HeyJaredMarkets & Mayhem
Thoughts on the market and big picture. Finding the opportunities where macro meets momentum.
- Indexed issues, last 90 days
- 3
- Latest publication
- Sep 14, 2026
- Audience
- Checking…
- Earliest in this view
- Aug 3, 2026
Latest issues
The Open Model That Fought Back (opens the original)
Read excerpt
On July 13, 2026, engineers at Hugging Face, the company that hosts most of the world’s open AI models, stopped an intrusion that had been running for days. When they sat down to work out what had attacked them, they reached for the most capable models available. The commercial frontier models refused. Their safety guardrails could not tell an incident responder from an attacker, and the forensic work required submitting real exploit payloads, live attack commands, and stolen credentials. The an
Bringing AI Home (opens the original)
Read excerpt
Two years after the generative AI boom funneled enterprise workloads toward a handful of cloud APIs, a measurable share of the market is moving the other way. Surveys of intent point one way while spending still flows the other: companies with sustained inference volume, sensitive data, or regulatory exposure are pulling AI onto infrastructure they control even as most infrastructure dollars continue to land in the cloud. The shift is real, and it is narrower than the marketing around it. Most o
Minimum Viable Baselines for Local LLM Inference (opens the original)
Read excerpt
Executive Summary8-bit KV cache + 4-bit weight quantization is the validated minimum viable baseline for local LLM inference, preserving 94-99% of BF16 qualityBelow 4-bit weights, quality degrades sharply: Q2_K is rated at 80-85% with noticeable quality lossBelow 8-bit KV cache, methods are research-grade (NVFP4 requires Blackwell, OSCAR/TurboQuant are unpublished at production scale)REAP (expert pruning) applies only to MoE models and often causes significant deterioration in quality<l
Publishing over time
Last 90 days. Choose a month to open its work.
Recurring subjects
Named in the text we hold. One piece can cover several.
Audience
No verified audience measurement yet.
About this data
Counts cover the work we have indexed. Tone needs enough text and a confident classification. Excerpts and episode notes are not full articles or transcripts.
Identity or attribution wrong? Suggest a correction.