Attention — Nov 11, 2026
What caught my attention on Nov 11 (4 items).
JensenGate and Anthropic’s IPO w/ Ed Elson | Better Offline
Better Offline (Article:11:17) - Published 2026-11-10 18:00 - Discovered 2026-11-11 11:17
In this episode of Better Offline, Ed Zitron and Ed Elson discuss the messy rollout of Anthropic’s IPO and what it signals about the broader AI market. Anthropic initially sought a valuation near $2 trillion, but the offering was ultimately priced much lower, around $1 trillion to $1.4 trillion, after a confusing filing process and revised disclosures. The hosts argue that the company’s S-1 reads…
Nemotron 1.5 Super VL 120B A12B
NoPlaceLikeLocalhost (Article:13:35) - Published 2026-11-10 18:00 - Discovered 2026-11-11 13:35
The video reviews Nemotron 1.5 Super VL 120B A12B, a 120-billion-parameter mixture-of-experts model with 12 billion active parameters, using an NVFP4 quantized version in a Strata fork designed for Blackwell GPUs. The host runs the model locally on a 24GB card, noting that it consumes about 22.5GB of VRAM and around 68GB of system RAM, while achieving roughly 50–70 tokens per second despite the mo…
The Hidden Bottlenecks Slowing Down Large Language Models
Modern Software Engineering (Article:13:39) - Published 2026-11-10 18:00 - Discovered 2026-11-11 13:39
In this episode, Dave Farley argues that large language model performance is shaped less by raw arithmetic than by mechanical sympathy—understanding the hardware beneath software. He draws on low-latency trading systems to show that bottlenecks often come from moving data, not doing math. LLM inference is largely memory-bandwidth bound: generating each token requires reading huge model weights fro…
‘But the AI is great at updating code!’ — no it isn’t
Pivot to AI (Article:16:54) - Published 2026-11-10 18:00 - Discovered 2026-11-11 16:54
The video argues that the common claim that AI chatbots are especially good at updating and refactoring old code is largely false. David Gerard begins with a Cisco executive’s decision to give an employee unlimited AI tokens after being told that a two-person team could rewrite Cisco’s legacy code. The video then examines a test in which several chatbot models were asked to upgrade 100 individual…