weekend ai reads for 2025-06-20

My strategy is to identify where the slowdown is happening and hand off just enough of the task to the LLM to regain momentum. Here’s what that looks like in practice for me

In trade, it’s now possible for machines to understand our thinking that we’ve recorded down into words over time. That means that I am on the cusp of the opportunity to uniquely scale myself by connecting “intelligence on demand for a few cents” with the written details of my thinking built over the past two decades of being a writer who operates.

LLM users also struggled to accurately quote their own work. While LLMs offer immediate convenience, our findings highlight potential cognitive costs. Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels. These results raise concerns about the long-term educational implications of LLM reliance and underscore the need for deeper inquiry into AI's role in learning.

I don’t doubt the relevance of item response theory or p-value in education research. The real problem is in putting these on steroids to reduce and explain every damn ed system of important concern. The end result is the commodification of learning outcomes in favour of what can be measured easily, and scaled cheaply (the cost effectiveness analysis nonsense), instead of what matters for long-term development of the child.

But Less, but better applies not only to outputs; it prompts us to reconsider inputs as well. Could fewer ingredients make healthier products? Could simpler supply chains make businesses more resilient? And the same applies to the way we work.

The lethal trifecta of capabilities is:

1. Access to your private data—one of the most common purposes of tools in the first place!
2. Exposure to untrusted content—any mechanism by which text (or images) controlled by a malicious attacker could become available to your LLM
3. The ability to externally communicate in a way that could be used to steal your data

Nate leads ChatGPT for Work which is now used by 92% of Fortune 500 companies. In our chat, he reveals how OpenAI runs with less than 30 PMs, what they look for in new hires, how he personally uses ChatGPT to save time at work, and more.

We also introduce three ways people engage with AI:

- Automation: The AI completes specific tasks based on your instructions.
- Augmentation: You and AI collaborate as creative thinking and task execution partners.
- Agency: You configure AI to work independently on your behalf, establishing its knowledge and behavior patterns rather than just giving it specific tasks.

With ambient agents, there is a shift in how humans fundamentally interact with AI to get the desired outcomes they need; the AI assistants rely instead on environmental cues.

I created the concept of Prompt 5 Sigma to show how to brief AI with total clarity, exact intent, and almost no room for misinterpretation.

Those deals highlight the strategic importance for legacy software players to own all aspects of data management, and M&A is often the fastest way to achieve it. Instead of building complex data systems from scratch, they are acquiring specialists that can help organize, clean, and connect data from across their business.

The current AI Trust Dilemma is really just the logical endpoint of a process that’s been going on for 40 years (or longer). When we decided to turn higher ed into a gatekeeper of credentials for a job rather than helping students to understand the purpose and value of becoming educated.

OK, I’m sold on multi-agent LLM systems now.

I’ve been pretty skeptical of these until recently: why make your life more complicated by running multiple different prompts in parallel when you can usually get something useful done with a single, carefully-crafted prompt against a frontier model?

This detailed description from Anthropic about how they engineered their "Claude Research" tool has cured me of that skepticism.

If you try to replicate RAG by just hooking your agent up to a bunch of federated MCP servers, your users will experience longer wait times, less relevant results, and inconsistent quality depending on which systems happen to contain their answers.

…

Instead, providing a centralized indexing and retrieval layer across heterogenous data sources and exposing that as a centralized MCP retrieval tool to the agent can enable rapid lookup of semantically relevant content.

During an emergency delivery, an expectant mother channels the strength of those who came before—past matriarchs to dying stars—turning her love into a cosmic force to save her daughter’s life.

‘Ancestra’ is the first of three films from Primordial Soup, a storytelling innovation venture founded by Darren Aronofsky. In collaboration with Google DeepMind, these films show how generative tools can be shaped by artists to expand creativity rather than replace it.

Google DeepMind has written in a report that Gemini 2.5 Pro resorts to panic when its Pokémon are close to death. This can cause the AI’s performance to experience “qualitatively observable degradation in the model’s reasoning capability,” according to the report.