Time-Blind Large Visual Models Machine learning models for video tasks have limited performance, and this new paper (+ benchmark) show clearly the issue lies in time-blindness: Temporal cues power everything from life-saving medical monitors to self-driving cars—yet today’s visual (eg video) language models can’t read them. This research team has released a new benchmark and dataset to test specifically for this temporal-blindness in state-of-the-art video-language models. Namely when all semantics live in frame-to-frame motion and individual frames look like noise, which results in 0% accuracy from the leading research labs (ie. GPT-4o, Gemini, etc). Without salient spatial features, the models have nothing coherent to integrate, yielding a hard failure rather than a graceful degradation - it is quite interesting to see that these type of open initiatives help set the foundation for step change improvements in these fields. |
|
|
|---|
|
Unified Metadata Model at Netflix Netflix has released their unified data architecture to standardise their data models across their organisation; given how data is the foundation for ML, reducing duplication, improving data quality and enabling discoverability is key: Netflix shared their eng blog how they used a self-describing metamodel to capture business entities once as graph-based “domain models,” then automatically transpile those models into consistent schemas (GraphQL, Avro, SQL, Iceberg, Java), which can then provisions the corresponding data containers and pipelines, and record bidirectional mappings that tie every conceptual attribute to its physical location. They already leverage this in production through a knowledge-graph control plane , which has significantly reduced model drift, duplicate effort, and brittle joins. These type of innovations may not sound big, however these give ML teams a single, queryable source of truth that keeps training data, features, and serving paths semantically aligned across the e2e model lifecycle. |
|
|
|---|
|
Mistral Reasoning Model Magistral Reasoning models are the next big unlock for high-stakes production AI - Mistral enters the race with their first reasoning model "Magistral": It is quite interesting to see the acceleration on the battle to the top across the various different foundation model domains - ie coding, reasoning, speed, image, video, etc. Although Mistral may seem is entering somewhat late to the party, this is only relative to the insane speed of the field. Overall it seems there's an open-weights 24b param version, and a closed source larger version. On benchmarks it seems like it brings competitive performance however we are seeing stagnation on order-of-magnitude improvements, and assume the major jumps in the near future will be due to innovations beyond raw-compute / params. |
|
|---|
|
Unreasonable Effectiveness of For-Loop in GenAI The unreasonable effectiveness of a for-loop in agentic systems; or "how to build an agentic system in under 10 lines of Python": Although an oversimplification, this provides quite a good intuition on a super simplified version of how many libraries approach agentic tool use; the code is so small that we can include it below: def run_agent(llm):
msg = user_input()
while True:
output, tool_calls = llm(msg)
print("Agent: ", output)
if tool_calls:
msg = [ handle_tool_call(tc) for tc in tool_calls ]
else:
msg = user_input() In most agentic systems, a loop coordinates a model to alternate between conversation and executing returned tool calls (bash, patch, web, etc.). Although each run is slower and increases the cost additionally (in tokens), containerizing these loops is able to "sandbox" side effects and enables parallel execution. This is an emerging field so it is always interesting to see how paradigms, design patterns and taxonomies are being developed in real time - check out the article (+ other blog posts) for a deeper dive. |
|
|
|---|
|
Human Quality Text-to-Speech High-quality voice-cloning and synthetic text-to-speech is going to impact a significant society as well as vast amount of industries + these OSS models are becoming indistinguishable vs humans: Resemble AI has released an MIT-licensed text-to-speech deep learning model which can perform voice-cloning and synthetic voice generation with sub-200 ms real-time inference, zero-shot voice replication from ~5 s of audio, and a dial-able emotion parameter, all trained on 500 k h of curated speech. There are also interesting innovations with watermarking which is implemented via through their system which enables for provenance of generated audio. It is mind blowing that these resources are openly available as OSS, as these will actively disrupt use-cases related to assistants, games, dubbing, or localization pipelines (+ many others). |
|
|
|---|
|
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
