weekend ai reads for 2025-02-14

This mindset tends to be good for #Capitalism, but betrays not only a lack of understanding of why people make music, but also a profound lack of imagination regarding how we could, or would want to live with our technologies in our lives. I, for one, would go as far as to say using generative AI for creative expression in this manner (“describe what you have in mind and AI will create it for you”) amounts to the least imaginative of use of AI that I can imagine.

But a voice is not just a sound. And I’d like to think that no matter how much an A.I. version of Moe or Snake or Chief Wiggum will sound like my voice, something will still be missing — the humanness. There’s so much of who I am that goes into creating a voice. How can the computer conjure all that?

“A.I. provided a way also to achieve this without intruding on real lives or placing real Costa Rican faces that people of the community might recognize,” he said. “Since the pegamachos culture remains hidden, these A.I. images serve as a mimicry of photography, a fiction, and a medium through which I can imagine and construct an imagined parallel history.”

Sauter Morera said the use of A.I. allows him to pose hypotheses that he can then answer through the use of the technology, such as whether pegamachos would have expressed themselves more freely if Costa Rican society had been different at the time.

“Would cowboy culture have embraced latex?” he wondered. “These speculative questions are at the core of my work.”

Quick reminder: no one is saying AI will make great art; it can make mediocre content, which the attention economy runs on now. The entirety of this post has (hopefully) shown how social media mediocrity has loomed over greatness, which has cowered in the corner, patiently looking for rocks to bang together.

You cannot be great without the greatness of others.

Good tools let the user choose when to switch between implementation and evaluation. When I work with a chatbot, I’m forced to frequently switch between the two modes. I ask a question (implement) and then I read a response (evaluate).

Overall, these workers self-reported that the more confidence they had in AI doing the task, the more they observed “their perceived enaction of critical thinking.” When users had less confidence in the AI’s output, they used more critical thinking and had more confidence in their ability to evaluate and improve the quality of the AI’s output and mitigate the consequences of AI responses.

The good news: ChatGPT appropriately selected and accurately summarized a set of recent court rulings, all of which exist. The so-so news: it missed some broader points that a competent human expert might acknowledge. The bad news: it ignored a full year’s worth of legal decisions, which, unfortunately, happened to upend the status of the law.

Aidan Toner-Rodgers of MIT, for instance, found that using an AI tool to assist with materials discovery nearly doubled the productivity of top researchers, while having no measurable impact on the bottom third. The software allowed researchers to specify desired features, then generate candidate materials predicted to possess these properties. Elite scientists, armed with plenty of subject expertise, could identify promising suggestions and discard poor ones. Less effective researchers, by contrast, struggled to filter useful outputs from irrelevant ones.

This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental models of how to think about their “psychology”, and how to get the best use them in practical applications. I have one “Intro to LLMs” video already from ~year ago, but that is just a re-recording of a random talk, so I wanted to loop around and do a lot more comprehensive version.

ROOST develops, maintains, and distributes open source building blocks to safeguard global users and communities. Backed by dedicated technical teams and leading experts, ROOST meets organizations where they are and provides hands-on support at every stage of their safety journey.

It is possible to develop a model with the best intentions, and for that model to be misused later on

Companies, look at the roadmap you have carefully developed and consider throwing out parts that no longer make sense. Start motions towards up-skilling how your employees think.

Five questions for every CEO

1. Am I moving fast enough?

2. How might AI change the future of my industry?

3. How can AI strengthen our competitive advantage?

4. How is this different, and how do I enable the tech foundation?

5. How should I lead the organization on this journey?

AI use is more prevalent for tasks associated with mid-to-high wage occupations like computer programmers and data scientists, but is lower for both the lowest- and highest-paid roles. This likely reflects both the limits of current AI capabilities, as well as practical barriers to using the technology.

Key policy measures include the development of guidelines for ethical GAI use, the design of authentic assessments to mitigate misuse, and the provision of training programs for faculty and students to foster GAI literacy. Despite these efforts, gaps remain in comprehensive policy frameworks, particularly in addressing data privacy concerns and ensuring equitable access to GAI tools. The study underscores the importance of clear communication channels, stakeholder collaboration, and ongoing evaluation to support effective GAI adoption. These insights provide actionable insights for policymakers to craft inclusive, transparent, and adaptive strategies for integrating GAI into higher education.

Workspace for Nonprofits users accessing the Gemini app now have enterprise-grade data and privacy protections, which means chats and uploaded files will not be reviewed by human reviewers or otherwise used to improve generative AI models. These same protections are available on NotebookLM for Workspace for Nonprofits.

Separated out like this, content and activities can be mixed and matched in both directions. For example, a learner can use different activities (like retrieval practice, reciprocal teaching, or worked examples) to study a single topic (like supply). Similarly, they could reuse a single activity across multiple topics (sometimes even across multiple courses).

Inconsistent information retrieval, chunking, and integration across multimodal solutions: Retrieval solutions and multimodal approaches can introduce data integration and engineering challenges, which can be resolved by setting up processes for retrieval augmented generation (RAG) integration with human oversight and improving chunking and advanced retrieval methods at all integration points.

In particular, we should prepare for the risk that AIs could be misaligned — that they might pursue goals that no one gave them and harm people in the process. We think that ML research today can help to clarify and mitigate the likelihood of this failure mode.

Canva says its “Magic Design” feature helps users effortlessly make beautiful designs. Their audience doesn’t need “AI-powered productivity.” They need the result of that AI-powered productivity. They need “beautiful marketing templates ready in 10 seconds.”

Scientists have spent thousands of hours sitting in front of stalls and cages observing the faces of animals in these painful or stressful situations, then comparing them against animals who were most likely pain- or stress-free. As a result, they’ve developed “grimace scales” for a variety of species, which provide a measure of how much pain or stress an animal is experiencing based on the movements of its facial muscles.

Writing in the MIT Technology Review, researchers Robert Mahari and Pat Pataranutaporn warned that sophisticated chatbots and other non-human agents posed new risks to human beings, a kind of artificial “addictive intelligence” that takes advantage of what we know about human behavior. “The allure of AI lies in its ability to identify our desires and serve them up to us whenever and however we wish,” they note. “AI has no preferences or personality of its own, instead reflecting whatever users believe it to be,” what researchers call “sycophancy.”

At the center of this book are five principles. These aren’t aspirational—they’re lessons we've learned through experience. But they’re also living ideas: they have tensions within them, and there are places where they don’t always fit.