Week in review: Nov 1 – Nov 8
A light week where the AI-money beat turned loud — DRAM rationing for data centers and a tech-stock sell-off dragging data-centre loans down with it — alongside MIT's gradient-free model editing and Gamers Nexus on tech money in elections.
A quiet week: five items across three rollup days (Nov 3, 6 and 7), but they line up behind one idea — attention on who pays for the AI buildout, and in what currency.
The AI-money beat took two sharp turns. Pivot to AI first went after Sam Altman’s suggestion that data centers get priority access to the world’s limited DRAM supply over ordinary consumers, then covered the sell-off that followed the US elections, where large investors dumped AI-linked shares and data-centre loans fell with them — the show’s framing being that the market runs on stories and speculative finance rather than underlying returns. Gamers Nexus came at the same question from the political side: a special report arguing that Nvidia, AMD, Intel, Micron, OpenAI, Anthropic and friends use donations, super PACs and lobbying to buy election outcomes and policy.
The one genuinely new technical idea was MIT CSAIL’s weight-space programming, where GPT-6 Astra edits a small model’s weights directly — no training, no gradients — moving Qwen3.5-0.8B’s MATH-500 score from 7.9% to 48.1% by touching 0.016% of its parameters. Rounding it out was a first local look at Apex 35B v1.0, a Qwen 3.6 derivative tested against a practical coding task.
Notable items
- GPT-6 Astra Made Qwen 6x Better at Math Without Training — MIT CSAIL research edits a small model’s weights directly, no training or gradients, multiplying its MATH-500 score.
- Sam Altman wants your RAM for the Compute Campuses — OpenAI’s CEO suggests data centers should get priority access to limited DRAM over ordinary consumers.
- Tech stocks crash — and so do data center loans — a post-election AI-share sell-off read as bubble economics finally meeting missing returns.
- Investigation: How Tech & AI Companies Corrupt Elections — Gamers Nexus maps the donations, super PACs and lobbying Nvidia, OpenAI and others use to shape policy.
- Apex 35B v1.0 - first look and test — a Qwen 3.6 fine-tune put through a real coding task to see if the improvement claims hold up.
From the blog
- Week in review: Oct 25 – Nov 1 — last week’s digest, covering the local-model compression race and the AI-money accountability beat that carried into this week.