May 14, 2025
Hi everyone!
If you’re new here, welcome! This is an extra-long issue because it’s been a couple of months since I sent one. This has been a super-busy spring for me.
In this issue we’ll look at:
Claude for education (now used at Northeastern University and a few other universities).
Google AI is free for college students in the U.S. through spring finals 2026.
Interesting thoughts: What is Scholarship in the Age of Deep Research AI? - Steve Covello.
Dividing up different types of skepticism about AI in a useful way: The Misunderstood Role of AI Skeptics - Alberto Romero
How I’ve been testing Mike Caulfield’s SIFT Toolbox prompt for fact-checking. (see below under My Offerings)
You can watch my video about image generation with 4o, in one of the videos I made for my AI Literacy course. Start at 3 min, 12 secs to skip to the part about 4o images.
And the usual tips, accessibility stories, what’s happening in education and libraries, a few “just for fun” items, thought-provoking stories, and stories about the future.
Enjoy!
Foundation Models
OpenAI
OpenAI scuttles for-profit transformation - Megan Morrone, Axios, May 5, 2025.
”A nonprofit will remain in control of OpenAI and its technology, according to a letter to employees from Sam Altman on Monday.”
ChatGPT search - OpenAI
They’ve improved the interface for how sources show up when it uses web search.
OpenAI overrode concerns of expert testers to release sycophantic GPT-4o - Carl Franzen, VentureBeat, May 2, 2025.
”Speaking perhaps more theoretically, for myself, it also indicates why expertise is so important — and specifically, expertise in fields beyond and outside of the one you’re optimizing for (in this case, machine learning and AI). It’s the diversity of expertise that allows us as a species to achieve new advances that benefit our kind. One, say STEM, shouldn’t necessarily be held above the others in the humanities or arts.”
Addendum to GPT-4o System Card: 4o image generation - OpenAI
It’s worth reading the system card to learn details of their new image creation model, including which guardrails have changed.
ChatGPT — Release Notes - OpenAI
More updates, including enhanced memory rolling out to all Plus and Pro users (including the EU). These features are OFF by default and must be enabled in Settings > Personalization > Reference Chat History.
Anthropic
Introducing Claude for Education - Anthropic, April 2, 2025.
”Today we're launching Claude for Education, a specialized version of Claude tailored for higher education institutions.” “Full campus access agreements with Northeastern University, London School of Economics and Political Science (LSE), and Champlain College, making Claude available to all students.”
Anthropic launches Claude web search API, betting on the future of post-Google information access - Michael Nuñez, VentureBeat, May 7, 2025.
Now in addition to web search in the paid versions of Claude, there’s web search in the API. I hope web search will come to the free version of Claude soon.
Claude takes research to new places - Anthropic, April 15, 2025.
Claude now integrates with Gmail and Calendar, in addition to Google Docs (in beta to all users in profile settings). Research is also available (like Deep Research), but only in more expensive accounts: Max, Team, and Enterprise plans in the United States, Japan, and Brazil.
Anthropic wins early round in music publishers' AI copyright case - Blake Brittain, Reuters, March 26, 2025.
This is about music lyrics. “Lee rejected the publishers' argument that Anthropic's use of their lyrics caused them irreparable harm by diminishing their licensing market.”
Apple
Apple is looking at adding Perplexity and other AI search engines to Safari - Emma Roth and Lauren Feiner, The Verge, May 7, 2025.
Apple and Anthropic reportedly partner to build an AI coding platform - Maxwell Zeff, TechCrunch, May 2, 2025.
Microsoft
Microsoft launches Phi-4-Reasoning-Plus, a small, powerful, open weights reasoning model - Carl Franzen, VentureBeat, May 1, 2025.
”During the supervised fine-tuning stage, the model was trained using a curated blend of synthetic chain-of-thought reasoning traces and filtered high-quality prompts. A key innovation in the training approach was the use of structured reasoning outputs marked with special <think> and </think> tokens. These guide the model to separate its intermediate reasoning steps from the final answer, promoting both transparency and coherence in long-form problem solving.”
Google
Google’s NotebookLM Android and iOS apps are available for preorder - Aisha Malik, TechCrunch, May 2, 2025.
”Google’s NotebookLM Android and iOS apps are expected to launch on May 20.”
The latest AI news we announced in April - Google, May 6, 2025.
”We made the best of Google AI free for college students in the U.S. through spring finals 2026. College students in the U.S. are now eligible to get tools like Gemini Advanced, NotebookLM Plus and 2 TB of storage free of charge for this and next school year. This means that students now have more tools to study, understand complex concepts and research new ideas.”
They also made Deep Research available on Gemini 2.5 Pro Experimental (for Gemini Advanced subscribers).
DeepSeek
China's DeepSeek releases AI model upgrade, intensifies rivalry with OpenAI - Reuters, March 25, 2025.
”The latest model demonstrates significant improvements in areas such as reasoning and coding capabilities compared to its predecessor.”
How DeepSeek erased Silicon Valley’s AI lead and wiped $1 trillion from U.S. markets - Nicholas Gordon, Fortune, March 30, 2025.
”But DeepSeek is just one player in a vibrant Chinese AI sector to which many U.S. CEOs have been oblivious. Big Tech companies like Alibaba and ByteDance (the parent of TikTok) are releasing AI models that have beaten Western products on reasoning benchmarks. And a new wave of smaller “AI dragons” are putting China’s cheap and efficient AI to work in the real world, through mobile apps, AI agents, and robots.”
Other models
Qwen swings for a double with 2.5-Omni-3B model that runs on consumer PCs, laptops - Carl Franzen, VentureBeat, April 30, 2025.
”While smaller in size, the 3B version retains over 90% of the larger model’s multimodal performance and delivers real-time generation in both text and natural-sounding speech.”
Meta’s first dedicated AI app is here with Llama 4 - Carl Franzen, VentureBeat, April 29, 2025.
”Facebook parent company Meta Platforms, Inc. has officially launched its own, free standalone Meta AI app, a move aimed at delivering a more personal and integrated AI experience across mobile devices, the web, and Ray-Ban Meta smart glasses. The app is available on iOS through the Apple App Store and on the web — with no mention of when an Android version could come.”
Deep Research
What is Scholarship in the Age of Deep Research AI? - Steve Covello, March 6, 2025.
“The metaphor that comes to mind to describe this kind of scholar is a symphony orchestra conductor. The conductor is a second-order scholar of first-order work characterized by the following competencies…”
Deep Research with AI: 9 Ways to Get Started - Jeremy Caplan, Wonder Tools, May 2, 2025.
Some easy and interesting ways to get started testing Deep Research.
These experts were stunned by OpenAI Deep Research - Timothy B. Lee, Understanding AI, Feb. 25, 2025.
AI Tools for Literature Reviews Benchmarking Report - Nuance, April 2025.
They tested over 20 tools and evaluated the top performers across five key criteria, including citation quality, writing clarity, and cost-efficiency.
The latest AI news we announced in April - Google, May 6, 2025.
”We made Deep Research available on Gemini 2.5 Pro Experimental. Gemini Deep Research is our personal AI research assistant built for you. … So we made it available to Gemini Advanced subscribers.”
I’ve been experimenting with Deep Research on different topics. Here’s a report I made using Deep Research on ChatGPT:
Overestimating the Energy Consumption of Digital Technologies: A Historical Perspective - PDF. May 6, 2025.
From the conclusion: “Looking across these decades, a clear historical pattern emerges: each new wave of digital technology has initially been accompanied by overblown predictions of energy consumption, which gradually give way to a more measured reality. This pattern is not coincidence – it stems from common pitfalls in how we imagine technology’s impact.”
(And here’s a version I copied into a Google Doc, which includes my prompt at the bottom of the page).
Images, video, voices, and music
Images
GPT-4o Image Generation: A Complete Guide + 12 Prompt Examples - Valeriia Kuka, Learn Prompting, April 2, 2025.
”OpenAI has released GPT-4o Image Generation, which is probably the most advanced, natively multimodal image generator to date.”
(See also my video about image generation with 4o, starting at 3 min, 12 secs. This is one of the videos for my AI Literacy course).
Ghibli, Ghiblification, Copyright and Style - Justin Bonfiglio, Authors Alliance, May 8, 2025.
”While it might be tempting to want to protect the Ghibli style, we should remain mindful of the core purpose of copyright: to promote creativity for the public good (“The immediate effect of our copyright law is to secure a fair return for an `author's' creative labor. But the ultimate aim is, by this incentive, to stimulate artistic creativity for the general public good.”) Overzealous protection of style risks enclosing the common building blocks creators rely on, preventing one creator from building on the style of another.”
Video
Awards Rules and Campaign Promotional Regulations Approved for 98th Oscars - Oscars Press Release, April 21, 2025.
”With regard to Generative Artificial Intelligence and other digital tools used in the making of the film, the tools neither help nor harm the chances of achieving a nomination. The Academy and each branch will judge the achievement, taking into account the degree to which a human was at the heart of the creative authorship when choosing which movie to award.”
Director James Cameron Is Open to Using AI to Reduce Filmmaking Costs - Jibin Joseph, PC Mag, April 10, 2025.
”James Cameron is open to the idea of using generative AI to reduce filmmaking costs. As Deadline reports, the development is a sharp turn from Cameron's earlier stance on the technology”….. “they have to figure out ways to cut costs in half—without laying off staff. ”
Natasha Lyonne Set to Make Feature Directorial Debut With AI Film — With Help From Jaron Lanier - Steven Zeitchik, The Hollywood Reporter, April 29 2025.
”Centered on a teenage girl who becomes unmoored by a hugely popular AR video game in a parallel present, the movie will blend traditional live-action and game elements. The latter will be created by Lanier as well as Lyonne and Marling. And the entire enterprise will draw on AI from Asteria partner Moonvalley via a model called “Marey,” which unlike systems from companies like Runway and OpenAI, is built only on data that has been copyright-cleared.”
Make great AI video right now using Gemini – it's quicker than using Sora - Graham Barlow, Tech Radar, May 7, 2025.
You can now generate 5 videos per day with Gemini Advanced, using their Veo 2 model.
Voice
Speech Transcription with parakeet-tdt-0.6b-v2 - demo on Hugging Face
Designed for high-quality English speech recognition. Open source. It’s fast and it works well for making time-stamped transcripts of audio files.
NotebookLM Audio Overviews are now available in over 50 languages - Google Labs blog, April 29, 2025.
”Audio Overviews are generated in your account’s preferred language. This update also introduces a new "Output Language" option in NotebookLM's settings; your Audio Overviews are always generated in the language you select here.”
Computer vision
Watching o3 guess a photo’s location is surreal, dystopian and wildly entertaining - Simon Willison, April 26, 2025.
”Watching OpenAI’s new o3 model guess where a photo was taken is one of those moments where decades of science fiction suddenly come to life.”
Music
Suno upgrades AI music generator with enhanced vocal capabilities, extended track length - Mandy Dalugdug, Music Business Worldwide, May 8, 2025.
”The v4.5 update enhances the range and emotional depth of AI-generated vocals, allowing users to “create everything from delicate, intimate performances to powerful deliveries with vibrato,” said Suno. Users will be able to create mashups combining different styles. …The v4.5 model is available exclusively to paid subscribers, with plans starting at $8 monthly for the Pro tier. Free users remain limited to the previous v4.0 model with a cap of 20 total songs, plus 10 daily creations using the older v3.5 model.”
Timbaland’s AI Reinvention: ‘God Presented This Tool to Me’ - Brian Hiatt, Rolling Stone, March 19, 2025.
”Timbaland uses one of them exclusively: a beta feature called “cover songs,” which allows you to upload beats or songs of your own creation, and then use Suno to generate infinite variations on them, in every imaginable genre.”
”He compares resistance from artists to early suspicion of Auto-Tuned vocals in the 2000s. “It was a big thing, to the point Jay-Z made ‘Death of Auto-Tune,'” he says with a laugh. “Come on now. T-Pain was the only one to step in. Same way I’m doing with this. T-Pain stood out there by himself.”
“It might give you something that you ain’t never thought of, and it becomes the biggest record of your life. Are you gonna criticize it then?… I never want to remove humans from what they do. I just want to inspire them to do more.”
Tips
3 Neat GPT-4o Image Tricks To Try - Daniel Nest, Why Try AI?, May 1, 2025.
AI for research: the ultimate guide to choosing the right tool - Amanda Heidt, Nature, April 7, 2025.
Search Google without the AI: &udm=14 and Does One Line Fix Google?
For another method, you can just add -a to the end of your Google query.OpenAI document explains when to use each ChatGPT model - Mayank Parmar, Bleeping Computer, May 4, 2025.
”OpenAI model names have been confusing, but the company is finally taking steps to make it easier for users to understand the different ChatGPT models.”ChatGPT’s Deep Research now has a PDF export button.
My offerings
Online course: AI Literacy for Library Workers - Nicole Hennig
This is the second time I’ve offered this 6-week asynchronous course, and we’re in week 5 right now.
I will offer it again in the fall (with some updated content), so you can sign up here to be notified if you’re interested in the fall session. Here’s some feedback from previous participants. I’m also thinking of creating a followup course for those who’ve completed the first one, or already have equivalent knowledge or experience. I’ll announce that on the same list (as well as in this newsletter).
If you’d like to see a sample video from the course, here’s one I made about generating images. It includes demos of the types of images you can make with the new version of ChatGPT 4o, so you can be familiar with what’s possible — and learn to be skeptical of any image you see these days. (The course also includes many hands-on activities).
Fact-checking bots
I’ve been testing Mike Caulfield’s SIFT Toolbox prompt for fact-checking. I created a custom “Gem” on Google Gemini. Gemini doesn’t make it possible to share a gem for others to use, but I can share the results of specific queries with the gem. Here the result of a question I asked about an image of a Studio Ghibli Cease and Desist Letter. I was suspicious, so I asked: “Is this a real letter about using Studio Ghibli's style to generate images?”
Look in the sidebar of this page for his full prompt that you can try with any model, without building a bot.
(His prompt is too long to turn into a custom GPT from OpenAI - they have a character limit of 8,000, including spaces, but Gemini doesn’t have this limit. I even tried using ChatGPT o3 to make his prompt shorter, but it lost some of the critical functionality).
Accessibility
This Startup Has Created AI-Powered Signing Avatars for the Deaf - Simon Hill, Wired, May 7, 2025.
”New technology from British startup Silence Speaks enables an AI-generated sign language avatar to effectively give the deaf and hard of hearing an interpreter in their pocket.”
If Apple gets apple intelligence right, the biggest beneficiary will be accessibility - Steven Aquino, Curb Cuts, March 20, 2025.
”Take Image Playgrounds, for instance. Most observers in the Apple community loathe the feature for being goofy and generally useless … What this perspective lacks is, of course, empathy for disabled people. Whatever you, able-bodied reader, may think of artificial intelligence and tools like Midjourney, for example, the reality is it’s extremely plausible the advent of Image Playgrounds gives an aspiring artist with disabilities—someone who may not be able to use an Apple Pencil on iPad Pro—a conduit through which to unleash their creativity and self-expression. This is not at all trivial, regardless of one’s philosophical views on art or their views on the quality of Image Playgrounds’ output. It’s perfectly okay for Image Playgrounds to not be your jam, but to sneer at it wholesale reeks of elitism and dishonesty.”
Recent Meta Ray-Bans Updates make smart glasses Even more alluring for accessibility - Steven Aquino, Curb Cuts, April 28, 2025.
”Take the Live Translation functionality, for example. It can be way more accessible to hear aural translations of language in a hands-free way; this could be important for people with limited motor skills who might have trouble holding their iPhone to, say, use Apple’s Translate app to make the conversion(s).”
Using AI to build Wordpress sites shows How AI makes web work more accessible - Steven Aquino, Curb Cuts, April 22, 2025.
”A person with disabilities who may have fine-motor delays, for instance, that limits their use of a mouse and keyboard very well could find the WordPress site builder a more accessible avenue to build their website(s).”
What’s happening in education
Addressing the Transactional Model of School - John Warner, May 11, 2025.
A response to the article in New York Magazine, “Everyone is Cheating Their Way Through School.”
Warner says: “But the presentation of the magnitude and intensity of the problem is unfortunately exaggerated in a way that I fear has the potential to induce a moral panic response which will not lead to a productive discussion about what to do. I hope everyone knows that not “everyone” is cheating their way through college.”
What the Alarm Misses: Why AI Isn’t Ruining Student Writing—But Shallow Reporting Might Be - Dr. Jeanne Beatrix Law, May 8, 2025.
”I read the New York Magazine Intelligencer piece1 “Rampant AI Cheating Is Ruining Education Alarmingly Fast” the way I imagine many educators did—tight-jawed, half-nodding, and wholly unsettled. Not because the piece got it all wrong. In fact, it got one thing deeply right: faculty are concerned, tired, and unsure how to respond to what feels like a seismic shift in student learning.
But I worry that headlines like these do something far more damaging than ChatGPT ever could. They flatten students into cheaters, flatten teachers into victims, and flatten a nuanced, discipline-rich, and rapidly evolving reality into a single narrative of collapse.”
Can an AI Partner Empower Learners to Ask Critical Questions? - Pratyusha Maiti, Ashok Goel, Proceedings of the 30th International Conference on Intelligent User Interfaces, March 24, 2025.
”Jill Watson is an LLM-powered conversational AI partner integrated with instructor-provided courseware, offering learners contextually relevant and immediately applicable support.” …”These findings demonstrate that Jill empowers learners to engage in critical questioning, thereby enhancing their educational experience by promoting depth, relevance, and application of course concepts.”
AI is transforming education into a collaborative interaction between humans and machines - B. Mairéad Pratschke, London School of Economics Impact Blog, May 7, 2025.
”While education is currently focused on the use of machine intelligence in the form of AI assistants, chatbots and tutors, in industry, the focus is on autonomous agents to scale service and operations and humanoid robots to perform formerly human labour. This increase in capabilities means that humans will increasingly interact with intelligent machines as presences in their personal and professional lives.” …”a future that educators need to take seriously.”
The effect of ChatGPT on students’ learning performance, learning perception, and higher-order thinking: insights from a meta-analysis - Jin Wang & Wenxiang Fan, Humanities and Social Sciences Communications, May 6, 2025.
”The results indicate that ChatGPT has a large positive impact on improving learning performance (g = 0.867) and a moderately positive impact on enhancing learning perception (g = 0.456) and fostering higher-order thinking (g = 0.457).”
AI as a conversation partner and coach - Mamie Rheingold, ChatGPT for Education, May 5, 2025.
”How Julia Minson of the Harvard Kennedy School trained a GPT on her research to help students master constructive disagreement.”
Thinking with AI: Machine Learning the Humanities - Open access book from Open Humanities Press, edited by Hannes Bajohr. 2025.
”This collected volume explores a novel approach to the intersection of artificial intelligence and the humanities, proposing that instead of merely writing about AI, scholars should think with AI. Rather than treating AI as an external subject of study, the essays explore how concepts from artificial intelligence, machine learning, and data science can provide ways to rethink core humanistic questions of meaning, representation, and culture.” (Free PDF)
Making ChatGPT Work for Me - Samantha Keppler, Wichinpong Park Sinchaisri, Clare Snyder, Proceedings of the ACM on Human-Computer Interaction, may 2, 2025.
”We focus our study on US K12 public school teachers who regularly design and complete text-generation tasks such as creating quizzes, slide decks, word problems, reading passages, lesson plans, classroom activities, and projects.” …”Our study contributes new data and knowledge about how teachers are coming to understand whether and how to integrate generative AI into their teaching preparation routines.”
The Evolution of AI Literacy: navigating the agentic shift - Carlo Iacono, May 2, 2025.
”The emergence of sophisticated multimodal and agentic AI capabilities renders baseline AI literacy, focused on simple prompting and text evaluation, demonstrably insufficient. Interacting effectively and responsibly with systems that can autonomously plan, act, use tools and process information across diverse formats demands a more advanced and critical form of literacy.”
Integrating AI Literacy into the Curricula: A New Cohort Project Gets Underway - Ruby MacDougall, et al., Ithaka S+R, April 9, 2025.
”The project will be conducted in partnership with librarians and educators at 45 colleges and universities with deep commitments to promoting AI literacy as a core learning outcome. Together, we will conduct institution-specific and landscape level research on faculty and student practices, priorities, and needs around AI Literacy, and develop actionable pathways to integrating AI literacy into the curricula.”
Making AI Generative for Higher Education - Claire Baytas, Dylan Ruediger, Ithaka S+R, May 1, 2025.
”In the time since these interviews were conducted, the number of individuals within higher education who are highly familiar with AI has only increased. New technologies are emerging— particularly in the realm of agentic AI, a term very few of our interviewees in spring 2024 referenced. Now, the crucial task is managing the transition from the phase of exploration to responsible, well-informed usage of this technology.”
What’s happening in libraries
Carla Hayden removed as Librarian of Congress and what this could mean for authors - Dave Hansen, Authors Alliance, May 9, 2025.
(Not about AI, but an important read).
Eight Hypotheses Why Librarians Don’t Like Retrieval Augmented Generation (RAG) - Frauke Birkhoff, The Scholarly Kitchen, May 8, 2025.
”There is reason to believe that in the next few years, discovery layers as we know them could be displaced by RAG-based search interfaces. Librarians currently working on discovery layers should get acquainted with these tools and ask themselves and their institutions how they should implement them for their communities, be it by using a commercially available tool or by investing in the development of an open-source version. Crucial to the success of these efforts, however, is the acceptance of the librarians implementing, maintaining and using these tools.”
Sensemaking with SIFT Toolbox - A research project by Mike Caulfield. (Caulfield is the creator of the SIFT method for information literacy, a method used by many librarians who teach info lit).
”SIFT Toolbox is a lengthy instruction prompt that outperforms unmodified LLMs in multiple dimensions. You paste it in at the beginning of a chat session (or add it to project instructions) to make the LLM act differently.
With the prompt in place, your LLM will come to better conclusions, hallucinate less, and source conflicting perspectives more systematically. It also models an approach that is less chatbot, and more research assistant in a way that is appropriate for student researchers, who can use it to aid research while coming to their own conclusions.”
Frenemies & That's OK: Useful Tensions & Possibilities Between Libraries and AI - Lance Eaton, May 5, 2025.
(from a talk for the University of Rhode Island’s Graduate School of Library and Information Science Annual Gathering)
”But it’s not all challenges; there are some significant possibilities. One is enhanced research support. A tool like Google’s NotebookLM changes the game for how people can organize and validate research, more effectively than something like ChatGPT, in some cases.
Accessibility is another area of transformation. We’re seeing AI actively at work in live-captioning what I’m saying right now. These tools are making the world more legible for many people, not just those with visual or hearing disabilities, but also those with cognitive differences, or those who prefer different modalities of communication”
Avoiding misconceptions - testing "AI search tools" - What not to do & some research questions to consider - Aaron Tay, Musings About Librarianship, Dec. 27, 2024.
”Stop trying to measure "hallucination" of sources retrieved. I know that when ChatGPT first hit the scene, there was widespread knowledge of the fact that the references it created tended to be wrong or worse yet fictional and I saw more studies that I could count that showed this (here's one).
The problem is that many people don't seem to realize this issue applies only when LLM (Large Language Models) that do not search are used, and they do not apply to search engines even ones that are "AI".
Below shows search results shown by a typical "AI search" tool, Elicit.com. Of the Top 8 results shown below they are 100% "real" even if you think the results are retrieved and ranked by AI!”
Just for fun
LegoGPT creates Lego designs using AI and text inputs — tool now available for free to the public - Jowi Morales, Tom’s Hardware, May 9, 2025.
”The tool is available for free on GitHub, and you can pair this with a computer vision model or image processing AI. For example, you can take a photo of your available LEGO bricks and let the AI give you a multitude of unique options for building with what you already have.”LegoGPT is new, but it reminds me of another tool from 2023 called Brickit.
”Just scatter your bricks and take a photo. Brickit will show you hundreds of ideas for what to build with them, along with the exact location of each piece you’ll need.”Two AI agents on a phone call realize they’re both AI and switch to a superior audio signal ggwave - YouTube
Kelly Boesch made these beautifully patterned origami pieces and books in MidjourneyV7 and animated them using Pika Labs. The song was made using Suno.
Thought-provoking
New research & techniques
Training as we know it might end - Alexander Doria, Vintage Data
“Synthetic playground are presented as a scientific and experimental design: "We draw inspiration from the physical sciences, where idealized settings—such as friction-less planes or vacuum chambers—reveal first principles by stripping away confounding factors." I now believe this is the future paradigm of training. There is a natural correspondence between playground to assess model capacities, emulators used by the emerging agent systems and, most importantly, the kind of complex pipelines recently used by DeepSeek to push for the frontier of math solving models.”
Absolute Zero: Reinforced Self-play Reasoning with Zero Data - Andrew Zhao et al, Machine Learning, May 7, 2025.
”To address these concerns, we propose a new RLVR paradigm called Absolute Zero, in which a single model learns to propose tasks that maximize its own learning progress and improves reasoning by solving them, without relying on any external data. Under this paradigm, we introduce the Absolute Zero Reasoner (AZR), a system that self-evolves its training curriculum and reasoning ability by using a code executor to both validate proposed code reasoning tasks and verify answers, serving as an unified source of verifiable reward to guide open-ended yet grounded learning.” (See also New "Absolute Zero" Model Learns with NO DATA - Matthew Berman on YouTube).
The walled garden cracks: Nadella bets Microsoft’s Copilots—and Azure’s next act—on A2A/MCP interoperability - Louis Columbus, VentureBeat, May 8, 2025.
”Nadella’s endorsement of A2A and MCP protocols shows how far the Microsoft senior management team has agreed that an open protocol approach is the best direction for the company.”
These Startups Are Building Advanced AI Models Without Data Centers - Will Knight, Wired, April 30, 2025.
”Distributed approaches could make it possible for smaller companies and universities to build advanced AI by pooling disparate resources together. Or it could allow countries that lack conventional infrastructure to network together several data centers to build a more powerful model.”
The interoperability breakthrough: How MCP is becoming enterprise AI’s universal language - Emilia David, VentureBeat, May 13, 2025.
”Unlike APIs, organizations can configure their MCP servers with custom instructions laying out what agents can or cannot access. The server can “ask” an agent for its identity and determine if it can tap information on the MCP client side. Companies have more of a say on what outside agents can access on their end, giving MCP more directionality from the enterprise.”
MCP and the innovation paradox: Why open standards will save AI from itself - Noah Schwartz, VentureBeat, May 10, 2025.
”Nobody likes using fragmented tools. No user likes being locked into vendors. And no company wants to rewrite integrations every time they change models. You want freedom to use the best tools. MCP delivers.”
Guardian agents: New approach could reduce AI hallucinations to below 1% - Sean Michael Kerner, VentureBeat, May 13, 2025.
”The guardian agents are functionally software components that monitor and take protective actions within AI workflows. Instead of just applying rules inside of an LLM, the promise of guardian agents is to apply corrective measures in an agentic AI approach that improves workflows. Vectara’s approach makes surgical corrections while preserving the overall content and providing detailed explanations of what was changed and why.”
OpenAI’s $3B Windsurf move: the real reason behind its enterprise AI code push - Matt Marshall, VentureBeat, May 9, 2025.
”Specifically, the maneuver underscores two imperatives for OpenAI: first, the need to arm the vital developer ecosystem with superior coding capabilities, and second, to win the broader, more defining battle to become the primary interface for a future shaped by autonomous AI agents.”
Surveys
Trust, attitudes and use of artificial intelligence: A global study 2025 - University of Melbourne & KPMG International, 2025.
“There are notable differences between countries with advanced and emerging economies: People in emerging economics report greater trust, acceptance and adoption of AI, higher levels of AI literacy, and more realized benefits from AI.”
Copyright
The Copyright Office Report about Fair Use in AI & the Dismissal of the Register of Copyrights: A Drama in Three Parts - Dave Hansen, Authors Alliance, May 12, 2025.
”So we think the efforts to push Perlmutter out are a shame. What exactly motivated them is still an open question. As I mentioned already, the immediate reaction to her removal was that it must have been related to this AI report and the policy positions the Office has taken on AI under her leadership, particularly as some important AI cases are currently hearing, or have just heard, oral arguments directly addressing points raised by the US Copyright Office's report. That doesn’t seem like a crazy deduction, but there may be some other factors at play.”
Copyright and the AI Action Plan - Matthew Sag, March 20, 2025.
”Adverse outcomes in U.S. litigation will not stop the development of AI, they will simply push AI innovation overseas. The reason is straightforward: AI models, once trained, are easily portable. Companies seeking to avoid restrictive copyright rules could simply move their training operations to innovation-friendly jurisdictions like Singapore, Israel, or Japan, and then serve U.S. customers remotely, entirely free of domestic copyright concerns.”
GenAI: Copyight's Unacknowledged Offspring - Lance Eaton, May 12, 2025.
Interesting thoughts: “There is no merit in rejecting AI simply to re-legitimise the same inequities that led us to piracy in the first place.”
Climate
What's the carbon footprint of using ChatGPT? - Hannah Ritchie, Sustainability by Numbers, May 5, 2025.
”Very small compared to most of the other stuff you do.”
What's the carbon footprint of using ChatGPT? - Simon Willison, May 6, 2025.
”Hanah is Head of Research at Our World in Data, a Senior Researcher at the University of Oxford (bio) and maintains a prolific newsletter on energy and sustainability so she has a lot more credibility in this area than Andy or myself!”
Why using ChatGPT is not bad for the environment - a cheat sheet - Andy Masley, April 27, 2025.
”All the changes we can make in our personal consumption choices are nothing compared to what we can do if we contribute to making the energy grid green. The current AI debate feels like we’ve forgotten that lesson. After years of progress in addressing systemic issues over personal lifestyle changes, it’s as if everyone suddenly started obsessing over whether the digital clocks in our bedrooms use too much energy and began condemning them as a major problem. It’s sad to see the climate movement get distracted. We have gigantic problems and real enemies to deal with. ChatGPT isn’t one of them.”
Replies to criticisms of my posts on ChatGPT & the environment - Andy Masley, May 8, 2025.
”I want to make something clear about why I decided to post about this more, because a lot of people were implying I have some nefarious connection to the AI labs. It’s fun to imagine being evil or a part of some big conspiracy, but the reality’s pretty boring: I was motivated entirely by a deep sense of unease as a former physics teacher that so many people were speaking in such apocalyptic tones about such small amounts of energy.”
More from Simon Willison - April 29, 2025.
”Andy Masley has pulled together by far the most convincing rebuttal of this idea that I've seen anywhere.”
AI & Climate: Seeing the Forest, Not Just the Kilowatt-Hours - Tim Dasey, April 25, 2025.
”The idea that we should unduly restrict or slow down the exploration of AI's potential solely because of its energy footprint feels shortsighted. Climate change is perhaps the most complex, multifaceted challenge humanity has ever faced. Frankly, I don't see how we solve it without AI. We need AI to model the complexities, find novel solutions, understand behavior change, and optimize adaptation. Asking educators and students to view AI primarily as a climate problem risks blinding them to its potential as a critical climate solution. Let's have more nuanced conversations. Let's push for policies that harness AI for climate action. Let's teach students to analyze the entire system of AI's impact, weighing the costs against the benefits and the potential for innovation.”
From waste to solutions: How an Australian startup is revolutionizing excess renewables with AI innovation - AI for Good, LinkedIn, May 2, 2025.
”By deploying decentralized AI compute nodes in regions with surplus renewable energy and leveraging its AI application for intelligent workload timing, Project Ohm actively contributes to affordable and clean energy, industry innovation, and climate action. This pioneering approach turns a global challenge into a valuable resource, bringing advantages like fewer emissions, a more stable energy grid, lower computing expenses, and the creation of AI systems powered by clean energy, even in remote areas.”
Explanations
Two publishers and three authors fail to understand what “vibe coding” means - Simon Willison, May 1, 2025.
”I think there is a real need for a book on actual vibe coding: helping people who are not software developers—and who don’t want to become developers—learn how to use vibe coding techniques safely, effectively and responsibly to solve their problems.”
Artificial intelligence learns to reason - Melanie Mitchell, Science, March 20, 2025.
“There has been substantial debate in the AI community on whether LRMs are “genuinely reasoning” or “merely mimicking” the kinds of human reasoning that is in the pretraining or post-training data.”
AI skepticism vs AI optimism
The Misunderstood Role of AI Skeptics - Alberto Romero, The Algorithmic Bridge, May 1, 2025.
”And it’s absolutely crucial to figure out who belongs to which camp—because while the former can be easily ignored, their irrelevance often matching their ignorance of the topic they’ve chosen to object to, the latter are, if anything, the most valuable force we have to counter the fever for power that afflicts the AI enablers..” Sorts skeptics into categories and promotes “those who question the execution by the optimists but not their forward-looking mindset.” Like Melanie Mitchell, François Chollet, and Arvind Narayanan.
Inspiration
AI Isn’t Only a Tool—It’s a Whole New Storytelling Medium - Eliot Peper, April 28, 2025.
”Again, we couldn’t use deterministic logic to control probabilistic technology. Our methods needed to evolve to match our means. The solution required following the advice of an unusual sage: Stephen King. King doesn’t believe in plot. As he explains in his book On Writing, his stories flow naturally out of a specific character finding themselves in a particular situation.”
Three Questions: Josh Vermillion - Scott Dickensheets, Nevada Public Radio, April 16, 2025. (Follow Vermillion’s work on Instagram)
”I see a lot of recycled, unserious critiques of AI, but there is really room to help ourselves understand what’s innately human, what can’t be automated. Things like ethics and empathy, those qualities that are probably more important now than ever. What I would hate, though, is to unequivocally dismiss a technology that is really going to be transformative in the long run. I don’t think we can do that.”
When ChatGPT Broke an Entire Field: An Oral History - John Pavlus, Quanta Magazine, April 30, 2025.
A fascinating history.
Beneficial uses of AI
Yes, LLMs Can Be Better at Search Than Traditional Search - Mike Caulfield, May 10, 2025. (He’s the creator of the SIFT method for information literacy)
”It takes a specialized prompt, but LLMs can significantly outperform search.”
AI Is Not Your Friend - Mike Caulfield, The Atlantic, May 9, 2025.
”But the technology has evolved rapidly over the past year or so. Today’s systems can incorporate real-time search and use increasingly sophisticated methods for “grounding”—connecting AI outputs to specific, verifiable knowledge and sourced analysis. They can footnote and cite, pulling in sources and perspectives not just as an afterthought but as part of their exploratory process; links to outside articles are now a common feature. My own research in this space suggests that with proper prompting, these systems can begin to resemble something like Vannevar Bush’s idea of the memex.”
Google rolls out AI tools to protect Chrome users against scams - Aisha Malik, TechCrunch, May 8, 2025.
How The Ottawa Hospital uses AI ambient voice capture to reduce physician burnout by 70%, achieve 97% patient satisfaction - Taryn Plumb, VentureBeat, May 8, 2025.
”…the tool has reduced after-hours, charting and documentation work for “all categories of physicians.” This not only saves them time but helps reduce burnout because they have less tedious work to do.
The tool has also improved clinician cognitive load during visits: Instead of focusing on inputting patient details and navigating documents and forms, they are able to “engage differently and better,” said Kearns. Additionally, “we saw an increased level of throughput, more patients per shift, per physician.”
The Cat, the X-rays, and the AI Who Saw What We Couldn’t - Dr. Jeanne Beatrix Law, May 6, 2025.
”ChatGPT didn’t save Biggs. But she gave me a new way to see his condition—with granular, trackable language. She helped me recognize the gap between “looks okay” and “is okay.” That matters.”
Our new AI strategy puts Wikipedia’s humans first - Chris Albon and Leila Zia, Wikimedia Foundation, April 20, 2025.
”We believe that our future work with AI will be successful not only because of what we do, but how we do it. Our efforts will use our long-held values, principles, and policies (like privacy and human rights) as a compass: we will take a human-centered approach and will prioritize human agency; we will prioritize using open-source or open-weight AI; we will prioritize transparency; and we will take a nuanced approach to multilinguality, a fundamental part of Wikipedia.”
DolphinGemma: How Google AI is helping decode dolphin communication - Google, April 14, 2025.
”..on National Dolphin Day, Google, in collaboration with researchers at Georgia Tech and the field research of the Wild Dolphin Project (WDP), is announcing progress on DolphinGemma: a foundational AI model trained to learn the structure of dolphin vocalizations and generate novel dolphin-like sound sequences.'“
The future
Voicing Change: How the next news interface sets a generational challenge – and opportunity – for newsrooms - World Association of News Publishers, May 8, 2025.
”‘Information is no longer a broadcast – it is now a dialogue.’ Nikita Roy illustrates how the next wave of conversational, voice-first AI is already upon us – boosted by Gen Alpha and GenZ’s active participation, with deep implications for the future of news.”
What Lovable teaches us about AI tools - Marcel Salathé, May 4, 2025.
”Two weeks ago, I started experimenting with Lovable, one of the so-called “vibe coding” platforms that let you generate web applications just by describing what you want.” … “Despite all these issues, I’m quite excited about these tools. The issues I encountered aren’t fundamental flaws, but rather growing pains. And most of them will be fixed. When they are, this will be absolutely transformative.” “We’re early. Very early. The problems today … will get smoothed out fast. When they do, the entire UX of web / app development may look completely different.”
It’s Me, Hi. I’m the Vibe Coder. - Katie Parrott, Every, May 1, 2025.
”It’s easy to dismiss these as edge cases, but I see them as early signals—prototypes of what the future of building might look like. Vibe coding feels similar to a workplace innovation like Excel, which gave non-programmers a way to automate insight and express their expertise in a scalable way.”
AGI is not a milestone: There is no capability threshold that will lead to sudden impacts - Sayash Kapoor and Arvind Narayanan (the authors of AI Snake Oil), May 1, 2025.
”Even if general-purpose AI systems reach some agreed-upon capability threshold, we will need many complementary innovations that allow AI to diffuse across industries to realize its productive impact. Diffusion occurs at human (and societal) timescales, not at the speed of tech development.”
These Startups Are Building Advanced AI Models Without Data Centers - Will Knight, Wired, April 20, 2025.
”Distributed approaches could make it possible for smaller companies and universities to build advanced AI by pooling disparate resources together. Or it could allow countries that lack conventional infrastructure to network together several data centers to build a more powerful model.”
The ‘era of experience’ will unleash self-learning AI agents across the web—here’s how to prepare - Ben Dickson, VentureBeat, April 30, 2025.
”The authors argue that the “pace of progress driven solely by supervised learning from human data is demonstrably slowing, signaling the need for a new approach.” And that approach requires a new source of data, which must be generated in a way that continually improves as the agent becomes stronger. “This can be achieved by allowing agents to learn continually from their own experience, i.e., data that is generated by the agent interacting with its environment,” Sutton and Silver write.”
Learn more
If you want to learn more about generative AI, contact me about doing a webinar or course for your group. This spring I offered multiple webinar series — for Boston Library Consortium, Penn State Libraries, and Mayo Clinic Libraries. I’m taking a break for the summer, but will be open to doing more in the fall.
»» Sign up here if you’d like to be notified of future courses I’m offering.««
And as always, you can follow me on Bluesky or Mastodon where I post daily about generative AI.