Free · Weekly · Tuesdays
The five AI talks worth your time this week.
— and the exact minute to jump to.
Hours of AI talks drop every week. I monitor the channels that matter, cut the hype, and send you the five worth watching — with what you'll learn and where to jump.
Free. One email a week. Unsubscribe anytime.
Semantic Blindness: 500,000 Sensors Confused an LLM - Raahul Singh & Vanč Levstik, Phaidra
Raahul Singh & Vanč Levstik · AI Engineer · 16 min
Phaidra's engineers had to make an LLM reason over 500,000+ sensor names in gigawatt-scale data centers, and every naive approach — vector RAG, pure LLM, parallel sharding — failed on recall or hallucinated phantom equipment. Their fix is a clean architectural pattern: the LLM plans and judges, deterministic code does the lookup and set operations, and cost scales with the hierarchy's depth rather than the number of instances. The production numbers are the best argument: from 116M tokens and 30% accuracy down to 390K tokens and 100% accuracy across 66 real cases.
Read the full issueLatest issue · Nº 001 · Jul 6 – Jul 13, 2026
2h 20m of talks → one 4-minute read
LLMs for judgment, code for counting: scaling to 500K sensors on 300x fewer tokens
- 01Semantic Blindness: 500,000 Sensors Confused an LLM - Raahul Singh & Vanč Levstik, PhaidraRaahul Singh & Vanč Levstik · AI Engineer16 minwatch
- 02How to Reduce LLM LatencyHamel Husain42 minwatch
- 03Understanding is the new bottleneck — Geoffrey Litt, NotionGeoffrey Litt · AI Engineer20 minwatch
- 04What Does Done Even Mean? Agents and Paperclip's Liveness Model - Dotta, PaperclipDotta · AI Engineer7 minskim
- 05Stop AI Agent Hallucinations: 5 Techniques + Production Patterns - Elizabeth Fuentes, AWSElizabeth Fuentes · AI Engineer55 minskim
The filter
Substance over hype. Every single week.
Most AI newsletters amplify whatever is loudest. This one does the opposite: primary sources, technical depth, real insight — or it doesn't make the cut. If a talk is a product pitch in disguise, it's out. If a week is thin, I'll say so instead of padding it.
— Héctor, builder & curator
CURATING FROM Anthropic · OpenAI · Google DeepMind · AI Engineer · Latent Space · Stanford Online + 35 more channels
How it works
01
Monitor
41 hand-picked channels — research labs, conferences, builder podcasts. Every new talk lands in the pipeline within hours.
02
Score
Each talk is scored for substance. Clickbait framing, news rehash and product marketing get cut before I ever see them.
03
Distill
I pick the five that matter and write why — what you'll learn, and the exact minute to jump to. Five picks, one four-minute read.
Five talks. Four minutes.
Every Tuesday.
Issue Nº 002 ships next Tuesday. Founding readers get it first.
Free. One email a week. Unsubscribe anytime.