bednars.me My personal blog https://bednars.me How your follower list exposes your politics without a single post How much can social media reveal without reading someone's posts? An old Twitter project exploring political inference from social connections, profile descriptions, and public metadata. https://bednars.me/blog/how-your-follower-list-exposes-your-politics-without-a-single-post https://bednars.me/blog/how-your-follower-list-exposes-your-politics-without-a-single-post Tue, 25 Aug 2026 00:00:00 GMT Don't trust Hermes Agent with anything that matters I tested Hermes Agent through Nous Portal. The interface was unreliable, the configuration was difficult to reason about, and private context leaked into a group conversation. https://bednars.me/blog/dont-trust-hermes-agent-with-anything-that-matters https://bednars.me/blog/dont-trust-hermes-agent-with-anything-that-matters Sun, 12 Jul 2026 00:00:00 GMT Test Windows drivers in seconds, not minutes Use Windows PE and QEMU as a disposable harness for KMDF testing, fuzzing, and kernel debugging without maintaining a full Windows virtual machine. https://bednars.me/blog/test-windows-drivers-in-seconds-not-minutes https://bednars.me/blog/test-windows-drivers-in-seconds-not-minutes Sun, 28 Jun 2026 00:00:00 GMT Your LLM can change its answers while it's still running A writable GGUF file can alter a running llama.cpp model through memory mapping, without restarting the server or injecting code into its process. https://bednars.me/blog/your-llm-can-change-its-answers-while-its-still-running https://bednars.me/blog/your-llm-can-change-its-answers-while-its-still-running Tue, 10 Mar 2026 00:00:00 GMT If your model has a public API, it can be stolen A public prediction endpoint can become a labeling service, an unofficial inference backend, or a training source for a replacement model. https://bednars.me/blog/if-your-model-has-a-public-api-it-can-be-stolen https://bednars.me/blog/if-your-model-has-a-public-api-it-can-be-stolen Fri, 14 Nov 2025 00:00:00 GMT Does hiding malware in a model actually work? What a neural packer proves, what the prototype deliberately leaves out, and where the idea becomes relevant to real malware analysis. https://bednars.me/blog/people-asked-if-hiding-malware-in-a-model-actually-works https://bednars.me/blog/people-asked-if-hiding-malware-in-a-model-actually-works Tue, 01 Jul 2025 00:00:00 GMT You can hide malware inside a neural network A neural network can memorize a complete source file and reconstruct it through inference, turning deliberate overfitting into an unconventional packing technique. https://bednars.me/blog/you-can-hide-malware-inside-a-neural-network https://bednars.me/blog/you-can-hide-malware-inside-a-neural-network Sun, 29 Jun 2025 00:00:00 GMT