Monday links: capable models

Quote of the Day
"There are and will be models eminently capable of mounting cybersecurity attacks on existing infrastructure, and those models will be — already are — widely available. The best defense — the only viable defense, in fact — will be to make sure defenders have access to the best models as well."
(Ben Thompson)
Sectors
- Why banks are leading. (trendlabs.com)
- Chip makers are sucking capital from the hyperscalers. (axios.com)
AI
- We live in a world with multiple capable AI models. (stratechery.com)
- It's possible to 'poison' open source models with bad training data. (futurism.com)
Stay in touch
- Don't miss a thing! Sign up for our daily e-mail newsletter. (abnormalreturns.com)
- Do you in wealth management? Sign up for our advisor-exclusive Talking Wealth newsletter. (talkingwealthpod.com)
Coffee
- That coffee habit of yours seems to be good for your heart. (wsj.com)
- Yemeni coffeehouses are having a moment. (theconversation.com)
Economy
- Some signs of improvement in the manufacturing economy. (bonddad.blogspot.com)
- Four questions about the American economy including 'Why hasn’t the housing market led to a recession?' (awealthofcommonsense.com)
Earlier on Abnormal Returns
- Adviser links: fee conversations. (abnormalreturns.com)
- What you missed in our Sunday linkfest. (abnormalreturns.com)
- Top clicks last week on Abnormal Returns. (abnormalreturns.com)

Facts Only

* There are and will be models capable of mounting cybersecurity attacks on existing infrastructure.
* Those models will be widely available.
* The best defense is ensuring defenders have access to the best models.
* Multiple capable AI models exist in the world.
* It is possible to 'poison' open source models with bad training data.
* Banks are leading in a specific sector.
* Chip makers are sucking capital from hyperscalers.
* There are signs of improvement in the manufacturing economy.
* Four questions regarding the American economy were presented, including one about housing market performance.

Executive Summary

A quote from Ben Thompson posits that the best defense in cybersecurity is ensuring defenders have access to the best available models for mounting attacks on existing infrastructure. This is contextualized by several disparate news items across finance, technology, and lifestyle. In technology, there are multiple capable AI models available, and the risk exists that these open-source models can be corrupted through poisoned training data. Related economic themes suggest areas of focus include the banking sector's leading role, the capital flows between chip makers and hyperscalers, and general signs of improvement in the manufacturing economy. The overall theme connects the accessibility and control of advanced models—whether AI or cybersecurity frameworks—to maintaining a defensible position.

Full Take

The juxtaposition of advanced AI capability and cybersecurity mandates immediate attention to epistemology: if models for attack and defense are widely available, the control mechanism shifts from possessing superior technology to controlling access to knowledge. The threat of poisoning open-source models introduces a systemic risk where foundational knowledge itself becomes a vector for deliberate manipulation, moving the battleground from perimeter defense to data integrity across the entire ecosystem. This dynamic echoes historical patterns where control over foundational knowledge determines power distribution. The focus on entities like banks and chip makers suggests that economic power structures are actively shaping which models (and their underlying infrastructure) become most contested or valuable. The implication for agency is that cognitive sovereignty requires not just technical defense, but a critical understanding of the data provenance and the incentives driving the creation and distribution of these capable systems. What mechanisms exist to ensure that access to high-fidelity defense models remains democratized rather than weaponized? What are the externalities borne by those who curate or withhold this knowledge?

Sentinel — Human

Confidence

This text appears to be an aggregated collection of headlines and links from various sources, characteristic of a personalized news feed or newsletter digest rather than a single piece of narrative writing.

Signals Detected

Moderate sentence length variance; the text shifts between declarative statements and direct quotes.

The text is a collection of disparate headlines and links rather than a cohesive argument, typical of a newsletter digest.

Simple juxtaposition of unrelated topics (cybersecurity, AI, coffee, economy) with minimal transitional language.

The structure strongly suggests a curated feed from various, independent sources rather than a single authored piece.

Human Indicators

The inclusion of specific, non-academic, and disparate source links (e.g., trendlabs.com, axios.com, bonddad.blogspot.com) points toward aggregation rather than generative synthesis.

The tone is fragmented and promotional, typical of a curated daily update or newsletter.