Just sitting here waiting for the day AI can order and wrap presents.

Building the Future of AI in Software Development // Varun Mohan // MLOps Podcast #195

Setting up and running a company takes courage.

Pivoting that company’s focus takes something else!

But that’s what the folks at Exafunction did. They shifted from GPU virtualization technology to developing their coding assistant, Codium.

And Varun is really honest in this chat about it, telling me all about the difficulties such a change comes with. But, more importantly, the reasons why they did. And, with 400,000 users so far, it seems like it’s paying off!

We also chatted about the challenges involved in developing large-scale evaluation systems and strategies for optimizing coding-related tools. Plus he really stressed the importance of bringing constant value in software tools, making sure they have useful and practical features instead of just flashy ones.

Klinton Bicknell from Duolingo // Bill Salak from Brainly // Yeva Hyusyan from Sololearn // Moderated by Paul van der Boor from Prosus

They say your school days are the best days of your life.

If that’s the case, you’ll want to tune in to this rewind edition from LLMs in Production. This fireside chat, moderated by Paul van der Boor hosting Yeva Hyusyan, Bill Salak, and Klinton Bicknell, unveils insights into how AI is shaking up the educational space.

It’s not just about generating information, but about presenting it in a way that students can understand, creating personized content for self-paced learning.

So, sharpen your pencil, put that chewing gum in the bin and settle down at the back before clicking through to view.

You can watch it here.

This time of year for the holiday season, it’s all about going big, so this mini summit on scaling delivers on that!

Ben hosts David Espejo from Union, Fabio Grätz from Recogni, and Arno Hollosi, from Blackshark.ai as they talk through the challenges of scaling, innovative solutions, infrastructural considerations, and process management.

Arno talks about data handling efficiency, David discusses fulfilling model development contracts, particularly in computer vision-intensive organizations, and Fabio shares experiences in developing inference systems for autonomous driving.

So, go big, or go home!

And then, when you’re at home, relax and watch this.

💡Job of the week

Senior Staff Applied Scientist (Natural Language Processing, LLMs) // Katanemo Labs, Inc (Various and Remote, US)

Katanemo is on a mission to build intelligent infrastructure services for developers building AI-native apps.
You will lead cutting edge research and development of NLU (transformer and otherwise) models and analytical solutions in machine learning (ML) that can be used for multi-task detection, intent classification, named entity recognition, etc.

What you will bring:
- 5+ years of modeling experience in one or more the following areas: Natural Language Processing/Understanding (NLP/NLU).
- A Ph.D Degree in Computer Science, Machine Learning, Statistics, Applied Mathematics or an equivalent highly technical field.
- Hands on experience developing, scaling and implementing machine learning using relevant programming languages (such as Python), state of the art deep learning frameworks and big data tools.

Competitive Differentiation for Foundation Models in the LLM Space

We’ve all spent more time scrolling through Netflix than actually watching anything, paralyzed by choice. Sometime you need help to guide your choices.

With an increasing number of companies developing their own foundational models and making them available to 3rd-parties, how do you chose?

Helpfully, this blog gives three main product characteristics that can be used as vectors to help make useful comparisons: Compute Performance, Safety and Alignment and RAG.

There’s also a General Purpose category too, so depending on your model needs, this blog will help you identify which models could be best for you.

Make the right choice, and be sure to read this blog!

With thanks to Alex Sandu for the contribution.

Back from Apply(ops) 23 conference

Conferences can be a bit like festivals, there are some acts you want to see, but it means missing out on others.

Thankfully, this blog does a great job of giving an overview of the whole Apply(ops) 23 conference, so even if you missed a headliner, it’s got it covered!

There’s loads in the blog to cover, but highlights of the highlights include:

• Uber's Michelangelo Evolution
• ML in a Multi-Cloud Environment
• Databricks/Tecton Discussion on Production ML Trends:
• Panel Insights on recommendation systems
  
  

So enjoy, knowing it’s better than a festival because you don’t have to stand in a field getting rained on then go back to sleep in a tent.

With thanks to Jean-Michel Daignan for the contribution.

Austin - December 14
Utah - December 26
Bristol - January 10
Stockholm - January 16