumn Breeze (SnapVPN, Signal Secure VPN - Robot VPN)
• Lemon Clove (SuperNet VPN, VPN Proxy Master Pro, VPN Proxy Master Lite)
• Matrix Mobile (Global VPN)
• ForeRaya Technologies (Melon VPN)
• Hong Kong Silence Technology (Super Z VPN)
• Yolo Mobile Technology (Touch VPN - Stable & Secure, VPN ProMaster - Secure your net)
• Wild Tech (3X VPN - Smooth Browsing, VPN Inf, Melon VPN - Secure Proxy VPN)
I initially thought they were writing custom kernels for proprietary models like GPT-5. They aren't - they're using proprietary models to write kernels for a set of ~250 open Pytorch modules.
Clear communication and transparent accountability are the way.
If you want to make pristine originals available to the masses, seed a torrent.
Grant developed the software originally as a personal tool for his YouTube videos. The software is optimized for his personal needs.
The community version tries to make the tool useful for more people. They’ve built out the docs and apparently improved testing.
An obsidian base is primarily defined as a code block in an existing markdown file. That code block defines a series of queries (filters, really) that produce a result set of markdown file names.
So… more a way to work with a large collection of markdown files than a relational database.
You can find many more options here https://www.co-optimus.com/games.php
Buy a book? You now own a the right to read that book from whatever file. Subscribe to Disney+? You have the right to watch all the marvel movies. Even when those movies are chopped and re-edited into a massive full chronology.
If a digital store ceases operations, you retain the rights to the bits. Maybe you have to do some work to get access to them, but you’re perfectly within your rights to torrent those bits.
Heck, maybe you could like the rights management to distribution. Have a right to the bits? Please seed.
Related, you have to make it benefit the current generation of engineers as well. Want to get your thing built? Deliver your designs in format X, which also happens to support long term reference.
None of this is easy… but it is actually possible to align incentives like this. You just have to do it from very high up, and with a very firm hand.
Meta’s Pyrefly announcement (may 2025)
Netflix post about their overall use of python (March 2013)
Google’s announcement of the Croissant ML metadata format (March 2024)
https://kagi.com/lenses/LdYine8hZtYmrt8yTMngOUtvTM9rmkRy
Kagi Lenses can be defined in many ways, one of which is specifying URLs to search. Unfortunately you can only provide 10 URLs per lens. Here are the ones I chose:
https://stripe.com/blog/engineering, https://engineering.fb.com/, https://www.uber.com/en-US/blog/engineering/, https://netflixtechblog.com/, https://research.google/blog/, https://technology.riotgames.com/, https://incident.io/blog, https://www.anthropic.com/engineering, https://openai.com/news/, https://shopify.engineering/
The main advancement in TFA is using the new Container Swift API for local tool use. That functionality would probably be a welcome contribution to any of these:
https://github.com/Renset/macai
https://github.com/huggingface/chat-macOS
He also explored several other open source AI scale out libraries, and reported that they’re generally way less mature than tooling for traditional scientific cluster computing.
https://www.jeffgeerling.com/blog/2025/i-clustered-four-fram...
Agree context is everything.
Which of course means this is never going to fly with any site that needs to show you ads.
Edit: emailed the address on the site in your profile, got an inbox does not exist error.
You would need to constrain the vocabulary to see any benefits, but that could be reasonable. For example, you an enumeration of numbers, units and metric names could handle dynamic time, temperature and other dashboard items.
For something more complex like offline navigation, you already need to store a map. You could store street names as tokens instead of text. Add a few turn commands, and you have offline spoken directions without on device pre-processing.
Other than the video, the only relevant content is on the about page [2]. It says the voice is a collaboration between 5 different entities, including advocacy groups, marketing firms and a music producer.
The video is the only example of the voice in use. There is no API, weights, SDK, etc.
I suspect this was a one-off marketing stunt sponsored by Copenhagen pride before the pandemic. The initial reaction was strong enough that a couple years they were still getting a small but steady flow of traffic. One of the involved marketing firms decided to monetize the asset and defaced it with blog spam.
It’s a good choice for a robot voice. It’s easier to understand than the formant synths or deliberately distorted human voices. The genderless aspect is alien enough to avoid the uncanny valley. You intuitively know you’re dealing with something a little different.
I don’t think models are doing that. They certainly can retrieve a huge amount of information that would otherwise only be available to specialists such as people with PhDs… but I’m not convinced the models have the same level of understanding as a human PhD.
It’s easy to test though- the models simply have to write and defend a dissertation!
To my knowledge, this has not yet been done.
You could have an arbitrarily large page size. You could use color to encode more data… maybe stack QR codes using each channel of a color space (3 for RGB, 4 for CMYK)
There are interesting accessibility and interoperability trade offs. If it’s print-ready with embedded metadata, you can recover the data from a printed page with any smart phone. If it’s a 1 inch by 20 ft digital page of CMKY stacked QR codes, you’ll need some custom code.
Playing “Where’s Waldo” with a huge field of QR codes is probably still way more tractable than handling PDF directly though!