Optimizing Tool Selection in LLM Workflows(Part 2): A DSPy and PyTorch Benchmark

Comments (1)

viksit · 1d ago

Based on the feedback from my last HN post on differentiable routing, I ran a follow-up benchmark: local RNN vs GPT-4o for tool selection in LLM workflows.

Same accuracy, 40% lower cost. Appreciate all the suggestions, this post builds on them.

Google ordered to pay Argentine pictured naked in garden (batimes.com.ar)

Judge Scraps Opinion After Lawyer Flags Made-Up Quotes (news.bloomberglaw.com)

Show HN: Blueboots – A retro themed Fedora OS built with one Containerfile (github.com)

Efrit: A native elisp coding agent running in Emacs (github.com)

Show HN: I built a notion ai agent (youtube.com)

Ask HN: Should HN introduce a "Tell HN" tab?

Distro-Hopping and RICEing (l-o-o-s-e-d.net)

Show HN: AI image generator with 6 artistic mentors for better prompts (createvision.ai)

Equilibrium in the Embedding Space: When Novelty Becomes Familiar (lightcapai.medium.com)

Show HN: Add viral TikTok audio to work meetings (soundboard.recall.ai)

RustMailer – Week 1 Update: 729 Views, 165 Clones, 13 Stars (in 9 Days) (indiehackers.com)

Good Docs Describe, Bad Docs Prescribe (rethinkingsoftware.substack.com)

Show HN: Crawell – Extract any page as Markdown or download images in bulk (chromewebstore.google.com)

Running Serverless WASM Functions on the Edge with K3s and SpinKube (fermyon.com)

Asciinema: Record and share your terminal sessions (asciinema.org)

Chinese drones carry 180ton of steel and concrete up mountain in pioneering feat (scmp.com)

Benchmarking LLMs on open source Vulkan (airlied.blogspot.com)

FCC approves Paramount-Skydance merger after Trump settlement, Colbert cancelled (turnto10.com)

PocketPages: No-Build Multi-Page Apps for PocketBase (pocketpages.dev)

The Economics of Superintelligence (economist.com)

A new way to build Trending filters using Elasticsearch (secalerts.co)

Handling request scoped dependencies in Golang without abusing the context (winsnes.io)

NextTurn – GitHub meets LinkedIn with XP, ranks, and Prestige for developers (nextturn.dev)

Spot-if-AI: detect if a track has been generated with tools such as Suno or Udio (chromewebstore.google.com)

Chinese Hackers Are Exploiting Flaws in Widely Used Software, Microsoft Says (nytimes.com)

Just built WeBuyBack – a resale marketplace (webuyback-marketplace-hub.lovable.app)

University of Maryland Linux Users Group Mirror (mirror.umd.edu)

US nuclear weapons agency 'among 400 organisations breached by Chinese hackers' (theguardian.com)

Kagan Says She Was Impressed by AI Bot Claude's Legal Analysis (news.bloomberglaw.com)

Microsoft says Chinese hacking groups exploited SharePoint vulnerability (cnbc.com)

Why Nvidia is going to have it's deepseek moment soon

I Made a Faster ChatGPT with Video and Image Generation (shiori.ai)

MeLE Overclock X2: pocket-sized PC with Intel N150, upgradeable RAM, storage (liliputing.com)

CasaOS: A simple, easy-to-use, elegant open-source personal cloud system (casaos.zimaspace.com)

What if made we Apache Iceberg feel like a RDMS? (mooncake.dev)

NASA astronaut described eerie moment he saw 'snakes' in space (uniladtech.com)

What if the structure of post-capitalism could be mathematically proven? (github.com)

I built an app to stop me from snoozing my alarm 20 times by looking at the sky (vakeup.app)

Tesla faces ban from selling cars in California (the-independent.com)

Ask HN: React vs. X

First Look: macOS Tahoe Public Beta (sixcolors.com)

Ask HN: Why Are Honest Devs Getting Punished by Algo?

SecureCRT (vandyke.com)

A vote targeting 'pro-China' lawmakers has split Taiwan (bbc.com)

Uber is testing a women-only driver service (washingtonpost.com)

Ask HN: Where can I find detailed course topics of reputed universities?

A new study sheds light on the moon's age and origin story (mag.uchicago.edu)

Benetton's Confusing Legacy of Brand Activism (2021) (jeffswystun.medium.com)

Redis-shield is a well-documented example of building Redis modules in Rust (github.com)

An introduction to fifth generation warfare (greydynamics.com)

Optimizing Tool Selection in LLM Workflows(Part 2): A DSPy and PyTorch Benchmark

Comments (1)