Joe Sack Substack
Subscribe
Sign in
Home
Notes
AI & Databases
Product & Leadership
About
New Course: SQL Server 2025 Vector Search Essentials
When keywords fail and intent matters.
Dec 16, 2025
•
Joe Sack
3
The RAG Tax
Sergii Voloshyn's L-RAG paper shows how to skip retrieval when the model already knows
Jan 14
•
Joe Sack
3
When Hallucinations = Malpractice
LLMs make confident claims without evidence.
Jan 12
•
Joe Sack
5
4
Most Popular
View all
Poisoned Documents and Redundant Context
Jan 5
•
Joe Sack
3
110 Days
Dec 31, 2025
•
Joe Sack
12
6
I Asked for 10 Results. I Needed 2000 Candidates.
Oct 24, 2025
•
Joe Sack
4
8
Beach Glass
Aug 23, 2025
•
Joe Sack
8
1
Before You Pick a Vector Database
Dec 12, 2025
•
Joe Sack
4
2
1
The Practitioner to PM Question
Nov 1, 2025
•
Joe Sack
7
1
Latest
Top
Discussions
Notes from Somewhere in the Context Window
I’ve been an early fan of Claude Code and recently I see it getting lots of well-deserved positive press.
Jan 9
•
Joe Sack
4
2
Poisoned Documents and Redundant Context
Two December papers on retrieval problems worth knowing about
Jan 5
•
Joe Sack
3
Kicking the Tires on BGE-M3
One model. Three modes. Three different answers.
Jan 3
•
Joe Sack
1
110 Days
In 2023, my boss at MongoDB sent me an article about LLMs and hallucinations.
Dec 31, 2025
•
Joe Sack
12
6
The Cross-Encoder Fix (And Its Cost)
In my last post, I showed that vector search treats “home with pool” and “home without pool” as nearly identical (0.82 similarity).
Dec 19, 2025
•
Joe Sack
3
1
The Negation Problem in Vector Search
I embedded two queries: “home with pool” and “home without pool.” The cosine similarity was 0.82.
Dec 19, 2025
•
Joe Sack
2
Before You Pick a Vector Database
Five production realities worth understanding first
Dec 12, 2025
•
Joe Sack
4
2
1
See all
Joe Sack Substack
I write and teach about Product, Databases, and AI
Subscribe
Joe Sack Substack
Subscribe
About
Archive
Sitemap
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts
