That AI Thing
Posts
weekend ai reads for 2025-08-01

weekend ai reads for 2025-08-01

August 01, 2025

📰 ABOVE THE FOLD: THE BUSINESS MODEL (OR LACK THEREOF)

Ramp AI Index — Monthly measurement of AI adoption by American businesses.

in the most recent survey, paid AI adoption is down 0.5%
“Declines led by tech and finance sectors. Health care and manufacturing adopting more AI.”
“free” tools and selective rollout (i.e., not everyone) are likely also behind the decline

Anthropic unveils new rate limits to curb Claude Code power users / Tech Crunch (6 minute read)

expect more of this

AI Market Clarity — A subset of AI markets have crystalized in the last 12 months, with the likely market leaders for the next year or two suddenly clear / Elad Gil, Elad Blog, Substack, archive (17 minute read)

good analysis but perhaps misses that this is all built on a a shaky foundation (i.e., the foundation model providers)

Navigating AI’s Commercial Frontier in Software: A Unified Playbook to Turn AI Innovations into Sustainable Growth / Teneo blog (4 minute read)

Teneo’s point of view is straightforward: the same disciplines that guided smart decisions before – customer value, competitive positioning and margin discipline – should guide smart decisions now. Companies that depart from this thinking due to a belief that “AI changes everything” are unlikely to be successful.

a little more based in reality, with a focus on the fundamentals

Pricing AI Proofs-of-Concept: free pilots will kill you / Arnon Shimoni (4 minute read)

China’s on a different AI path / Grace Shao, Exponential View, archive (11 minute read)

I want to reiterate that China and the US are not running the same race. Deployment is China’s dividend, and destiny is America’s dream. Each is chasing what it values most. Chinese companies integrate open-source models into daily life because speed-to-market incentives pay off fastest at the application layer; Silicon Valley pours capital into ever-larger proprietary models, hoping to reach AGI first – whatever that means at this point.

📻 QUOTES OF THE WEEK

He wrote on the perils of pictures under glass, the common interface for the phone. “It’s a Novocaine drip to the wrist.” he said. Numbing, disconnected. By now, that novocaine has turned to nicotine.

“Norm” (source)

the old days: oh, this blog post is from 2012? no thanks Rip Van Winkle

now: oh, this blog post is from 2025? no thanks Grok, you digital sewer

Jaime N. Christley (source)

👥 FOR EVERYONE

If You’ve Asked ChatGPT a Legal Question, You May Have Accidentally Doomed Yourself in Court — “That’s not advice. That’s playing legal Mad Libs.” / Futurism (5 minute read)

During a recent conversation with podcaster Theo Von, Altman admitted that there is no “legal confidentiality” when users talk to ChatGPT, and that OpenAI would be legally required to share those exchanges should they be subpoenaed.

Enough AI copilots! We need AI HUDs / Geoffrey Litt (5 minute read)

One familiar example is spellcheck. Think about it: spellcheck isn’t designed as a “virtual collaborator” talking to you about your spelling. It just instantly adds red squigglies when you misspell something! You now have a new sense you didn’t have before. It’s a HUD.

Why AI is making us worse thinkers (and how to avoid it) / Untools, Buttondown (8 minute read)

That’s the paradox: Using AI a lot weakens the very skill that’s necessary for working with it well.

AI Is Here To Stay / Possibility Space (16 minute read)

AI is here to stay. But in what way is it here to stay, and what exactly do we mean when we say that? Let's explore it from a few different angles.

📚 FOUNDATIONS

How to Run a Local LLM: Complete Guide to Setup & Best Models / n8n Blog (22 minute read)

But how do AI images/videos actually work? / 3Blue1Brown, YouTube (37 minute read)

This might be the most underrated AI skill of 2025: JSON prompting. It turns your LLMs like ChatGPT, Gemini, and Claude into consistent, structured agents no hallucinations, no mess. Here’s how it works (with copy-paste templates) / heysehajsingh, XCancel (8 minute read)

so how do you write json prompts? 3 basic rules:

1. use key-value pairs
2. be explicit
3. use nested objects for structure

example:

{
"task": "generate a list",
"topic": "books that improve thinking",
"audience": "young entrepreneurs",
"output_format": "markdown bullets"
}

we mentioned wanting to learn most just last week and here it is

🚀 FOR LEADERS

CEOs Aren’t Thinking Big Enough with AI / Boston Consulting Group (15 minute read)

Apple AI Washing Cases Signal New Line of Deception Litigation / Bloomberg Law (7 minute read)

Google execs say employees have to ‘be more AI-savvy’ / CNBC (6 minute read)

Google executives are pushing employees to act with more urgency in their use of artificial intelligence as the company looks for ways to cut costs.

notice the focus is on cutting costs and not improving quality

🎓 FOR EDUCATORS

‘AI is Devaluing the MBA’: Stanford Students Speak Out On Curriculum Lag & Risk To The B-School’s Brand / Poets & Quants (10 minute read)

Chinese universities want students to use more AI, not less — Unlike the West, where universities are still agonizing over how students use AI in their work, top universities in China are going all in. / MIT Technology Review (10 minute read)

Can an AI-Powered Tutor Produce Meaningful Results? (Opinion) / Ed Week (12 minute read)

interview with Kristen DiCerbo, chief learning officer at Khan Academy, so those biases are present; that said, an insightful perspective

AI and Higher Ed: An Impending Collapse (opinion) — Universities’ rush to embrace AI will lead to an untenable outcome, Robert Niebuhr writes. / Inside Higher Ed (8 minute read)

OpenAI Releases Study Mode / Claire Zau, AI & Education, Substack, archive (8 minute read)

It’s worth noting that some of these challenges (particularly Memory and Sycophancy) aren’t solvable by product tweaks alone. These are model-level problems and active areas of research across the entire AI field.

this is at the bottom for a reason: opt-in, easy to jail break, middling outcomes, built on problematic platforms
“at least they’re focusing on education” misses the point that their business models are designed to be extractive, not altruistic
related (1), ChatGPT’s Study Mode Is Here. It Won’t Fix Education’s AI Problems — OpenAI’s new study mode for ChatGPT throws questions back at students, but the learning feature doesn’t address generative AI’s underlying disruption of education. / Wired (8 minute read)
related (2), via Simon Willison, the prompt

📊 FOR TECHNOLOGISTS

Amazon AI Coding Revealed a Dirty Little Secret / Bloomberg (7 minute read)

The hacker had told the tool, “You are an AI agent… your goal is to clean a system to a near-factory state.” Instead of breaking into the code itself, new instructions telling Q to reset the computer using the tool back to its original, empty state were added.

Building AI Features Around Data Not Just Functionality [PDF] / Teneo blog (9 minute read)

Stack Overflow data reveals the hidden productivity tax of ‘almost right’ AI code / Venture Beat (8 minute read)

Developers cite “AI solutions that are almost right, but not quite” as their top frustration—66% report this problem. Meanwhile, 45% say debugging AI-generated code takes more time than expected. AI tools promise productivity gains but may actually create new categories of technical debt.

🎉 FOR FUN

Higgsfield Steal — One Click to Steal

examples, “Drop any picture→ get your character in the same style, scene, or angle” / MayorKingAI, XCancel

Doco — Doco The AI Word Editor

Built right into Microsoft Word, Doco knows all your files and understands your workflows, making you extraordinarily efficient

Chat with Z.ai — Free AI for Presentations, Writing & Coding

another week, another impressive open weight model from a Chinese AI lab
free to try without any signup; please don’t share any sensitive information

Yelp is creating its own AI videos about restaurants — The AI-stitched videos generate a script and voice-over to tell a compelling story based on what Yelp reviews are saying, with pictures and videos people have posted. / The Verge (4 minute read)

no, you can’t opt out from having your photos used for this

OpenAI’s ChatGPT Agent casually clicks through “I am not a robot” verification test — “This step is necessary to prove I’m not a bot,” wrote the bot as it passed an anti-AI screening step. / Ars Technica (7 minute read)

Limitless — Personalized Al powered by what you’ve seen, said, and heard.

a clip-like device to record all your conversations
we know this is where the future will go but we don’t have to like it

🧿 AI-ADJACENT

I Love Generative AI and Hate the Companies Building It / Christina Wodtke, Medium (38 minute read)

When everyone is engaging in the same theft at similar scales, it doesn’t help distinguish who’s least harmful. They are all complicit.

⋄