Apparently the DENIC team was on a party this evening! Party hard, but not too hard. https://bsky.app/profile/denic.de/post/3ml4r2lvcjg2h

A real party killer if I have ever seen one.

At least all of the appropriate people were in a room together when the outage happened

Sounds like poor risk pooling. If that room crashed, we'd have nobody to fix this.

nation state actor picking right time to sabotage a tiny part of the key rotation process. on monday someone cut major fiber lines, on tuesday DENIC is failing.

maybe someone is showing off?

Unironically yeah, we are at the level of weaponizable sophistication that this metaphorical dick waving you are suggesting is probably something that happens

Interesting "bus problem" to have in a scenario where everyone who is qualified, experienced and trusted enough to commit lives changes (or perform a revert, undo results of a botched maintenance, etc) in an emergency situation is not completely sober.

Sobriety is just factor to be weighed in an emergency situation. 30 years ago I was at a ski resort with about 50 friends having a drinking competition in the resort's main bar. Late that night two ski lodges collapsed, trapping people inside. Around midnight, soon after the winner was announced, the police entered and asked "who's able to drive a crane truck?" The winner of the competition put his hand up and informed them of how much he had had to drink. Don't care they said, so he drove a crane big enough to lift a building up a single lane 35km mountain road in nighttime ice conditions. (The crane made it, but sadly most of the people in the ski lodges didn't. https://en.wikipedia.org/wiki/1997_Thredbo_landslide )

Sounds like Australian police. I remember 15 or so years ago being in a big team assisting the Australian police with something on a remote farm. There were 20 people that needed to be taken back to base and one 10 seater car. Someone asked the police if everyone could get in the car and policeman shrugged and said you can try. So the policeman drove a four wheel drive across farmland with 16 people stuffed into the back.

Sounds like Europe, yes.

Just ask it about the "Tiananmen Square protests and massacre". :-)

On the other hand, asking ChatGPT about "Hiroshima US atomic bombs", isn't too much better.

run it hosted on another provider - it's less censored there

Especially not as a solo dev. That would be way to expensive.

I am using DS4 via cortecs.ai. There is no training and it is GDPR-compliant. The flip side, it is expensive.

This. I tried Zed for an entire month, but this "search thing" drove me nuts. It is also slow. If you work in a large project search is absolutely essential. Too bad.. Back to Visual Studio Code.

They both use ripgrep, weird for it to be slower? Especially with the multibuffer it's more pleasant to use

I think VSCode use ripgrep and Zed has its own ripgrep-like search. Zed likely still do more work per match due to the multibuffer. A normal nested tree-based result should be faster.

I think multibuffer can be good in edit/renaming use cases, but it's very annoying for fast lookups/navigation across different files (as mentioned elsewhere).

IDK if both use ripgrep it is even more strange why it's so slow in Zed.

Yes, that is insane. Or said in another way, they simply didn't had any working backup strategy!

To be 100% fair, having only one provider for backups is really risky. A minimum 3-2-1 would be better

Is that why they call it S3?

Principle of most surprise.

At least from an European perspective it is impossible to use DeepSeek v4 right now, as there are no privacy safe offerings.

Where or how can I use this model with a DPA and better privacy terms? Are there EU friendly hosters already? Would love to use it.

> better privacy terms

DeepInfra, as far as I'm aware, doesn't log your prompts and doesn't retain them in most cases, except "debugging purposes". As their per their privacy policy[1]: "We understand that the inputs you provide to our API and the outputs it generates may contain your Personal Information. We will not store, sell, or train using this data unless we have your explicit consent. We might sometimes store, for a limited period of time, the inputs and outputs to API calls for debugging purposes."

They're not EU-based, though. And I'm not sure how "private" their inference actually is. The throughput is also not the best everywhere, sometimes it can be really slow (although right now both DeepSeek-V4 models seem to be doing fine). However, they have a good pricing, probably on of the best on the market.

I'm not affiliated with them in any way, but when I want to test (I'm not a power user of LLMs, chatbots and agents, not at all; I'm doing it just out of the curiosity) something that is too big for my local hardware, DeepInfra is usually being my go-to provider.

[1] https://deepinfra.com/privacy

Fascinated, a bummer that DeepSeek does not offer a DPA or opt-out for training. This renders it unusable for my use cases unfortunately. At least z.ai GLM has a somewhat DPA in Singapore.

The weights are open and you can use the model with any third party provider that gives you the DPA you want.

For my use-case, I want the providers to get my tokens as long as they plan to keep releasing open-weight models

Yes, looking at things from the same angle every time and not really representing the alternative view is indeed a crusade