Why do you think that the companies training the models will let that happen as opposed to mitigating the problem when they detect it? Many of them have been deliberately using carefully curated synthetic data for a few generations of models now
@simonw
-
iPhone autocorrect changes prompting to promoting in AI workflows
By
–
Yes, when I'm typing on the iPhone it turns prompting into promoting all the time and sometimes I miss it before I hit post
-
Training Models: Balancing Demonstrations with Real-World Performance
By
–
See my original comment Sure, it's easy to deliberately demonstrate on small datasets, but for it to affect real-world models the people training them would have to ignore the problem entirely What makes you think they're just going to let their models get worse?
-
AI Security: Mitigating Prompt Injection and System Vulnerabilities
By
–
If there's no known fix for those then they are indeed similar to prompt injection What's the recommended mitigation for people with security concerns that are serious enough for this to be a concern? Air-gapped machines? Back to pen and paper messages sent using one-time pads?
-
Hardware Bug Raises Questions About AI System Reliability
By
–
Surely that's a bug in the hardware at they point?
-
Anthropic Launches New Memory Feature Similar to OpenAI
By
–
And an update, since it turns out Anthropic announced a new memory feature yesterday that's more similar to how OpenAI's works https://
anthropic.com/news/memory -
Claude vs ChatGPT: Memory Implementation Differences Explained
By
–
Posted some of my own notes on Shlok Khemani's (excellent and comprehensive) notes on how Claude and ChatGPT's memory implementations differ from each other
-
GitHub Domain Allow-listing Security Risk Analysis
By
–
I don't think there are any endpoints on http://
GitHub.com itself that could expose logs of incoming GET requests to anyone outside of GitHub/microsoft employees It does make me nervous that all of that domain is allow-listed though, feels like the riskiest entry in there -
FFmpeg sandbox integration security vulnerability analysis
By
–
I don't see why it would add any new vulnerabilities, the main safety feature of the sandbox is that it can't make network calls and ffmpeg wouldn't change that
-
Model weights now available on Hugging Face 151GB
By
–
Looks like the model weights are up on Hugging Face, about 151GB