weekend ai reads for 2025-04-18

šŸ“° ABOVE THE FOLD: PATIENT CARE

But it turns out that AI may also be better, under some conditions, at providing the most human parts of doctoring — compassion and empathy. This revelation, supported by a growing body of research, is reshaping what patients expect of their doctors and, increasingly, how doctors interact with people they care for.

AI models miss disease in Black and female patients — Analysis of chest x-rays underscores need for monitoring artificial intelligence tools for bias, experts say / Science (8 minute read)

Participants’ walking distance and navigation time improved by 25% compared with using a cane when completing a 25-metre-long indoor maze.

Briscoe used to go to therapy weekly. After being briefly laid off for about a month in late February, she decided to turn to OpenAI’s ChatGPT chatbot for mini therapy sessions. She now uses it several times a day, bumping up against the 10-free-message limit. ā€œTherapy is expensive,ā€ she says. ā€œNo one has an extra $200 to spend to talk to someone.ā€

 

šŸ“» QUOTES OF THE WEEK

Compare the feeling of doomscrolling to kneading dough, playing an instrument, sketching... these take effort, but they're also deeply satisfying. When you strip away too much friction, meaning and satisfaction go with it.

Amelia Wattenberger (source)

 

And when you have the mind virus, you become committed to your human baby. But it’s much more important to be committed to the AI of the future. And so to have human babies is fundamentally unethical.

Jaron Lanier, paraphrasing others (source)

 

šŸ‘„ FOR EVERYONE

ā€˜We tried to train it like it was a kid in art school’: artist David Salle on using an AI model to enhance his painting practice — The New York-based painter’s work with machine learning generates backgrounds based on his previous work, which he then transforms into new paintings—with some of the results now on show in London / The Art Newspaper (10 minute read)

Nate said its app’s users could buy from any e-commerce site with a single click, thanks to AI. In reality, however, Nate relied heavily on hundreds of human contractors in a call center in the Philippines to manually complete those purchases, the DOJ’s Southern District of New York alleges.

ā€œAI-generated content continues to flood streaming platforms like Deezer and we see no sign of it slowing down,ā€ said Aurelien Herault, the company’s innovation chief.

Herault added that a detection tool launched in January was helping the company filter fully AI-generated tracks from the algorithmic recommendations for its 9.7 million subscribers.

AI praise-giving tool promises ā€˜authentic’ insights — Irish company Workhuman says new technology can help workers feel more valued / Financial Times, archive (14 minute read)

  • gross

Furthermore, Meta says, that while the company ā€œhas invested hundreds of millions of dollars in LLM development,ā€ they see no market in paying authors to license their books because ā€œfor there to be a market, there must be something of value to exchange, but none of Plaintiffs works has economic value, individually, as training data.ā€ (An argument essential to fair use, but that also sounds like a scaled up version of a scenario in which the New York Philharmonic board argues against paying individual members of the orchestra because the organization spent a lot of money on the upkeep of David Geffen Hall, and also, a solo bassoon cannot play every part in ā€œThe Rite of Spring.ā€)

  • also gross but not surprising

  • we tend to agree with James Cameron that perhaps plagiarism accusations should be based on outputs, not inputs

 

šŸ“š FOUNDATIONS

What are Gemini Gems? And how to use them / Zapier blog (8 minute read)

Gemini Gems are custom versions of Google Gemini that you can program with specific instructions and knowledge files for it to consult every time it responds.

RAG vs Fine-Tuning: Choosing Optimal AI Strategies / Clickup blog (18 minute read)

RAG equips LLMs with real-time, external data retrieval, reducing retraining costs

Fine-tuning optimizes LLMs by training on specialized datasets, improving accuracy for domain-specific tasks

RAG is best for fast-changing data environments like finance, legal updates, and customer support

Fine-tuning is ideal for brand-specific AI, compliance-heavy industries, and sentiment analysis

via george, Voice AI & Voice Agents — An Illustrated Primer

If you are building conversational AI applications, 800ms voice-to-voice latency is a good target to aim for. Here's a breakdown of a voice-to-voice round trip from a user’s microphone, to the cloud, and back.

 

šŸš€ FOR LEADERS

The appointments come as companies look to move beyond experimenting with AI, and find ways to actually realize the business efficiencies they hope it can deliver – it’s become tablestakes. They’re finding they need a driving personality to get those changes off the deck and into motion.

But the promise of agentic AI involves coordinating numerous AI systems, orchestrating multiple intelligent systems to connect agents across teams, clients and platforms. Without it, the risk of conflicting behavior, redundancy, or outright failure goes up fast.

In other words, part of WPP’s bet is that the real competitive edge won’t come from using AI so much as it will come from responsibly managing, integrating and scaling it before rivals catch up.

Stop Managing AI Projects Like Traditional Software — We'll show you why conventional approaches to product development break down when building AI and what to do instead. / Hamel Husain, YouTube (59 minute video)

"If your mission is to make customers' lives better and easier every day, and you believe every customer experience will be reinvented by AI, you're going to invest deeply and broadly in AI," Jassy wrote in his letter to shareholders, an annual rite of passage for the top boss at the Seattle retailer.

  • and

The company removed all references to "diversity, equity and inclusion" in its proxy statement, which had numbered 21 in last year's report.

 

šŸŽ“ FOR EDUCATORS

Educators can select a Google Drive file or manually input text, select which skills they want students to demonstrate, then generate questions to assess grade-specific comprehension and critical reasoning skills of students. Once generated, educators can then export the questions to a Google Form or Google Doc to assign to students.

  • this will (should?) change the value proposition of OER

While some bots simply don’t submit classwork and hope they can skate by, they also frequently use AI programs to generate classwork that they then submit. Determining whether a student is a bot can be a confusing task.

One of the most interesting trends we have observed is how the introduction of AI tools and the uncertainty they bring can sometimes test the limits of instructors’ stated dedication to co-creation, leading to the structure we call faux-creation. In the AI faux-created class, instructors enact many components of co-creation. They may give students opportunities to share input on course content, develop personalized learning outcomes or objectives, and design different ways for them to express their learning through flexible assignment structures. … However, they limit student co-creation around AI.

  • and

We have developed a four-stage pathway of improving engagement that involves increasing levels of trust and responsibility in students concerning AI in the learning environment: 1. Entering; 2. Exploring; 3. Evaluating; and 4. Integrating.

 

šŸ“Š FOR TECHNOLOGISTS

How AI will Disrupt BI As We Know It / Tristan Handy, The Analytics Engineering Roundup, Substack archive (16 minute read)

1. AI is going to be meaningfully better at exploratory data analysis than any BI tool.
2. If you take away EDA (ed: ā€œexploratory data analysisā€) from BI, the ā€˜conveyor belt’ model breaks down. And the conveyor belt model is the primary reason you use your current BI tool.
3. It is not yet clear how the BI ecosystem will adapt to this new reality.

That’s it. That’s my entire argument. Let’s see if it holds up.

As coding itself becomes commoditized, developer roles connect more directly to business strategy.

ā€œDevelopers are taking supervisory roles, guiding agents doing work on their behalf,ā€ Govindarajan explained. ā€œBut they remain responsible for what gets deployed. The buck still stops with them.ā€

This elevation places developers closer to decision-makers and further from implementation details—a promotion rather than an elimination.

How I reduced 90% errors for my Cursor (+ any other AI IDE) / AI Jason, YouTube (15 minute video)

  • use a PRD for task management

AI code suggestions sabotage software supply chain — Hallucinated package names fuel ā€˜slopsquatting’ / The Register (8 minute read)

Larson said that there are many reasons a developer might attempt to install a package that doesn’t exist, including mistyping the package name, incorrectly installing internal packages without checking to see whether those names already exist in a public index (dependency confusion), differences in the package name and the module name, and so on.

 

šŸŽ‰ FOR FUN

The Vegan Couple Accused of Killing for the Zizian ā€˜Cult’ / New York Magazine, archive (38 minute read)

And they believed that if those who worked on AI were meat-eaters, that super-intelligent AI would use the meat industry as a blueprint for how society worked: enslaving and torturing humanity for all eternity in factory farms spread across the galaxy. To prevent this, they would have to work to turn everyone, and especially those who worked on AI, vegan.

I Tested The AI That Calls Your Elderly Parents If You Can’t Be Bothered — inTouch says on its website ā€œBusy life? You can’t call your parent every day—but we can.ā€ My own mum said she would feel terrible if her child used it. / 404 Media (registration required) (2 minute read)

  • A.i. can raise your kids, call your dad, do your job … we’re not too far off from getting plugged into the power plant from The Matrix, are we?

DolphinGemma: How AI can decipher dolphin communication — DolphinGemma, a large language model developed by Google, is helping scientists study how dolphins communicate — and hopefully find out what they're saying, too. / Google blog (7 minute read)

  • they chose to invest time and money to get A.i. to speak ā€œdolphinā€ over Navajo or any other indigenous language

Users on X quickly discovered that o3, in particular, is quite good at deducing cities, landmarks, and even restaurants and bars from subtle visual clues.

  • you’ll never guess what Jesus’s favorite movie is

 

🧿 AI-ADJACENT

But what if I really want a faster horse? / Rakhim, Exotext (3 minute read)

Overall, consistency, user control, and actual UX innovation are in decline. Everything is converging on TikTok—which is basically TV with infinite channels.

 

ā‹„