logoalt Hacker News

bpavuk • yesterday at 9:48 PM • 1 reply • view on HN

that they control and not you, re: tightening cyber safeguards.

earlier I commented on "Gemini 1.5 through 2.5 being decent RE sidekicks for free before Google's rug pull." now I decided to throw a buck for API tokens, and I discovered that if I want to use a Gemini model even for semantic decompilation (aka dump control flow graph into a LLM, together with Ghidra output, and let it figure out a cleaner C/C++ pseudocode), I'm hit with "prohibited content" errors on Gemini 4 models and 3.8 Flash.


Replies

CamperBob2 • today at 5:51 AM

That's a problem, all right. I run those sorts of tasks on local models, which are (so far) relatively easy to uncensor.