The Weekly Review

Nvidia loses its founder as Treasury yields climb back above 5%

Week of 9 November 2026

Published · Updated

The week in five lines

  1. 1

    Jensen Huang resigned as Nvidia’s chief executive on 12 November, and the shares ended the week 12.75% lower at $162.85, days before results on Tuesday.

  2. 2

    Anthropic opened its IPO book at $70–78 a share, with Nasdaq trading expected on 19 November, and an amended filing said it now meets its own threshold for AI-accelerated research.

  3. 3

    Anthropic said 1,000 agents running Claude Fable 5.2 formalised Fermat’s Last Theorem in Lean in 30 hours, and Google DeepMind used agents to verify security properties of the Signal protocol’s code.

  4. 4

    October consumer prices rose 0.4% and core prices 0.3%, and the 10-year Treasury yield rose 29.5 basis points over the week to 5.154%.

  5. 5

    The S&P 500 still gained 2.51% on the week as investors sold Nvidia and bought AMD and Broadcom.

The 10-year Treasury yield rises 29.5 basis points in a week US 10-year Treasury yield (Cboe index), daily close, %
4.4% 4.6% 4.8% 5.0% 5.4% Sep 2026 Oct Nov 5.154%

Source: Yahoo Finance. Latest data: 13 November 2026.

The yield closed at 4.859% on 6 November and at 5.154% on 13 November, its highest close since 14 October. The largest daily rise of the week came on 10 November, the day of the October inflation report, when it went from 4.864% to 4.983%.

Jensen Huang resigns as Nvidia chief executive after 33 years

Nvidia’s board said on 12 November that Huang had resigned as president, chief executive and director with immediate effect. Colette Kress, the finance chief, is interim chief executive and Harvey Jones becomes the first chair of the board. The shares fell 12.75% over the week, and results are due on Tuesday.

· Nvidia: CEO transition · Yahoo Finance: live coverage

Anthropic opens its IPO book and says it meets its AI research threshold

Anthropic is offering 825 million shares at $70–78, which would raise up to $64.4bn, with Nasdaq trading under the symbol ANT expected on 19 November. An amended filing on 10 November said the company now meets the AI research threshold in its own Responsible Scaling Policy.

· Anthropic: roadshow launch · Reuters: order book · The Next Web: research threshold

AI agents formalise Fermat in 30 hours and verify Signal’s code

Anthropic said 1,000 agents running Claude Fable 5.2 completed a Lean formalisation of Fermat’s Last Theorem in 30 hours, against 11 days for its first attempt. Google DeepMind’s AlphaProof Code verified security properties of the Signal protocol’s implementation in four weeks.

· Anthropic: Claude Fable 5.2 and Mythos 5.2 · Anthropic: formalising Fermat’s Last Theorem · Wired: AlphaProof Code and Signal

Hot October inflation lifts the 10-year Treasury yield to 5.15%

Consumer prices rose 0.4% in October and core prices 0.3%, above forecasts, and markets moved to price a Federal Reserve rate rise at the 8–9 December meeting. The 10-year yield closed the week at 5.154%, up 29.5 basis points.

· FatTail: October CPI · Wolf Street: yields and core services · Yahoo Finance: US 10-year Treasury yield

How a 29.5 basis point rise in yields takes 2.3% off a 10-year bond

The 10-year Treasury yield rose from 4.859% to 5.154% this week. For a bond priced at par at the start, a modified duration of \(D \approx 7.85\) gives a loss of 2.31% to first order, and convexity brings it to 2.28%, the exact repricing.

The Race Is Decided Early! A Kernel Law Predicts Lock-In in Majority-Vote Test-Time RL

Hwang et al.

It turns self-consistency training into a quantitative prediction: which answer a question will lock into can be computed before training, from the starting logits and a kernel probe, with no fitted constants. For readers interested in the probability of test-time reasoning it is a clean example of a diffusion-style race analysed in closed form.

Prerequisites
Gaussian tail probabilities, diffusion approximations of noisy gradient updates, and the neural tangent kernel.
Focus on
The two-answer race with separate logits first, then the kernel spillover between answers and its closed form; the LoRA experiments last.
Reading time
About 2 hours
arXiv:2611.11165

Sharp and explicit Berry--Esseen inequalities over convex sets

Zhilova

Normal approximation over convex sets is what justifies the high-dimensional bootstrap and simultaneous confidence regions. The paper gives explicit constants and shows that four moments are the threshold: bounds on the fourth coordinate moments give a worst-case error linear in \(d\), and higher moments do not improve the rate.

Prerequisites
Lindeberg and Stein-type normal approximation, Gaussian smoothing of indicator functions, and Rademacher sums.
Focus on
The main theorems and the lower-bound construction first, then the Gaussian-smoothing argument for Rademacher sums on Parseval frames and the random swaps that extend it to general sums.
Reading time
About 3 hours
arXiv:2611.13106
Learning theory

A Theory of Policy Optimization in Post-Training: Coverage Collapse and Reference Resets

Huang et al.

REINFORCE, the basis of PPO and GRPO, can drive the probability of optimal responses towards zero even as reward rises, and KL regularisation to the base model does not prevent it. Repeatedly resetting the KL reference to the latest policy does, and gives the first global convergence guarantees for these families of algorithms.

arXiv:2611.14467
Optimization

e-means: outperforming k-means (in one-tenth the epochs)

Bessa et al.

A mini-batch k-means with exponential moving average updates, whose decay rate and batch size are fixed by fitted scaling laws, beats the loss of 25 epochs of Lloyd’s algorithm in 1.4 to 2.6 epochs, a 10–45 times speed-up on GPUs.

arXiv:2611.09501
Large language models

Latent Sparse Attention: Towards Subquadratic Ultra-Long Context Attention

Kubicek et al.

Keys are clustered hierarchically in a learned latent space and queries score clusters rather than every key. At 10 million tokens one attention layer runs 22.7 times faster in prefill than GLM-5.3-Flash’s own sparse attention and 72.4 times faster than full attention in Qwen-3.8-27B, with similar accuracy.

arXiv:2610.42147
Economy

Organizing Intelligence

Imas et al.

A model of firms in which human and AI agents sort into competing organisations. When AI coordination improves, the Coasean ceiling on firm size can break and the market goes through a phase transition towards large, concentrated, AI-heavy firms.

arXiv:2611.05417

Mollick argues Fable 5.2 is close to a jagged general intelligence

Ethan Mollick, a professor of entrepreneurship who writes One Useful Thing, defined general AI as a system as good as most experts at most intellectual work done on a computer, and concluded that Fable 5.2 comes close, with gaps in taste and judgement. Greg Brockman, president of OpenAI, wrote that frontier models with heavy test-time compute are becoming the engine of scientific discovery.

· One Useful Thing: Fable 5.2 and jagged AGI · Discussion on X: Ethan Mollick · Greg Brockman on X

Chollet says scale multiplies generalisation rather than replacing it

Francois Chollet, creator of Keras, answered Richard Sutton by arguing that more data and compute multiply the generalisation power of a method, so they cannot rescue a weak one. Yann LeCun separated joint-embedding from joint-embedding predictive architectures and defended the second against critics who point to weaker benchmarks.

Francois Chollet on X · Yann LeCun on X · Discussion on X: LeCun

AI Futures Project calls for a pause after METR’s Anthropic report

The authors of the AI 2027 scenario said that, on the evidence of METR’s report, they all now expect a superhuman AI researcher during 2027 unless the industry slows down, and called for an industry-wide pause. Anthropic’s founders wrote that they may choose to move more slowly than they could.

· The Next Web: Anthropic’s threshold and the reaction

Investors sell Nvidia and buy AMD and Broadcom

Nvidia fell 12.75% over the week to $162.85 while AMD rose 18.93% to $534.06 and Broadcom 18.16% to $347.01, so the AI chip trade split rather than collapsed.

Nvidia share price, $$162.85
Daily closes, 14 Aug to 13 Nov 2026 · Yahoo Finance
Yahoo Finance: Nvidia · Yahoo Finance: AMD · Yahoo Finance: Broadcom

S&P 500 gains 2.51% in a week of rotation

The S&P 500 closed at 7,561.67 on Friday, up 2.51% from 7,376.86 a week earlier, after a 2.3% rise on Monday. The Nasdaq Composite gained 2.27% to 25,136.27.

S&P 500, index points7,561.67
Daily closes, 14 Aug to 13 Nov 2026 · Yahoo Finance
Yahoo Finance: S&P 500 · Yahoo Finance: Nasdaq Composite

Brent falls for a second week, to $95.84

Brent crude futures settled at $95.84 on Friday, down from $98.32 a week earlier and well below the $117.64 close of 30 October.

Brent crude futures, $ a barrel$95.84
Daily closes, 17 Aug to 13 Nov 2026 · Yahoo Finance
Yahoo Finance: Brent crude futures · FatTail: October CPI

Gold falls 4.63% as Treasury yields climb

Gold futures settled at $3,886.30 on Friday, down 4.63% from $4,074.90 a week earlier, with most of the fall on Thursday as markets priced a December rate rise.

Gold futures, $ an ounce$3,886.30
Daily closes, 17 Aug to 13 Nov 2026 · Yahoo Finance
Yahoo Finance: gold futures

October consumer prices arrive without an official annual rate

The Bureau of Labor Statistics published no 12-month change for headline or core prices, because October 2025 data were never collected during the 2025 shutdown. Monthly figures and shorter annualised rates are what remain.

· FatTail: October CPI · Wolf Street: October CPI

Economists expect the Fed to raise rates in December

All 101 economists in a Reuters poll expect the Federal Reserve to raise rates by 25 basis points at its 8–9 December meeting. Kevin Warsh, chair since 22 May, speaks on Monday and the minutes of the October meeting follow on Wednesday.

FatTail: October CPI · Kiplinger: the week’s economic calendar

Gamarnik writes a human-friendly account of OpenAI’s k-SAT proof

David Gamarnik published a detailed exposition of OpenAI’s proof of the satisfiability conjecture for random k-SAT, which says the probability of satisfiability drops sharply from near 1 to near 0 at a fixed ratio of clauses to variables.

· arXiv: Gamarnik