weekend ai reads for 2025-02-28

1. China, India, and Southeast Asia are the ones to watch.

2. Data centers are a somewhat minor factor globally, but they can’t be counted out.

3. What this all means for climate change is complicated.

The average Google search uses 0.3 Watt-hours (Wh) of energy. The average ChatGPT question uses 3 Wh, so if you choose to use ChatGPT over Google, you are using an additional 2.7 Wh of energy.

How concerned should you be about spending 2.7 Wh? 2.7 Wh is enough to

- Stream a video for 2 minutes

- Watch an LED TV (no sound) for 3 minutes

- Upload 30 photos to social media

- Drive a sedan at a consistent speed for 15 feet

- Leave your digital clock on for 3 hours

- Run a space heater for 2.5 seconds

- Print half a page of a physical book

Stakeholders can then use these models to understand and plan for specific what-if scenarios far beyond what could be achieved with existing data alone. For example, generated data can predict the potential load on the grid if an additional 1,000 households were to adopt solar technologies, how that load might change throughout the day, and similar contingencies vital to future planning.

I don’t think I have ever seen so much technical debt being created in such a short period of time during my 35-year career in technology.

This shift in meaning – from machine learning to artificial intelligence – creates a kind of linguistic sleight of hand, where the promise of true in­tel­li­gence – which would be AGI – is implicitly attached to current AI systems, even though they are fundamentally incapable of becoming that. This rebranding leads to a lot of mis­under­standings and over­hyped expect­ations.

But he’s also concerned about the potential for AI to “hallucinate” quotes based on its training data or to extrapolate what was said into something that wasn’t — and about what happens to information after the meeting.

No. It’s a machine, why would I be polite?: 13%

OpenAI offers two types of models: reasoning models (o1 and o3-mini, for example) and GPT models (like GPT-4o). These model families behave differently.

This guide covers:

- The difference between our reasoning and non-reasoning GPT models

- When to use our reasoning models

- How to prompt reasoning models effectively

Lean into the chat aspect when you’re learning a new topic. If you use LLMs like a better Google search, you’re only capturing a fraction of their value. My best experiences have come when asking followup questions - just straightforward, down-the-line, naive questions, such as “so you’re saying that X is true” or “does this have consequence Y”?

Google Colab, short for Colaboratory, is a cloud-based Jupyter notebook environment that runs entirely in your browser. It’s a free service provided by Google that allows you to write and execute Python code, create detailed documentation, and share your work with others seamlessly. Think of it as a Google Docs for programming – but with far more powerful features under the hood.

Indeed, organizations that successfully deploy AI at scale devote the bulk of their effort to process and people. We call it the 10/20/70 rule: allocate 10% of the effort to data science capabilities for developing and implementing algorithms, 20% to a modernized and scalable tech stack, and 70% to effective processes supported by talent and robust change management. That crucial 70% is often overlooked or undercooked.

But here’s what caught my attention: while traditional opportunities decreased, something interesting emerged. The study found 903 new job listings specifically requiring ChatGPT expertise. And 82% of these were in software and web development.

Read the 2025 AI Decision Brief to learn from chief AI officers and Microsoft executives as they share guidance and strategic perspectives for leaders on the state of generative AI in the current business landscape.

The tool also lets you choose what type of pedagogy strategies you’d like it to embrace, for instance, active learning or the Socratic method. My experience with LearnLM bore this out. It was easier to get the model to do what I wanted, and instead of being frustrated by the answers it was generating, as frequently happens when I've tested other AI models, I was intrigued.

As cases of chatbots hallucinating or dispensing dangerous advice have made headlines, schools are wary of steering students to AI-only solutions. Sonar says Sonny’s selling point is that humans with backgrounds in psychology, social work and crisis-line support are always in the mix, reviewing the chats and taking cues from AI to inform their own replies to students.

Google forces companies like Chegg to “supply our proprietary content in order to be included in Google’s search function,” said Schultz, adding that the search company uses its monopoly power, “reaping the financial benefits of Chegg’s content without having to spend a dime.”

One standout example comes from China — and it’s not DeepSeek. There, an AI-powered tutoring platform called Squirrel AI has scaled personalized learning to an unprecedented level. Across thousands of after-school centers in China, students use Squirrel AI for one-on-one tutoring. The system adapts each lesson to individual needs by analyzing a vast database of student interactions. According to the World Economic Forum, Squirrel AI “currently serves more than 24 million students” and has collected over 10 billion data points on learning behaviors to continuously improve its teaching methods.

Notably, Code Assist for individuals offers 180,000 code completions a month, which is 90 times the usage cap of the free GitHub Copilot plan (2,000 code completions a month). Code Assist for individuals also comes with 240 chat requests a day, close to 5 times the number of requests the free GitHub Copilot plan offers.

Before this A.I. boom, start-ups generally burned $1 million to get to $1 million in revenue, Mr. Jain said. Now getting to $1 million in revenue costs one-fifth as much and could eventually drop to one-tenth, according to an analysis of 200 start-ups conducted by Afore.

For some affected companies, Copilot could be prompted to return confidential GitHub archives that contain intellectual property, sensitive corporate data, access keys, and tokens, the company said.

It’s perhaps most interesting that the future robots felt the need to take a human-like form. Is that implying that the very form of human beings is important to sentient intelligence? That's sort of a key to the whole film, in a way.

In games where players are assigned information hidden from their opponents (for example, in card games like bridge or poker, where others cannot see a player’s hand), designers must decide whether to give the AI the ability to memorise play so far and to count the pack perfectly, or else to act in a sloppier—and more humanlike—manner.

As a result, more and more, graphic design has begun to resemble collage: a patchwork of AI-generated textures, pre-made renderings, repurposed signage, and public domain imagery. Rather than referencing the past, designers are stripping it for parts to be mixed and matched – treating design history like a visual junkyard (or maybe a reanimated corpse).