@GergelyOrosz Yep, I failed to read what you quoted before writing the reply. Then, I understood why you were assuming stuff.
I highly doubt Revolut would risk so much, but I agree they need to release a post-mortem.
I hope they do.
@GergelyOrosz I can only say from earlier this year AWS outage experience that you can have failovers and even with these, you can still fail to recover.
AWS main region (us-east-1) service was so deteriorated that you couldn’t basically do anything.
So unless you hear from someone who lived
@mattpocockuk@AidenAllred2 Codex does this too. It is aware of how many tokens remain before compaction so it efficiently managed stuff.
To be honest I stopped worrying about compaction like a month ago and increased context window to 400K in GPT models because their evals showed no degradation.
Security through obscurity is not a solution. It’s better to have a vulnerability found, than have one replicated without even knowing it exists.
The only measure we have is that more code = more vulnerabilities. I have yet to believe that LLM can write code without them (kinda have the opposite experience).
Hardened code has a lot of scars and it’s much safer.
Projects with good threat models that triage vulnerabilities protect everyone.
You guys release so many changes that I don’t actually even know how to keep up 😭
Alchemy
@samgoodwin89
Effect
@MichaelArnaldi
Any suggestions? Should I have a clanker summarize your continuous updates based on what I’m developing or what?