saladtoes

6 ms·

1.
The Security Company of the Future(lakera.ai)
2 points by saladtoes 1y ago|0 comments
2.
by saladtoes 1y ago
Agreed on CaMeL as a promising direction forward. Guardrails may not get 100% of the way but are key for defense in depth, even approached like CaMeL currently fall short for text to text attacks, or more e2e agentic systems.
3.
by saladtoes 1y ago
https://www.lakera.ai/blog/claude-4-sonnet-a-new-standard-fo... These LLMs still fall short on a bunch of pretty simple tasks. Attackers can get Claude 4 to deny legitimate requests easily by manipulating third party d
4.
Gandalf the Red: Adaptive Security for LLMs(arxiv.org)
1 points by saladtoes 2y ago|0 comments
5.
by saladtoes 3y ago
I've been playing Gandalf in the last few days, it does a great job at giving an intuition for some of the subtleties of prompt engineering: https://gandalf.lakera.ai Thanks for putting this together!
6.
Safety is All you Need(sites.google.com)
1 points by saladtoes 4y ago|0 comments
7.
by saladtoes 4y ago
Is that really a universal fact? In any case, my statement goes in the opposite direction: is a reliable system necessarily "well understood", in the sense that it can explain its decisions? Most complex systems powering our lives
8.
by saladtoes 4y ago
Explainability is not a given in many more traditional complex systems. Decisions are often an aggregation of a large number of signals, and one can often not conceive of a single intuitive explanation for the system's decisions. A lot
9.
by saladtoes 4y ago
Interesting, thanks for sharing! Somehow I'm not surprised. My experience building systems for real world applications is that choosing a simple CNN is usually the way to go, and it's all in the data. Choosing more complex model c
10.
Detecting Data Bugs(lakera.ai)
1 points by saladtoes 5y ago|0 comments
11.
Detecting Data Bugs(lakera.ai)
4 points by saladtoes 5y ago|0 comments
12.
by saladtoes 5y ago
“The latter being when the training or test data follows a different distribution to the in-operation data” This form of ML bug is the most challenging to catch. The true in-operation distribution is often unknown which makes testing for su
13.
Free of bias? We need to change how we build ML systems(lakera.ai)
4 points by saladtoes 5y ago|0 comments
14.
Algorithms Are the New Drugs(lakera.ai)
3 points by saladtoes 5y ago|0 comments
15.
by saladtoes 5y ago
Spot on, the EU seems to agree: https://techcrunch-com.cdn.ampproject.org/c/s/techcrunch.com...
16.
by saladtoes 5y ago
Oh no!! I really wanted to know how they develop their AI so reliably :). This is a massive issue today. Hope we develop tools and processes to get us there soon.”