@sama Sam, this is a great step forward. One thing labs should consider is a much more rigorous malware analysis solution to complement the great code analysis capabilities of the models. Democratizing access to this would be game changing.
Replit, turned 10 🎂
To celebrate we’re making it totally free for 24 hours starting at 5am PT.
But our work—to make coding accessible for all—goes back to 2011.
Watch the highlights from the journey:
It’s been an honor to help millions learn & ship. Here is to the next 10!
@teortaxesTex feel like the class of models prior to 4.7 had some real issues around vision from an adversarial input perspective as well. I noticed this as a side effect of my security research on VLMs. I wrote it up here and open-source the methodology as well: https://t.co/98paUmimoM
@benhylak The Red blog is all the confirmation you need on the power of that model. It is found in days no collective group of individuals could. If there was a human collective smart enough to find and exploit those vulns...they would have. Mythos cooked bro
@levie Right. Scaffolding is just a temporary bridge between model releases at this point. As models get smarter the scaffolding life span gets shorter.