I find a llm in a harnes combined with manual ripgrep exploration is really effective of getting codebases. But you font always want to find what you find.
Interresting, Have you done any Møre testing trying to trigger this behaviour?
If it is semi konsistent or even just alone it makes a god case to switch.
Amazing. Next is automating this, automatically or manually gathering all domains they have registered, and spinning up Claude to register a similar domain and creating the counter website.
Just because a measure is less useful doesn't mean it is useless.
There is always a compromise between convince and security and in this case one would have to do even more screening (less convince) for the same security level. Even if airgaps like before increases effort needed for exfiltration.
Its nice if you used local, but needed å beefier modell, or more context Window.
It will eat input tokens, but you do that all the time unless you have input caching.
We'll see. The qwen team has always released a few close to sota but proprietary models in between tgeir open releases. We did get 3.6 35B and 27B so its not all set in stone yet.
Its higley unlikely we get another open llama model though after the llama4 flop, even if their muse spark seems pretty good.