Voting isn't open yet, but even Big JC's getting excited about our March Model Madness knockout comp.

When voting does open on 25 March, it will be a blind vote—the models are only revealed once they're defeated.

While you wait to get voting, submit your prompts for a chance to win prizes and be part of the biggest showdown since Magic Johnson Vs. Larry Bird.

Or, for those not into basketball, Hulk Hogan Vs. Andre the Giant in WrestleMania III.

Why Purpose-built Vector Databases Matter for Your Use Case // Frank Liu // Jiang Chen // Yujian Tang

With the recent talk about the next James Bond, I thought I'd throw my straw hat into the ring, so I need a suit. Off-the-rack just won't cut it; sometimes, you need something bespoke, purpose-made.

This nicely leads to this chat on the importance of purpose-built vector databases! We covered the limitations of long context models and how vector databases can improve real-world search, use cases, potential issues, and best practices. They also shared some development insights, upcoming features for Milvus, and the challenges of generating embeddings and managing vector databases.

The name'sh Demetriosh, from the MLOpsh Community. Thanksh for lishening. 🍸 (That was Sean Connery, not too many Martinis!)

Security and Privacy Panel // Diego Oppenheimer // Ads Dawson // Katharine Jarmul // David Haber

I've got the suit sorted, but to be in the Secret Service, I'm going to need to know all about Security.

Well, this Panel from our AI in Production Conference covered that. Thinking about the shift from traditional ML to LLMs, they discussed the underlying challenges, such as privacy engineering, traditional vulnerabilities, the need for new evaluation criteria, and the lack of user knowledge about AI risks.

They emphasized the importance of frameworks like the OAS Project for managing AI risks and debated AI memory vs human learning and real-world implications like content licensing and legalities.

Sorted. Now I just need to get the gadget filled car.

💡Job of the week

Senior Machine Learning Software Engineer // Grok (US, remote)

As a part of the ML Systems team, you will work closely with the sales, applications, and engineering teams to develop and optimize ML models and systems for Groq's hardware and contribute to original research.

Responsibilities:

• Develop and optimize ML models and kernels for Groq hardware across various domains using proprietary and popular frameworks.
• Analyze and compare the performance of models on Groq and competitor hardware to identify optimization opportunities.
• Influence Groq hardware features based on optimization insights from performance analyses.
• Publish research on model optimizations and hardware improvements in leading ML conferences.

Requirements:

• Foundational knowledge in neural networks and mathematics, with proficiency in at least one area among computer vision, natural language processing, recommendation engines, reinforcement learning, and linear algebra.
• Experienced with programming languages (Python, C/C++), machine learning frameworks (TensorFlow, PyTorch, Caffe), and hardware accelerator languages (CUDA, MKLDNN), including programming on FPGAs or DSPs from evaluation to production.
• Understand processor architectures and distributed systems, recognizing their impact on machine learning model performance.

Madrid - March 21
Bristol - March 26 (cheers to Berkeley Square 🙌)
Stockholm - April 23 (shout out to Weights & Biases, Stormgrid,  ...................................Crowd Collective, and AI Sweden)