Optimizing Tool Selection for LLM Workflows with Differentiable Programming

Comments (1)

viksit · 43m ago

I was experimenting with how local, learnable routers can reduce token overhead, and lower costs, and decided to publish a post about it. The main goal is to delegate tool calls via a PyTorch based learner and examples of how to integrate this into a DSPy pipeline. Feedback welcome!

Automatically Evaluating AI Coding Assistants with Each Git Commit (tensorzero.com)

WIP Silent Hill decompilation project (github.com)

Coding with AI agents using the Breadcrumb Protocol (dasith.me)

Engineer caught juggling multiple startup jobs is a cautionary tale (cnbc.com)

OpenMW 0.49.0 released (open-source Morrowind reimplementation) (openmw.org)

How to crack FAANG coding interviews (mlwhiz.com)

Artificial Intelligence in Miniature Format for Small Devices (tugraz.at)

The Self That Never Was (hedgehogreview.com)

Techno-Feudalism and the Rise of AGI: A Future Without Economic Rights? (arxiv.org)

How to Crack ML System Design Interviews (mlwhiz.com)

WinUAE 6.0.0 Amiga Emulator (winuae.net)

Show HN: Web Metadata search. Search for headers, web apps, CMSs, and versions (dnsarchive.net)

A100 is a puzzle game inspired by the mobile game „1010 " by Gram Games (cobour.itch.io)

How Tether became money-launderers' dream currency (economist.com)

Inertial forces (indirect terms) in problems with a central body (astro.theoj.org)

Video captured during Alberta storm could be rare ball lightning event (globalnews.ca)

How to Network as an Introvert (aginfer.bearblog.dev)

Free Lunch Is over for the AI That Broke the Web (gizmodo.com)

AI helps find formula for paint to keep buildings cooler (theguardian.com)

The Memory Wall (semianalysis.com)

Ask HN: How do you sell to B2B in current state of AI?

Optimizing Tool Selection for LLM Workflows with Differentiable Programming (viksit.substack.com)

Free tool for engineering managers: Never show up unprepared to a 1-on-1 again (tndm.app)

Drunk Tourist Damages Kyoto Temple Tied to Samurai Legend (tokyoweekender.com)

Cap and Trade, but for Rent Control (nodumbideas.com)

Building accurate postal address matching systems (robinlinacre.com)

Approach to LLMs and Other Reflections (matttproud.com)

Litestar 2.16.0 (docs.litestar.dev)

The Internet as Our Nervous System: From Cloud to Local-First (2024) (blog.anytype.io)

Holding Cellphone while driving is illegal, California court rules (latimes.com)

Lefties Aren't as Creative as We Thought (nautil.us)

The New Interstellar Object 3I/Atlas Is Smaller or Rarer Than It Looks (avi-loeb.medium.com)

Saving energy with efficient code [video] (youtube.com)

The Prime Reasons to Avoid Amazon (blog.thenewoil.org)

China helped Pakistan with 'live inputs' in conflict with India (reuters.com)

What if This is it? (prickly.oxhe.art)

Ask not for whom the Louvre of Bluesky tolls, it tolls for thee (techcrunch.com)

Ask HN: Would you use AI software to assist you in an interview?

"Nothing in biology makes sense without teleology" by Michael Levin (youtube.com)

Energy-Based Transformers Are Scalable Learners and Thinkers (energy-based-transformers.github.io)

Show HN: touchgrass.fm, reclaim screen time for quality time (touchgrass.fm)

Hacker News MCP Server (github.com)

The Growing Threat: The Dark Side of AI and LLMs (blog.sofiane.cc)

The Anthropic Economic Futures Program (anthropic.com)

Examining a Copyright Claim from Copytrack (bentasker.co.uk)

Tell HN: A fake, highly obfuscated Solidity VSCode plugin found on marketplace

Cheese may give you nightmares (medicalnewstoday.com)

XGH: EXtreme Go Horse Methodology (2019) (medium.com)

Context Engineering as Code – Systematic approach to reliable AI development

India bars Jane Street from securities market, citing stock index manipulation (reuters.com)

Optimizing Tool Selection for LLM Workflows with Differentiable Programming

Comments (1)