DSPy Official Hot Take #2: Model providers need to standardize a "tell me about yourself" API. We'd like to know a model's context length (input & output), any special structure (e.g., <think>), special APIs (e.g., structured outputs), and perhaps latency, rate limits, and cost.

It's honestly mindblowing. transformer-circuits.pub/2025/attributi…

DSPy is the highest-bandwidth language to talk to computers in a just-precise-enough way. First 36 lines below: 1. Take arbitrarily long content: `chunks`. 2. Build a global Table of Contents. 3. Distribute chunks into sections to be written. 4. Recursively write each section.

The paper investigates how LLMs perform in realistic, multi-turn conversational settings where user instructions are often underspecified and clarified over several turns. I keep telling devs to spend time preparing those initial instructions. Prompt engineering is important.

I love the simplicity of "just putting it into context" - and it may work for some scenarios. But for majority of non trivial use cases information retrieval, context synthesis and context window management steps will stay with us (in some form).

Code reuse isn't free. Nor is building everything from scratch.

Context synthesis to fork existing chat is difficult as it relies on understanding of a broader context, eg user goals or constraints which may not be explicitly stated in the data sources. It may have to be proceeded with another step / steps to discover user's intent behind

Version 1.0 of Instructor for PHP is coming with **many** breaking changes. Multiple APIs from 0.x have been remodelled. If you're using Instructor in production and we haven't spoken yet, please ping me via DM. I'd like to discuss the best way to roll it out to limit the

2025.1 versions of Jetbrains IDEs (Pycharm, PhpStorm) are working super slow on my machine. Does anybody experience similar issue?

Don't use structured output mode for reasoning tasks. We’re open sourcing Osmosis-Structure-0.6B: an extremely small model that can turn any unstructured data into any format (e.g. JSON schema). Use it with any model - download and blog below!

A 340 page huge report on AI trends - released by BOND Some wild findings from this report. 🧵1/n

InstructorPHP v1.0.0 released! 🎉 First stable release of InstructorPHP - MIT/OS library for structured data extraction and LLM integration for PHP apps. Framework agnostic, batteries included. 🔥 StructuredOutput, Inference, and Embeddings classes with fluent, cohesive APIs.

the future is more engineers using AI to ship 10x, not fewer because of AI.