weekend ai reads for 2025-02-21

Many large companies do not tolerate AI use by candidates. An April survey from Resume Genius found that AI-generated resumes were the biggest red flag for 625 U.S. hiring managers.

But that doesn't mean companies oppose using it to make hiring decisions — over 97% of Fortune 500 companies use AI software to filter candidates.

One of the things we can do, however, is change the nature of the interviews themselves. Coding interviews today are quite basic, anywhere from FizzBuzz, to building a calculator. With AI assistants, we could expand this 10x and have people build complete applications. I think a single, longer interview (2 hours) that mixes architecture and coding will probably be the way to go.

I am what happens when you try to carve god out of the wood of your own hunger.

Deep Research has the same problem as every other generative AI product. These models don't know anything, and thus everything they do — even “reading” and “browsing” the web — is limited by their training data and probabilistic models that can say “this is an article about a subject” and posit their relevance, but not truly understand their contents.

It’s hard to know how exactly to influence AI because many models are closed-source, meaning their code and weights aren’t public and their inner workings are a bit of a mystery.

In a study of nearly 600 people, published in Science on 20 February, the tool identified whether participants were healthy or had COVID-19, type 1 diabetes, HIV or the autoimmune disease lupus, as well as whether they had recently received a flu vaccine.

To cut through the noise, TechCrunch has compiled an overview of the most advanced AI models released since 2024, with details on how to use them and what they’re best for. We’ll keep this list updated with the latest launches, too.

“Give me 3 options”: Whenever I’m generating text that will be used in a document or email, I always ask for multiple options. This allows me to either pick the best one or, more often, combine the best parts of each to create something that feels more natural and human. I don’t trust any model to one-shot human-sounding text.

It covers alignment problems we can expect as AI capabilities advance, and our current approach to these problems (on technical and governance levels). If you would like to learn more about AGI safety but have only an hour to spare, this course is for you!

Is your company positioned to take advantage of access to a country of geniuses in a datacenter?

Is your company positioned to compete with the company that does take advantage of access to a country of geniuses in a datacenter?

API pricing currently faces a huge problem since it turns LLMs into commodities (raw materials used by other products for value adds), opening them to price wars. Since commodities are very tied to price, this leads to massive price wars and low margins for everyone involved. This is why estimates say that LLM API margins have been dropping-

Focusing on tools over processes is a red flag and the biggest mistake I see executives make when it comes to AI.

What’s really going on with campus-wide AI adoption is a mix of virtue signaling and panic purchasing. Universities aren’t paying for AI—they’re paying for the illusion of control.

Respondents from smaller institutions are remarkably similar to respondents from larger institutions in their personal use of AI tools, their motivations for institutional use of AI, and their expectations and optimism about the future of AI. Respondents from small and larger institutions differ notably, however, in the resources, capabilities, and practices they're able to marshal for AI adoption.

Our aim is not to teach students the mechanics of how large language models work, nor even the best ways of using them in various technical capacities.

We view this as a course in the humanities, because it is a course about what it means to be human in a world where LLMs are becoming ubiquitous, and it is a course about how to live and thrive in such a world. This is not a how-to course for using generative AI. It's a when-to course, and perhaps more importantly a why-not-to course.

To do that we need to have some experience ourselves, but that also gives us an opportunity to demonstrate to our students that learning is indeed a lifelong pursuit. This is a great opportunity to engage with our students in what is an important and complex problem.

Lack of AI support prevents a technology from gaining the required critical adoption mass, which in turn prevents a technology from entering use and having material made for it, which in turn starves the model of training data, which in turn disincentivises selecting that technology, and so on and so forth.

Opinions on this risk varied: some participants believed that combining high-quality synthetic data with real data could sustain model development without severe degradation, while others viewed this distribution shift as a slippery slope that resulted in compounding errors that threatened the integrity of models over time. The lack of consensus around this topic was described as a challenge to both understanding the problem and formulating appropriate solutions.

[A software developer’s] task is moving from writing code to being a context manager.

Since the coefficients of a logistic regression model provide a natural ranking for its features, we leverage these coefficients to highlight important phrases in the classification task. Figure 6 presents the top 10 phrases with the largest logistic regression coefficients for each of the five chat API models. Notably, these phrases often serve as transitions or emphasis in sentences.