COMPUTING

Handling Dynamic Document Updates in AI Retrieval Systems

By

–

24 April 2025 15h38

For (2), documents that change and where you want the updates to be reflected quickly in the retrieval system can be challenging. You usually end up building a "fresh" layer of your system that has very different data structures that can handle updates/insertions, but that is

Knowledge Graphs Infrastructure Impact Web Corpus Scale

By

–

24 April 2025 15h34

For (1), knowledge graphs help with lots of higher level kind of things, search presentation aspects, etc., but from an infrastructure perspective, they don't have a huge effect because they just aren't very large relative to the web corpus.

→ View original post on X — @groqinc

Groq’s LPU Chip: Shaping AI Inference Future Today

By

@groqinc

–

24 April 2025 1h06

Groq saw inference coming almost a decade ago – it's why the LPU, our purpose-built chip, is shaping what's possible for AI inference today. Learn more in this @PracticalAIFM episode where @djays shares his insights for tech leaders. Listen here:

→ View original post on X — @cerebras

60% Compute Savings Achievement for Large Model Training

By

@cerebras

–

23 April 2025 22h51

The result: up to 60% compute savings (!)
That’s a massive impact for training large models efficiently.

→ View original post on X — @latentspacepod

SF Compute: Commoditizing the Compute Market

By

@latentspacepod

–

23 April 2025 22h35

SF Compute: Commoditizing Compute https://
latent.space/p/sfcompute We're excited for our latest deep dive into the compute market with @evanjconrad of @sfcompute! It should not be normal for the prices of one of the world’s most important resources right now to swing from $8 to $1

→ View original post on X — @aymericroucher

Text embeddings rediscover BM25 core principles during training

By

@aymericroucher

–

23 April 2025 12h01

-> Researchers show that text embeddings methods re-discover BM25 from scratch during training! Far from abandoning standard search techniques like BM25, it appears that neural search methods naturally re-implement their core principles. Let's you have a search query like

In-Memory Index Redundancy and Distributed Cluster Architecture

By

–

23 April 2025 1h18

The problem was that an in memory index, because of its size, did not have the same level of redundancy on a per-cluster basis as a disk-based index (e.g. usually every page was present in the memory of just a single machine).

Unpublished Search Systems Work from 2009 Conference Talk

By

–

23 April 2025 0h33

We never really published any papers about lots of interesting systems work that underlies our search systems, so this relatively little known talk I gave at the Web Search and Data Mining conference in 2009 is the best public source of info.

Google Search Evolution: From Disk to In-Memory Systems

By

–

23 April 2025 0h08

Someone just reminded me of this lecture I gave in 2009 that described the evolution of Google Search from 1999 to 2009. People who are interested in how our search systems work might find this interesting. It touches on disk-based serving systems, in-memory indices,