The applications of AI are numerous,
yet they suffer from vulnerabilities.
The importance of this is even greater in safety-critical systems,
where AI-based decisions can have a significant effect on users or operators.
We solve these problems.
keepn.ai maximizes the potential of your product by minimizing risk related to AI safety.
The solution
Using our first generation solution, stakeholders, product, and R&D can define static and dynamic policies to govern the productโs behavior.
Come on board as an early adopter,
Using our solution, Product Owners and Domain Experts gain complete control over business logic, regardless of software updates and configuration changes, assuring Stakeholders that compliance requirements will be met.
Partner with us.
Complete the form for all details.
Already a LinkedIn follower? Check out our content.
Yesterday, I visited the AWS floor 28 in Azrieli Sarona, Tel Aviv. There was a really fun event called How To Secure Your Generative AI Application.
This workshop was a great opportunity to learn more about AWS GenAI compliance philosophy and how it aligns with the EUs AI act, watch a panel disc...
Can copyright be protected by diffusion models (DMs)? The CVPR 2024 paper suggests watermarks.
It is impressive how well DMs generate images.
However, DMs replicating unauthorized creations may violate copyright.
To prevent DMs from replicating unauthorized content, the framework embeds person...
Machine-learning-based Windows PE malware detectors can significantly be improved with explainability techniques.
In malware, it has become increasingly apparent that crafting optimal adversarial perturbations is critical for maximizing the safety of AI systems.
This study aims to improve adver...
Did you know? On-device deep learning models are being used in finance, social media, and driving assistance.
--
On-device models on Android smartphones were proven to be extremely vulnerable because of the transparency of their platform.
Although iOS apps are as popular as Android, there are ...
An informative post we post on behalf of. Hope you find it useful ๐โก๏ธ๐
#ml #safety #mlsafety #machinelearning The `SB 1047` is a new bill in the California Legislature. It generally states that if an ML model might have hazardous capabilities then engineers must take precautions.
Credit: twitte...
Did you know? Compression using neural networks can reduce storage space by a factor of 1000.
As an example, a 2Gb hash-table can be compressed into a 2Mb neural network.
You can learn more by reading the attached article which shows how a neural network was used to compress a Markov decision p...
The Lasso (L1), Ridge (L2), and Elastic Net (L1+L2) techniques are regularization methods (and feature selection methods) in Machine Learning.
Obviously, each technique has pros and cons. The illustration might help you figure it out.
In machine learning, regularization prevents overfitting and...
JailbreakBench is an open robustness benchmark for jailbreaking language models.
Language models can generate harmful, unethical, or otherwise unwanted content as a result of jailbreak attacks.
The JailbreakBench provides a repository, an evaluation framework, a leaderboard and a dataset for ja...
Anyone who has studied Machine Learning knows the โNo-Free-Lunch theoremโ and why prior knowledge is essential. This paragraph from Understanding Machine Learning: From Theory to Algorithms by Shai Shalev-Shwartz and Shai Ben-David is a great reminder:
It turns out that the incorporation of prio...
Need open-source LLM guardrails? Check out this post by Dor, our founder, from 4 months ago.
Want to stay up to date? Follow keepn.ai
#llm #guardrails #safety #security #mlops Securing your LLM system? Here are 3 open-source tools to get you started.
1. (NVIDIA) ๐๐๐๐จ-๐๐ฎ๐๐ซ๐๐ซ๐๐ข๐ฅ๐ฌ: ๐ต๐ฐ๐ฐ๐ญ๐ฌ๐ช๐ต ๐ง๐ฐ๐ณ ๐ข...
Some more bits from our founder, we hope you find it useful.
Keep it up, and follow keepn.ai ๐ Quantization of neural networks (NN) is a technique to reduce the NN required computational resources.
One simple yet effective method to produce a quantized NN out of existing one, is to decrease th...
When Sam Altman and Ilya Sutskever were in Tel Aviv last year, Sam explained why ChatGPT was useful to him - Wikipedia is his favorite website.
#chatgpt #llm #attentionnetworks --> Dear ChatGPT, describe the Invisible Hand concept by Adam Smith in a simplified way
The concept of the invisible ...