OWASP Agentic AI Security A gem in ML Security has been released last week! The OWASP Top 10 Agentic AI Threats and Mitigations report has put together a threat-model-based framework to secure autonomous AI systems powered by GenAI agents. This has to be one of the most comprehensive threat-model-based frameworks for agentic AI, covering key emerging security risks. It is interesting to see also a taxonomy of these security risks across memory poisoning, tool misuse, privilege escalation, cascading hallucinations, and goal manipulation. We had the honour to contribute to this framework as one of the reviewers, and certainly are looking forward to contributing across the many upcoming iterations! |
|
|
|---|
|
Sakana AI Making CUDA Slower Sakana AI launched a GenAI agentic framework that promised optimizations on CUDA kernels up to 10-100x+, however after the community got their hands on the benchmark code mistakes were pointed out showing that in some cases the kernels were 3x slower. Nevertheless it is still an interesting conceptual approach to code-optimization which likely we will only see increasingly prevalent - for good or for bad! This agentic framework basically converts standard PyTorch modules into highly optimized CUDA kernels using evolutionary techniques. |
|
|
|---|
|
Karpathy Deep Dive on LLMs Out of nowhere, Andrej Karpathy has launched an absolutely must-watch 3-hour deep dive into the foundations of building LLMs end-to-end - this is really a great reminder of how lucky we are to have access to top quality knowledge in the internet for free! This is a hands on and intuitive deep dive that goes across the end-to-end lifecycle, including pre-training on cleaned text data using techniques like byte pair encoding and Transformer architectures, as well as fine-tuning base models on curated conversational datasets to create robust assistants. |
|
|---|
|
AI Code Increase Tech Debt AI Code Assistants are making your company's code worse, and there is now clear data to back this up - this report has analyzed 211 million lines of code across 5 years to find the impact of AI coding: 1) 63% of developers using AI tools, code is being generated faster, yet this leads to more "copy/paste" and duplicated code rather than proper refactoring.
2) The metrics show a clear significant increase in defects as duplicated code blocks become prevalent. In 2024, copy/pasted lines surpassed refactored lines, which correlates with higher bug propagation. 3) Over-reliance on AI-generated code is also impacting modularity and code reuse, and this trend suggests that long-term stability might be compromised if human-led refactoring isn’t prioritized. 4) The Google’s DORA benchmarks are also showing a measurable decrease in delivery stability as AI adoption rises. This is something that we already were aware about, and bad code has existed for decades even before AI came in, however as we now begin to find clear trends it will be important that organisations as well as even code-assistant providers are able to ensure we focus these tools where they can drive the most value, as opposed to worsening conditions through bad practice. |
|
|
|---|
|
Bluesky Lossy Timeline / Ranking It is interesting not only to see Bluesky’s meteoric rise, but also their ability to serve to the increasing massive scale in innovative ways that leverage clear learnings from existing players - in this case it is interesting to see their appoach with "lossy timelines": Lossy Timelines basically was their approach to scaling to the growing number of users, which basically accepts controlled imperfection by probabilistically dropping some timeline updates for users with extremely high follow counts to prevent system overload (hot shards) and reduce latency. They introduce a loss factor which allows them to calibrate this accordingly to ensure efficient ranking algorithms, which shows really interesting improvements in performance in latency and fanout times. These type of innovations are what really showcase the value of elegant solutions to already tested AI systems (i.e. search, ranking, etc). |
|
|
|---|
|
Upcoming MLOps Events The MLOps ecosystem continues to grow at break-neck speeds, making it ever harder for us as practitioners to stay up to date with relevant developments. A fantsatic way to keep on-top of relevant resources is through the great community and events that the MLOps and Production ML ecosystem offers. This is the reason why we have started curating a list of upcoming events in the space, which are outlined below. Upcoming conferences where we're speaking:
Other upcoming MLOps conferences in 2025:
In case you missed our talks:
|
|
|---|
| | |
Check out the fast-growing ecosystem of production ML tools & frameworks at the github repository which has reached over 10,000 ⭐ github stars. We are currently looking for more libraries to add - if you know of any that are not listed, please let us know or feel free to add a PR. Four featured libraries in the GPU acceleration space are outlined below. - Kompute - Blazing fast, lightweight and mobile phone-enabled GPU compute framework optimized for advanced data processing usecases.
- CuPy - An implementation of NumPy-compatible multi-dimensional array on CUDA. CuPy consists of the core multi-dimensional array class, cupy.ndarray, and many functions on it.
- Jax - Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
- CuDF - Built based on the Apache Arrow columnar memory format, cuDF is a GPU DataFrame library for loading, joining, aggregating, filtering, and otherwise manipulating data.
If you know of any open source and open community events that are not listed do give us a heads up so we can add them! |
|
|---|
| | |
As AI systems become more prevalent in society, we face bigger and tougher societal challenges. We have seen a large number of resources that aim to takle these challenges in the form of AI Guidelines, Principles, Ethics Frameworks, etc, however there are so many resources it is hard to navigate. Because of this we started an Open Source initiative that aims to map the ecosystem to make it simpler to navigate. You can find multiple principles in the repo - some examples include the following: - MLSecOps Top 10 Vulnerabilities - This is an initiative that aims to further the field of machine learning security by identifying the top 10 most common vulnerabiliites in the machine learning lifecycle as well as best practices.
- AI & Machine Learning 8 principles for Responsible ML - The Institute for Ethical AI & Machine Learning has put together 8 principles for responsible machine learning that are to be adopted by individuals and delivery teams designing, building and operating machine learning systems.
- An Evaluation of Guidelines - The Ethics of Ethics; A research paper that analyses multiple Ethics principles.
- ACM's Code of Ethics and Professional Conduct - This is the code of ethics that has been put together in 1992 by the Association for Computer Machinery and updated in 2018.
If you know of any guidelines that are not in the "Awesome AI Guidelines" list, please do give us a heads up or feel free to add a pull request!
|
|
|---|
| | |
| | | | The Institute for Ethical AI & Machine Learning is a European research centre that carries out world-class research into responsible machine learning. | | | | |
|
|
|---|
|
|
This email was sent to You received this email because you are registered with The Institute for Ethical AI & Machine Learning's newsletter "The Machine Learning Engineer"
|
| | | | |
|
|
|---|
|
© 2023 The Institute for Ethical AI & Machine Learning |
|
|---|
|
|
|
The Institute for Ethical AI & Machine Learning
