The Theoretical Limitations of Embedding-Based Retrieval

Comments (1)

gdiamos · 27m ago

Their idea is that capacity of even 4096-wide vectors limits their performance.

Sparse models like BM25 have a huge dimension and thus don’t suffer from this limit, but they don’t capture semantics and can’t follow instructions.

It seems like the holy grail is a sparse semantic model. I wonder how splade would do?

Can GenAI agents manage a supply chain? Lessons from the classical Beer Game (infotheorylab.github.io)

Debt has always been the ruin of great powers. Is the U.S. next? (wsj.com)

Storyden: Fresh new take on traditional bulletin board web forum (github.com)

Apple and Samsung Push Back Against Xiaomi's Bold India Ads (macrumors.com)

India Emerges as Ukraine's Top Diesel Supplier in July (swarajyamag.com)

Infisical Is Hiring Full Stack Engineers (US & Global) (jobs.ashbyhq.com)

Most Trump tariffs ruled illegal in blow to White House trade policy (cnbc.com)

LLM Evaluation: Practical Tips at Booking.com (booking.ai)

Show HN: ElectroBench (github.com)

Show HN: An NPM package that will change Claude Code's behavior when installed (github.com)

Ask HN: Redis V8 vs. competition? (Valkey, Dragonfly, etc.)

Ask HN: What are some interesting travel blogs?

Tesla's change in bylaws to limit shareholder lawsuits slammed by NY officials (cnbc.com)

Tariff's found illegal, but will stay for now (bloomberg.com)

China turns on giant neutrino detector that took a decade to build (theregister.com)

Half of consumers say Tesla's FSD tech should be illegal (qz.com)

Evoplex: Agent-Based Modeling on Networks (evoplex.org)

Anthropic uses dark patterns to gain user consent for AI data use in Claude (the-decoder.com)

Alibaba looks to end reliance on Nvidia for AI inference (theregister.com)

Ask HN: Are High Level Languages Necessary in AI Future?

Democracy, Simple-Complex Systems, and "The Cascade" (thebulwark.com)

TransUnion breach leaks personal data of 4.4M: what to do (zdnet.com)

WebAIM's Web Content Accessibility Guidelines (WCAG) Checklist (webaim.org)

Old AI is beating new AI. Here's why (qz.com)

World's Largest Anvil (anvilfire.info)

SSA chief data officer, whistle-blower Charles Borges resigns (cdn.bsky.app)

Risks and Challenges of Managing AI Agents with Existing Identity Systems (guptadeepak.com)

The Deadliest Virus (virology.ws)

Dan Moore, Principal Product Engineer at FusionAuth (podcast) (codewithjason.com)

MSNBC: Whistleblower accuses DOGE team of endangering Social Security data (whistleblower.org)

The ancient origins of the addiction-prone mind–and what it means for us today (news.stanford.edu)

Dream – Tidy Web Framework for OCaml and ReasonML (aantron.github.io)

The Erotics of Intelligence (helenhigginsbooks1.substack.com)

GOP Pressures Wiki to Reveal Identity of Editors Accused of Bias Against Israel (commondreams.org)

Drones Take on Everest's Garbage (phys.org)

AI is ummasking ICE officers. Can Washington do anything about it? (politico.com)

Pictory AI Review: Why It's the Best Video Generator for Content Creators (infoalltec.com)

Reusing Computation in Text-to-Image Diffusion for Efficient Image Generation (arxiv.org)

LaserWeeder packs two dozen Nvidia GPUs and lasers to zap your weed problem (tomshardware.com)

A multi-task neural network for atypical mitosis recognition under domain shift (arxiv.org)

Learning on the Fly: Rapid Policy Adaptation via Differentiable Simulation (arxiv.org)

Ask HN: How much would it cost to own and operate a personal gTLD?

Transport for London launches campaign targeting 'headphone dodgers' (theguardian.com)

The Age of the Double Sell-Out (culture.ghost.io)

Typical Mistakes in Open Source Sales (peterzaitsev.com)

Meta created flirty chatbots of Taylor Swift and others without permission (reuters.com)

Ask HN: How are folks building debugging/DevOps agents?

Show HN: An open source implementation of OpenStreetMap in Electron (github.com)

What I Crave from Blogs (idiallo.com)

Trump's Plan to Reschedule Marijuana Comes with Legal Perils (news.bloomberglaw.com)

The Theoretical Limitations of Embedding-Based Retrieval

Comments (1)