IT'S HERE! This is Imp: declarative self-Improving language model programming in Elixir. It's a port of DSPy to the BEAM ecosystem. That means you get signatures, optimizers, agent loops, retrieval and more, all with the reliability and concurrency of the actor model and the elegance of Elixir. This started as a test of the models' ability to port things across languages, but it's become much more advanced, with Optimize-Anything support and full-fledged MCP and ACP adapters as well. Basically a whole kit of building blocks for anything you want to make. If you're interested in the DSPy paradigm, Elixir, text optimization, or weird code ideas in general, give it a look and tell me what you think! Imp is experimental, it's 0.5.0, it probably needs some expensive benchmarking to prove its optimizers really work. I'd love to get the kinks worked out before doing that haha. Imp is open source, MIT licensed, and hopefully easy to read. Given enough agents, all bugs are shallow, right? Now, without further ado: Imp! github.com/deepfates/imp

DSPy 3.4.0 was just released! This release includes native support for Jev and System one models inside of DSPy! Use it with compatible signatures. This release also includes a brand new optimizer, ReAnchor, specifically for calibrating outputs with confidence.

SmallReason-ColBERT: An Ultra-Small Late-Interaction Retriever for Reasoning Intensive Retrieval @perdactor et al. present a 32M retriever for reasoning-heavy queries, adding a learned token-weighting head. 📝arxiv.org/abs/2609.29652 👨🏽‍💻github.com/DataScienceUIBK/S…

OBLIQ-IR: Training a Dense Retriever for Oblique Queries @perdactor et al. train a dense retriever for oblique queries, where relevance depends on latent traits like stance or style rather than topic overlap. 📝 arxiv.org/abs/2609.29649 👨🏽‍💻 github.com/DataScienceUIBK/o…

DSPy methodology 🤝 System One Program, don't prompt!

We’re considering hosting a short, casual @DSPyOSS webinar next week to run through Jev support and design patterns. Let me know if you’re interested, and what you’d like to see covered.

And why I'm talking about why Embeddings Don't Solve RAG on Monday

The embedding for this perfectly answers a very narrow question: "Paris is a city on the Seine" Compare to the longer passage about tourism... "Paris is a city on the Seine, it's tourist attractions are..." This probably would match Paris tourism queries, but the first sentence suddenly stops being available for "What river is Paris on?" query. It's document embedding is now muddled more away from being able to answer the question.

Kudos to @LightOnIO for being ChatGPT's preferred late interaction model ;)

I have just realesed lm15 1.0 on pypi. I hope it will serve you as well as it is serving me. lm15 is going straight to the network layer, in python, it has 0 dependencies. As such, it is smaller and faster then any other sdks. If you want a, liteweight, provider agnostic way to call llms in a universal and typed way. You should give it a try. It was developped specifically as a delighful lower level abstraction for other AI framework to build on top. It open sourced, mit licensed, already depended on by other AI framework (dspy). lm15.dev/llms.txt pip install lm15 npm install @lm15/lm15 cargo add lm15 go get github.com/lm15-dev/lm15-go LM15 already supports over 10 providers, including typesafe. It supports continuation, live, image gen, image inputs, audio, citation, tools, everything.

DSPy 3.4.0 out, with: (1) native support for Jev and System One models - in the timeless DSPy syntax. (2) a brand new optimizer, ReAnchor, specifically for calibrating outputs with confidence. (3) lightning-fast import speeds for the LLM abstraction, via sibling library LM15)

this is an honor! 🥹

Introducing ReAnchor, a DSPy optimizer built for Jev and other System One models. cmpnd.ai/blog/building-jev-p…

whoa! so clean. the more the world changes, the more dspy stays the same at the interface level and very very different below the surface

Just released Jev support for DSPy, with a new optimizer that calibrate’s Je’s output thresholds against your metric. Works with existing DSPy signatures and metrics, can switch between LMs and Jev with one line.