From the archive
Tuesday, 27 October
Folded from 10 sources that day.
Sorted by quality and by how many sources covered it, the same for everyone.
-
Why an AI agent's revenue query was $3.5 million wrong, and how AWS fixed it with dbt Semantic Layer
AWS shows a raw SQL query overstating revenue by 30.8% and proposes letting agents pick only certified metrics from the dbt Semantic Layer. Read-only Redshift MCP access is enforced at three layers, and each answer carries its compiled SQL.
-
How Ecolab used an AI migration agent to shrink 84,000 lines of legacy SQL by 6x
Ecolab moved a legacy Snowflake DML system into more than 40 dbt projects using a Datafold AI migration agent. A flagship dimension dropped from 409 models to 170, with row-level data diffs checking each model.
-
Build specialized multimodal models with LFM3
LFM3 is a family of about 2B-parameter models for text, vision, audio, omni and decision tasks on one backbone. LFM3-1.7B-Base beats LFM2.5-2.6B with 35% fewer parameters, and the text model fits in about 1 GB at 4-bit.
-
AI SRE: what is it, how could it change on-call – and does it work?
Examines the rise of AI SRE tools for faster incident investigation, using input from three vendors on agent design, measuring accuracy, cost and latency, and how autonomous they can realistically be.
-
Genie 4: A new generation of world simulation
Genie 4 is a natively multimodal world model that generates synchronized audio and responds to real-time text and keyboard input. It is in research preview for trusted testers and Google AI Ultra subscribers.