2 bugs will occur at 100 thousand years later and 2.1 billion years later

[kshadow1000]

100 thousand years later the date will take 2 lines.

2.1 billion years later a "Segmentation fault." will occur.

[Cheaterman]

That gives them plenty of time to fix it, then.

[AmyrAhmady]

I'm coming from future and I can confirm tty-clock does not exist anymore. I would've stayed here to talk but I'm afraid I must go right now.

[livcher]

Longtermists be like:

[pompompur1nn]

yeah. Can confirm this exploded 900 systems. In the future. Please fix now.

[Jelly-Fish-Dev]

Tested it with libfaketime, appears to have been fixed?

faketime "101059-11-18" ./tty-clock

[Citrullin]

Please tell me that's AI slop.

[EverStarck]

valid concern btw

[lowlune]

Please tell me that's AI slop.

you are AI slop

[sumedhahire]

This is a critical bug we have a very short amount of time left

[NecA970]

when this fails there can easily be 800 trillion dollars in damages (currently 5$ adjusted for inflation)

[transdeveloper]

billions must patch future bugs!!

[AusaafAhmad]

fair

[exilonium]

The only reason we're not using it in prod cause this will break out system in the future

To mitigate this issue
We're making an api request to claude to get the current time.
It works like a charm 🫡

[yarrseni]

The only reason we're not using it in prod cause this will break out system in the future

To mitigate this issue We're making an api request to claude to get the current time. It works like a charm 🫡

You shouldn't rely on only one provider, it's quite unsafe. The requests should go to all major LLM providers and fed into a locally deployed triage agentic system of open weights LLMs (3+, but tailor to your SLAs).

[exilonium]

You shouldn't rely on only one provider, it's quite unsafe. The requests should go to all major LLM providers and fed into a locally deployed triage agentic system of open weights LLMs (3+, but tailor to your SLAs).

That's a quite nice system that we tried to implement but claude seems to be most accurate
We've created a weighted chain of the delay in request and response so the
Time comes out to be with delay of +-2 milliseconds that's quite accurate

Openai gpt 4.1 has a variable response rate so we get the time with +-8ms accuracy

Tho i think implementing the traige system will be most helpful

Either way we're using tty clock as a fail safe in case the connection drops or we're out of ai quotas

(It's currently costing us ~2$ per minute)

Tho i really love the engineers we're working with cause they found the solution this problem with only a single cursor prompt 🥰
(Saved us a dollar)