Anthropic on LLM "Biology" Anthropic's new paper on the "Biology" of a Large Language Model has some really interesting takes on the inner workings of large models when performing complex tasks: This is a great deep dive to demystify the complex internal behaviour across multi-step reasoning, forward/backward planning and interventions. 1) The model internally executes sequential reasoning steps (e.g., deducing “Dallas” → “Texas” → “Austin”) rather than relying solely on memorized patterns. 2) In tasks like poem generation, the model pre-selects candidate rhyming words and then structures the line to fit these plans. 3) Various circuits within the models themselves contribute to the final output which can replace opaque neurons with interpretable features via a cross-layer transcoder. 4) Using intervention experiments can confirm that altering specific circuit components (e.g., swapping “Texas” for “California”) results in predictable changes in outputs, offering a practical path toward debugging. For production ML practitioners it is interesting to see some conceptual frameworks that allows reasoning and discourse as these are still emerging. |
|
|
|---|
|
Architectural Patterns in Python A brand new (+ free!) O'Reilly Book on Architecture Patterns with Python - check it out: This is a great resource for practitioners that are interested to take their knowledge to the next level through practical architectural patterns that are used growingly across real-world codebases. These include patterns like domain modeling, the repository pattern, and unit of work, as well as event-driven architectures like including message buses, command handling, and CQRS. Overall, while the examples stem from an e-commerce context, these principles are directly transferable to structuring and deploying machine learning systems that need to evolve gracefully with business needs. |
|
|
|---|
|
Tencent's New Ultra-Large Model Another Chinese "Giant Model" enters the arena with Tencent’s Hunyuan-T1, an ultra-large that beats the DeepSeek-R1 models and ChatGPT 4.5 models on a clear race to the top. This model is quite interesting as it would be the largest Mamba MoE architecture, which leverages advanced long-text capture and sequence processing to minimize context loss and reduce compute usage. Interestingly enough over 96% of its post-training is dedicated to reinforcement learning using classic RL techniques like data replay and periodic policy resetting. The model achieves competitive scores on industry benchmarks - namely 87.2 on MMLU-PRO and strong results in mathematics, coding, and logical reasoning tasks. |
|
|---|
|
Improving RecSys in LLM Age Eugene Yan has put together a great compendium of recommender systems in the age of large language models, with interesting insights on how breakthroughs in the field are leading to major leaps in this space: It is quite interesting to see that the RecSys & search space is now being tangibly improved from research that arises on the LLM space, working together with traditional ID-based and behavioral recommender systems, and integrating through hybrid architectures that utilize dense, multimodal content embeddings to address challenges like cold-start and long-tail items. It is interesting to see key areas as well on LLM-assisted data generation and analysis such as synthetic metadata creation to refined query segmentation, enhancing training data, as well as innovative training paradigms and inference best practices. |
|
|
|---|
|
OpenAI's Image Gen Models The internet has been increasingly discussing and showcasing OpenAI’s latest GPT‑4o image updates, it is interesting to see some of the details on the internals from these techniques, particularly the evolution on the architectures themselves. This includes integrating multimodal capabilities directly into language models themselves which allows for precise and context-aware rendering of both text and images, and in this case also enabling for nuanced context aware image generation. Although the models are closed source, from the description we can see that these are built on a unified autoregressive transformers paired with diffusion-based decoding which is quite interesting as this is enabling for reasoning at the image level whilst enabling for multi-modal interfaces. We can only bet that we will see in the coming weeks and months many more innovations in the space of image models from players around the world. |
|
|
|---|
|
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
