THE INSTITUTE FOR ETHICAL AIALIGNMENT + SAFETY
JOIN / CONTACT

THE ML ENGINEER — WEEKLY NEWSLETTER

The MachineLearning EngineerIssue #344

Join 70k+ AI professionals receiving weekly curatedarticles, tutorials and blog posts onproduction machine learning.

#343
Issue #34420/07/25ai-agentsml-education
#345
Issue #344 đŸ€– - The State of GenAI in 2025, Neural Net Operating System, Chain of Thought Monitoring, Crawling a Billion Pages, Principal Engineering at Amazon + more 🚀

Thank you for being part of over 70,000+ ML professionals and enthusiasts who receive weekly articles & tutorials on Machine Learning & MLOps đŸ€– You can join the newsletter for free at https://ethical.institute/mle.html ⭐

If you like the content please support the newsletter by sharing with your friends via ✉ Email, 🐩 Twitter, đŸ’Œ Linkedin and 📕 Facebook!

This week in Machine Learning:

If you’re looking for an interesting career opportunity, I’m hiring for a few roles including Data Science Manager (Forecasting), as well as Data Scientist (Forecasting) - check them out and please do share with your network!

The State of GenAI in 2025

Our GenAI keynote at the 2025 WADev World Congress is live! We dive into current limitations on GenAI productivity, and best practices on building for GenAI. One of the key take-aways on the current tooling for dev producitivity is that most sit at the top of the development funnel (ie writing code), whereas the lower parts of the funnel (QA, operating, monitoring, etc) are still much less mature. In order to ensure we invest in mature GenAI tooling, this requires investment across the entire development funnel, and going beyond purely chat‑style code copilots. Achieving this means building agentic stacks that fuse orchestration frameworks, context gateways, guardrails and GPU‑aware schedulers while baking in security defences against prompt injection, data poisoning and other novel threats.

Neural Net Operating System

I never would have imagined an operating system-like experience running (literally) in a neural network; this is quite a creative approach to training a network on user inputs and visual OS GUI outputs: NeuralOS basically brings together an RNN that keeps internal application state with a diffusion UNet that renders the next (OS GUI) frame. The model has been trained on roughly 40 k thirty‑second Ubuntu‑XFCE sessions captured at 15 fps and compressed so the full pipeline fits on an eight‑GPU H100/H200 node. The public demo shows how it reproduces window launches, cursor motion, and coarse keyboard states - despite the huge lag (due to inference time) it’s pretty impressive, and the text hallucination is quite amusing. This particular use-case seems more of a toy implementation, however the concept of GenUIs is definitely growing (last time we saw this simulating DOOM) - certainly an area to keep an eye particularly as these type of simulations evolve.

Chain of Thought Monitoring

The UK AI safety institute released an interesting paper proposing monitoring of chain-of-thoughts where LLMs are used in particularly critical or sensitive topics, as this can allow for stronger guardrails compared to only looking at input/output: This is an interesting paper that shows that when language models emit natural‑language chains of thought, those tokens can be routed through lightweight monitors to flag malicious or undesired plans, such as reward‑hacking shortcuts or policy violations. This can be triggered before any external action occurs which provides an extra safety layer, however the method also has weaknesses, as methods that internalise these chains of thought would remove this capability (e.g. reinforcement fine‑tuning, direct CoT supervision, or future architectures) There are some practical takeaways, such as ensuring teams log the chains-of-thought by default, consider tracking a dedicated metric during training, stress‑test for deliberate obfuscation, and treat CoT monitoring as just one layer in a broader set of tools for observability.

Crawling a Billion Pages

Web‑scale crawling is the raw fuel that powers today’s data‑hungry ML models, and this is a great deep dive in the current state of large-scale (billion+) web crawling. As of today it’s impressive how much can be achieved with relatively low hardware; in this case this is a practical experiment on 12 i7i.4xlarge nodes crawlling over 1-billion web pages in a single day. The setup consists of nine asyncio fetchers and six parsers with redis for caching, which can crawl over 1B HTML webpages in about 25.5 hours for just $462 by sharding the top‑million domains and writing straight to local NVMe instead of S3. The new bottle-neck is no longer the pipes, but the parsing itself; swapping libraries like lxml foroptimized C++ libraries lifts per‑process throughput from 55 page/s to roughly 160 page/s. It is interesting to see the new challenges that emerge as crawling at such a scale becomes so accessible; particularly in the era of AI, many organisations are still figuring out the best balance to restrict unexpected DDOS whilst enabling access for the numerous AI services popping up.

Principal Engineering at Amazon

The principal engineering community at Amazon is renown for having a high bar on technical knowledge and outcome-driven thinking - this is a great interview that provides a deep dive into the world of principal engineers at Amazon: One interesting highlight is the emphasis on the challenges on jumping from Senior (L6) to Principal(L7) is widely seen as Amazon’s toughest promotion, apparently as scope and compensation-wise is presented as a two-level jump; it seems often requires multiple attempts and demands showcasing significant tangible impact (ie already working at that level). Once engineers reach principal, the scope for ownership increases to department-wide, ensuring key focus on scaling impact through technical contributions that have quantifiable outcomes. It is interesting to see how organisations shape these communities internally, as well as the similarities and differences between these - it is particularly interesting now to see how organisations are wrestling to enable what is the intersection of their principal engineering and science IC communities to support the scale of machine learning needs in the era of GenAI+.

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:

Open Source MLOps Tools

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!

OSS: Policy & Guidelines

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!

About us

The Institute for Ethical AI & Machine Learning is a European research centre that carries out world-class research into responsible machine learning.

Check out our website

✉ Email, 🐩 Twitter, đŸ’Œ Linkedin