Yes, a link requires user interaction. But what if the attacker decides to render an image instead and put the secret data in the query params? Loading an image is a way to trigger a request without user interaction
I would even go one step further: why would an AI tasked with summarizing external comments require any tool access, especially being able to reach into non public data?
The solution to prompt injection is not more AI on top of it, it starts with data access controls
> SANA-WM uses only ~213K public video clips with metric-scale pose supervision, completes training in 15 days on 64 H100s, and generates each 60-second clip on a single GPU; its distilled variant runs on a single RTX 5090 with NVFP4 quantization to denoise a 60s 720p clip in 34s.
How does the archive they provide look like? Many zip files?
I would like to retrieve them and offload to another storage service but I don’t have local storage enough to hold all of it at the same time, unpack and then reupload. I would need to do it in stages.
All of this is meant for 100% public data, right? Or is there a concept of visibility control? Can I create private communities, with data flowing just inside?
I’m always surprised when they (GPT 4.1 in my case) manage to get all of the closing parentheses right! I’d have guessed that having such a sequence of same characters would be a challenge for the LLM to use the right amount
It’s not that bad, you are still able to control your position and nudge the spawned enemies away, and force the boss to kill these enemies for you. And even if you get cornered here and there there’s plenty of time and space in this fight to heal back.
Do you approve on every update of the package? Do they offer a way to quickly review what’s going to run and what has changed since the last approval? Otherwise it’s just like another checkbox of “I confirm I read the terms and conditions”