Check out the release notes via the Zeta Alpha blog.

www.zeta-alpha.com/post/zeta-al...

Enjoy Discovery!

Our new release allows your teams to not only create and schedule, but also to share powerful enterprise-ready AI Agents that connect securely to your internal knowledge, public research, and the web. Zeta Alpha empowers your teams to achieve more together, and faster.

The great thing with AI agents is that they can do work in the background while you are busy doing something else, which is why we have built controls so that you are able to schedule them and they send you a notification when they are done.

Everyone is experimenting with AI Agents, but few are able to show real success yet. One of the main reasons is that AI Agents need a secure connection to your company’s internal and external knowledge sources, and that’s exactly where Zeta Alpha comes in.

We'll cover how AI agents are accelerating research today, how close LLMs are to exploring new scientific ideas, and how they can support researchers’ knowledge work. As always, we'll round it out with the latest news in AI R&D, open-source releases, and the most trending papers of the month!

Good advice. Don’t select AI vendors based on capital raised, news headlines or Gigawatts of compute, but on ability to partner and to deliver.

The most interesting takeaway from the keynote? 84% of IT leaders don’t track accuracy. How can you govern and improve something that you don’t measure.

Also cool: Context Engineering, the fastest hype cycle rise from Andrej Karpathy tweet to Gartner CIO strategy advisory in less than 5 months. (With due credit for Tobi Lutke)

And as always, expect the freshest AI R&D news, the best open-source releases, and the most trending research papers of the month. Don’t miss it!
👉 lu.ma/trends-in-ai...

We will dive into:
🔒 Secure AI infrastructure & data sovereignty
🧱 Guardrails & moderation for LLM outputs
💥 Jailbreaks & prompt injections in the wild
📜 Regulation like SB 53 and ISO 42001 compliance
🧠 Information integrity & the “AI workslop” productivity killer

Check out the full EmbeddingGemma paper on Zeta Alpha, where you can easily explore more related work using the "Find similar" functionality. Until next time, enjoy discovery!

search.zeta-alpha.com/documents/4b...

Performance
✦ SOTA on MTEB multilingual/English/code
✦ Minimal int8/int4 quantization degradation
✦ Graceful drop from 768d to 128d, still leading its class
✦ Cross-lingual: beats many larger open & API models

Tricks
✦ "Spread-out" regularizer for quantization robustness
✦ Model souping across diverse data mixtures
✦ Quantization-aware training & MRL during finetuning

Training / Architecture
✦ Turn Gemma 3 into an encoder-decoder LLM using the UL2 objective (like GemmaT5)
✦ Keep the encoder, pre-finetune it on a large unsupervised dataset, then finetune on retrieval datasets with hard negatives
✦ Embedding distillation from Gemini Embedding

What is EmbeddingGemma?
✦ 300M parameter text embedding model
✦ derived from Gemma 3
✦ SOTA on multilingual, English & code retrieval tasks
✦ matching performance of models 2x its size
✦ supports dimensionality reduction
✦ comes with quantization-aware checkpoints