These results are an example of generalization. Generalization occurs in benign ways in the training of all AI models: training a model to solve math problems turns out to make it better at, say, planning vacations and a whole range of other useful tasks. But as we show here, it can happen for more concerning behaviors, too: when we accidentally reward the model for one kind of “bad thing” (cheating), this makes it more likely to do other “bad things” (deceiving, aligning itself with malicious actors, planning to exfiltrate its own weights, and more). As in previous work studying emergent misa
From shortcuts to sabotage: natural emergent misalignment from reward hacking…This setup lets coders build custom behaviors, keeping customers happy with direct answers. Speaking of agents, you might have heard that hackers recently broke into Instagram accounts by simply asking the Meta…
…PyPI package with 1.1M monthly downloads hacked to push infostealer New npm supply-chain attack self-spreads to steal auth tokens Backdoored PyTorch Lightning package drops credential stealer Hackers compromise Axios…
…Agama 21 also improves the network management options from its web-based UI, supports installing now to an existing Logical Volume Manager (LVM) setup, and adds support for systemd-boot as an…
…From there, the initial setup is fairly simple. You're asked a few questions about where to store your data and whether you want to download firmware for a PlayStation 3 emulator…
I often run `docker run hello-world` after setup to do the basic check. But I am getting bored. Here's what I hack on and do instead: docker run --rm -it warachet/hello-world You get Matrix digital rain, lmao. Benefits a…
Embed real Chromium (CEF) in SwiftUI apps on macOS with a single package. No WKWebView hacks. No manual CEF setup. Just Swift Package Manager.
Hello people, I need some help with my current unraid server, it's driving me mad. What I have: A custom domain registered with cloudflare Nginx-proxy-manager Tailscale installed What I'd like: To setup something like `n…
I’ve been dealing with a massive headache for a long time: video on my second monitor would constantly stutter or lose smoothness when I was gaming. It’s that classic Windows bug that hits when you have a large gap in re…
Received my new mini PC yesterday to move my World of Warcraft server over from my Raspberry Pi5. Everything setup nice and clean, working locally but for whatever reason my EE smart hub would not forward the ports no ma…
…Microsoft Exchange? Hackers are actively exploiting a new zero-day flaw I’ve spent almost a year testing the Corsair Platform 4 — and it’s my gaming desk setup endgame 9 laptop…
…Sign in to your XDA account Porting modern software to old hardware that shouldn't be able to run it is a fun way for hackers to challenge themselves. You may have…
…All it takes is a few unique hacks. The method involves tricking an iPhone into thinking that a payment terminal is actually a mass transit terminal that’s using Apple’s ‘Express…
…If your Wi-Fi network isn't properly secured, you could be leaving your Wi-Fi wide open to hackers, data thieves and other digital threats. When your home network is breached…
…and Google Drive together and it's been a game-changer The ultimate workflow hack The setup is only two steps You set it once and forget about it You need the…
…On Tuesday, Valorant players began to report that an update to Riot's Vanguard anti-cheat software was preventing DMA firmwares — specially made pieces of hacking hardware that connect to your PC…
