weekend ai reads for 2025-09-19

As with all technology, there are pros and cons to AI for kids, but parental involvement in navigating it is key. Kucirkova notes: “AI introduces what has been called the ‘third digital divide’: families with resources can guide their children’s use of technology, while others cannot. Parents who come home exhausted from long hours or multiple jobs may see AI-powered chatbots as a way for their child to have someone responsive to talk to.”

If you’ve never had to shovel [manure], not only do you lack empathy for those who do, but you also tend to create more [manure] for others to shovel.

now that coding’s been solved i spend most of my time thinking and thinking is honestly so much harder than writing code

my brain hurts

Just a week ago Anthropic raised $13 billion in a deal that valued the company at $183 billion; the settlement sum represents less than one percent of that valuation. In order to achieve that valuation, Anthropic stole the work of many thousands of people, and then paid less than 1/100th of their valuation to simply make the whole thing go away—the settlement does not even require Anthropic to admit any wrongdoing. As a country, we should not allow multibillion-dollar companies to commit theft at vast scale and then settle out of court for a negligible fraction of their ill-gotten gains.

No, but it does mean that conversation-fixated humans have latched onto the most conventional and conversational aspect of language models, and thrown that at every application in sight. While it sometimes works, it often does not. To this way of thinking, ChatGPT was the original error: a seductive service that appealed to our biases, to a fault.

However, once you introduce AI, the machine has to parse an input, break it into tokens, reason about meaning, generate an output, and then render it. Unless you have enough processing power, you can’t do that in real-time at 800 words per minute.

The price point reflects the “intuition” and technical skills needed to keep pace with a rapidly-changing technology, Tanmai Gopal, PromptQL’s cofounder and CEO, told Fortune.

Gopal said the company hourly wage for AI engineers as consultants is “aligned with the going rate that you would see for AI engineers,” but that “it feels like we should be increasing that price even more,” as customers aren’t pushing back on the price PromptQL sets.

Learn Your Way is grounded in learning science and powered by LearnLM, our best-in-class pedagogy-infused family of models, now integrated directly into Gemini 2.5 Pro. It adapts content to a learner's selected grade level and personal interests, and generates multiple representations based on the source material, from mind maps and audio lessons to interactive quizzes which enable real-time feedback and further content personalization. It gives students agency over their learning process.

That dwarfs the budgets of any computer science department’s GPU budgets, where grad students are often forced to scrounge for the graphics processing units required to train and run AI models. Plus, many universities are tightening their belts, making it harder for computer science departments to justify spending money on expensive facilities.

As the scope of AI research grows, some professors say GPUs could determine recruiting outcomes. “I think some universities are realizing that, more and more, this type of research requires those kind of resources and they’re allocating resources there,” said Alvarez-Melis, who is also part of the Kempner Institute faculty.

Elise Porter, the executive director of Harvard’s Kempner Institute, said finding funds to grow their cluster is important for the organization. But it’s not as easy as pursuing gift-giving with naming rights. “It’s a critical tool for us to be able to do this, and it is extraordinarily expensive,” she said. “I think those donors are out there, but they’re few and far between.”

At Origin, our breakthrough was architecting a system that pairs the contextual reasoning of frontier LLMs with deterministic computational engines. The LLM layer interprets complex financial scenarios and orchestrates tasks, while deterministic modules handle the underlying math with absolute precision.

Malekzadeh estimates he spends around 50% of his time writing requirements, 10% to 20% of his time on vibe coding, and 30% to 40% of his time on vibe fixing — remedying the bugs and “unnecessary script” created by AI-written code.

He also doesn’t think vibe coding is the best at systems thinking — the process of seeing how a complex problem could impact an overall result. AI-generated code, he said, tries to solve more surface-level problems.

Things are now getting a bit clearer. Essentially, Ontology is Palantir’s way of refining, structuring, and connecting the myriad kinds of data and information pipelines companies constantly use, creating a new stable foundation on which the company can run Palantir software.

You can see why this could be a pretty big advantage.

A day later, my swing was different and self recorded video sent to Google’s Gemini confirmed the change. Swing errors that were decades in the making were corrected in the span of minutes. I’m not saying that I’ve suddenly made a leap from “Struggles to break 100” to “scratch golfer.” I’m just saying that a process that could have been expensive and arduous was instead efficient and relatively cheap. I apply the LLM’s fix, and it tells me whether I’ve actually applied it. The feedback is instant and objective.