Nicole Hennig

Text Shot: To build a useful AI model, you need to journey into the wild base model and stake out a region that is amenable to human interests: both ethically, in the sense that the model won’t abuse its users, and practically, in the sense that it will produce correct outputs more often than incorrect ones. What this means in practice is that you have to give the model a personality during post-training1. Human beings are capable of almost any action at any time. But we only take a tiny subset of those actions, because that’s the kind of people we are. I could throw my cup of coffee all over the wall right now, but I don’t, because I’m not the kind of person who needlessly makes a mess2. AI systems are the same. Claude could respond to my question with incoherent racist abuse - the base model is more than capable of those outputs - but it doesn’t, because that’s not the kind of “person” it is. In other words, human-like personalities are not imposed on AI tools as some kind of…

Giving LLMs a personality is just good engineering - Sean Goedecke www.seangoedecke.com/giving-llms-a-… #AI #training #personalities

BookReconciler P - Metadata Enrichment and Work-Level Clustering • Who is this for? Digital humanities researchers, librarians, metadata specialists, and more. • What does it do? Finds, clusters, and enriches records for books. Adding ISBNS, HathiTrust IDs, subject headings, descriptions, page counts, publication dates, and more. To learn more about this tool's design, motivations, and related work, or to cite this tool, please see the following paper: "BookReconciler * •: An Open-Source Tool for Metadata Enrichment and Work-Level Clustering". ' Matt Miller, Dan Sinykin, and Melanie Walsh. Joint Conference on Digital Libraries. December 2025.

GitHub - Post45-Data-Collective/BookReconciler: BookReconciler, A Tool for Metadata Enrichment and Clustering of Book Data github.com/Post45-Data-Co… #AI #metadata #libraries #books

Text Shot: Four people with different risk tolerances and technical backgrounds all landed in the same place: Personal AI agents are going to be a basic part of how we live and work. A month ago, none of these agents existed. Now Felix writes its own tweets, Zosia orders butter with the right preferences, and Polly reschedules meetings from a Target parking lot. They’ll be better next month. If you listen in on the camp or follow this setup guide, yours could be, too.

OpenClaw: Setting Up Your First Personal AI Agent every.to/source-code/op… #AI #agents

I admire Anthropic for its principled stance, but ultimately I'm not sure even strong contractual restrictions would have made much difference. The Pentagon already has a deal in place with xAI that puts few restrictions on military use of AI. Moreover, open-weight models are already good enough for many surveillance activities, and they '11 presumably become suitable for even more in the coming months and years. Indeed, even Dario Amodei believes that contractual agreements are only a stopgap solution to preventing abuse of AI models. "In the long run, I actually do believe that it is Congress's job," Amodei said in a Saturday interview on CBS. He urged Congress to "catch up" with laws to limit domestic mass surveillance. And that mav

The Pentagon’s bombshell deal with OpenAI, explained - Timothy B Lee www.understandingai.org/p/the-pentagon… #AI #Pentagon #OpenAI #Anthropic

Text Shot: With uniform documents, Rodriguez and his team use Optical Character Recognition — a long-standing technology that turns scanned documents into machine-readable, searchable text — to tackle analysis. But the Foster cards have important stamped information, including “accredited,” “accepted” and “rejected,” that wasn’t always picked up by the traditional method. Rodriguez and his colleague, Matt Jansen, turned to AI. Using ChatGPT and custom coding scripts, they reanalyzed the cards — 179,481 PDF scans in total. With AI, Rodriguez was able to extract more data from the collection and format that information, including production company, genre, subject matter, actors’ names and more, in a digestible way. For Rodriguez, the application is a win for the preservation of film history and for humanities researchers who might be interested in how AI can advance their work.

AI powers research into film’s past - College of Arts and Sciences college.unc.edu/2026/03/ai-fil… #AI #FilmHistory #Humanities

Text Shot: We find that users submit longer and more complex queries than in traditional search, and treat the system as a collaborative research partner, delegating tasks such as drafting content and identifying research gaps. Users treat generated responses as persistent artifacts, revisiting and navigating among outputs and cited evidence in non-linear ways. With experience, users issue more targeted queries and engage more deeply with supporting citations, although keyword-style queries persist even among experienced users. We release the anonymized dataset and analysis with a new query intent taxonomy to inform future designs of real-world AI research assistants and to support realistic evaluation.

New From Allen Institute For Artificial Intelligence (Ai2): Understanding Usage and Engagement in AI-Powered Scientific Research Tools: The Asta Interaction Dataset (preprint) - Library Journal infoDOCKET www.infodocket.com/2026/02/26/new… #AI #ResearchTools

Text Shot: This was using Claude Opus 4.6, which turns out to have quite good taste when it comes to building explanatory animations. If you watch the animation closely you can see that for each word it attempts to place it somewhere on the page by showing a box, run checks if that box intersects an existing word. If so it continues to try to find a good spot, moving outward in a spiral from the center. I found that this animation really helped make the way the algorithm worked click for me. I have long been a fan of animations and interactive interfaces to help explain different concepts. A good coding agent can produce these on demand to help explain code - its own code or code written by others.

Interactive explanations - Agentic Engineering Patterns simonwillison.net/guides/agentic… #AI #agents

Fact Checking a Misleading Iran War Meme - Wes Fryer wfryer.substack.com/p/fact-checkin… #AI #FactChecking #MikeCaulfield

Notion tests custom agents with Computer Use and Claude Code Notion is developing custom agents with direct computer and Claude Code connectors, enabling coding within managed environments.

Notion tests custom agents with Computer Use and Claude Code www.testingcatalog.com/notion-tests-c… #AI #Notion #agents

Text Shot: Google is gearing up for a major overhaul of the Build feature in AI Studio, with signs pointing to a release as early as next week. The update is expected to transform AI Studio's vibe coding environment from a simple app prototyping tool into a more capable platform for building full-stack, multiplayer applications with proper authentication and deeper Google service integrations. Logan Kilpatrick, who leads developer relations for Google's AI products, has recently teased the upcoming changes publicly.

Google AI Studio will be able to generate full-stack apps www.testingcatalog.com/google-ai-stud… #AI #Google

Text Shot: ProducerAI is designed for “artists of every level,” catering to both first-time creators and established names. According to Google, the tool assists in writing lyrics, refining melodies, remixing tracks, and exploring genre blends. It was developed with input from a community that includes artists like Lecrae and The Chainsmokers. 🚨 Big news: @producer_ai is officially joining Google Labs! 🎶 ProducerAI is a creative collaborator, whether you’re writing lyrics, developing a melody or experimenting with genres. With ProducerAI, you can turn your imagination into dynamic tracks. We share ProducerAI’s… — Google Labs (@GoogleLabs) February 24, 2026 The technology stack is explicitly Google DeepMind. Gemini provides the conversational layer, Lyria 3 generates music, Veo is used for music video creation, and Nano Banana handles image outputs such as artwork. Google highlights that ProducerAI utilizes a preview version of Lyria 3, which is described as high-fidelity and…

Google adds ProducerAI for music creation to Labs platform www.testingcatalog.com/google-adds-pr… #AI #music #Google

Text Shot: Code references consistently point to "learning goals" as the primary use case, which aligns closely with Google's investment in LearnLM, its education-focused AI initiative built on learning science principles and already integrated into Gemini. This suggests the feature could allow users to set a learning target and have Gemini periodically guide them through structured progress, quizzes, or resource curation.

Google tests new Learning Hub powered by goal-based actions www.testingcatalog.com/google-tests-n… #AI #Gemini

Text Shot: Perhaps more interesting is a "Remix" button currently hidden in the interface. This would allow users to clone someone else's public notebook and use it as a starting point for their own research. Since Google introduced public notebook sharing in mid-2025, over 140,000 public notebooks have been created by the community. A remix function would create a natural feedback loop: users build notebooks, share them publicly, and others fork them to extend or adapt the work. While Google is unlikely to launch a full public feed or marketplace for notebooks, the remix concept still lowers the barrier for collaborative knowledge-building across the platform.

Google tests custom banners and Remix feature for NotebookLM www.testingcatalog.com/google-tests-c… #AI #NotebookLM

Text Shot: Anthropic is advancing Cowork deeper into enterprise workflows by launching an updated plugins system for Claude Enterprise. This system allows admins to build private, organization-specific plugin marketplaces and distribute internal tools across teams. The aim is to transform Claude into role-specific “agents” for departments such as finance, HR, design, engineering, legal, and operations, while maintaining tighter administrative control over the tools and data accessible to each team.

Anthropic upgrades Cowork and plugins on Claude Enterprise www.testingcatalog.com/anthropic-upgr… #AI #Claude #plugins

It sounds like a catastrophe. Citrini's piece had a 3.4-to-1 ratio of negative/crisis words to positive ones, with 198 subjective/ emotional words. It was designed to elicit an emotional response. And it did. I take two issues with their essay - Citrini assumes no adaptation and that everything happens everywhere all at once. As investor Gavin Baker observed, the scarcity of watts and wafers will significantly slow the scenario. He

Citrini, of course, is not all wrong. The first flywheel has kicked off - cheap, fast software is already undercutting mid-market SaaS. But this focuses on the wrong part of the story. This month, someone on our team replaced a $40/month Saas product in a couple of hours by building something that actually worked how they work. In this case, it's not a job loss but a capability gain. We need to look at the gains and the pinch points of the transition. The collateral damage of railroads cannot be ignored, but nor can the fact that they didn't just kill the canal companies; they also created an entirely new geography of economic possibility. In creative destruction, markets always show what is destroyed first. And the markets are selling the canal phase of an Al transition that, like railroads, will destroy visible incumbents while building new economic possibilities. But that destruction will be slowed by the compute crunch, which buys time for adaptation that Citrini doesn't consider.

🔮 Exponential View #563: The Citrini craze; human cognition; Anthropic’s moral machine; COBOL returns; bye‑bye tax filing and Moon mining++ www.exponentialview.co/p/ev-563 #AI #CitriniReport

Text Shot: In line with our recent Economic Index, we find that the most common expression of AI fluency is augmentative—treating AI as a thought partner, rather than delegating work entirely. In fact, these conversations exhibit more than double the number of AI fluency behaviors than quick, back-and-forth chats. But we also find that when AI produces artifacts—including apps, code, documents, or interactive tools—users are less likely to question its reasoning (-3.1 percentage points) or identify missing context (-5.2pp). This aligns with related patterns we observed in our recent study on coding skills. These initial findings present us with a baseline that we can use to study the development of AI fluency over time.

Anthropic Education Report: The AI Fluency Index www.anthropic.com/research/AI-fl… #AI #Claude #use

For those of us who work in education, in libraries, in institutions that still believe knowledge carries obligations, there is a sharper lesson buried in the wreckage. We talk about AI governance as though it were a matter of getting the frameworks right, of building the appropriate guardrails. The February crisis showed what happens when guardrails meet a state that refuses to be guarded. They don't hold. They get redesigned into something that looks like a guardrail but functions as a courtesy. Anthropic will likely survive this. It will pivot to European markets, to enterprise privacy, to the growing global demand for AI systems with binding ethical commitments. It may even emerge stronger in the markets that value what the Pentagon punished. But the precedent is set. In the contest between principle and power, power didn't destroy principle. It did something more efficient. It found a partner willing to translate principle into architecture, where it can be quietly renovated

The Handshake That Replaced the Red Line hybridhorizons.substack.com/p/the-handshak… #AI #OpenAI #Anthropic

Text Shot: OpenAI cautioned that any breach of its contract by the U.S. government could trigger a termination, though it added, "We don't expect that to happen." The company also said rival Anthropic should not be labeled a “supply-chain risk,” noting, "We have made our position on this clear to the government."

OpenAI details layered protections in US defense department pact www.reuters.com/business/media… #AI #OpenAI

Market dilution and mass surveillance could be a toxic mix I've highlighted market dilution and large-scale inquiry into individual uses of AI to spotlight two areas where we may be on the cusp of significant change in fourth-factor analysis. What worries me is a version of the future where intrusion into personal creative spaces becomes normalized. Where we protect works in ways that attempt to protect humans from machines but do so by sacrificing the ways humans have traditionally consumed, studied, remixed, and learned from works.

Some might argue that we can simply move these activities offline or out of the emerging Al space. But that misunderstands the concern. As I stated at the beginning of this post, we grant rights to rightsholders subject to a host of rights we reserve to ourselves. We reserve and should continue to reserve the freedom to experiment, to test technologies, to rework and analyze the art of others (sometimes commercially, often not) in ways that make culture dynamic and knowledge cumulative. It would be a diminished world if we quietly surrendered those rights through doctrinal drift and expansive discovery in intrusive litigation, to an ever-watching apparatus that finds new ways to meter, monitor, and monetize the ordinary ways we learn, play, and create.

Markets Markets Markets (Value!) - Authors Alliance authorsalliance.substack.com/p/markets-mark… #AI #copyright (good points)

Trump moves to blacklist Anthropic AI from all government work The penalty is usually reserved for companies from adversarial countries, like China's Huawei.

Trump moves to blacklist Anthropic AI from all government work www.axios.com/2026/02/27/ant… #AI #Anthropic

Text Shot: After Anthropic CEO Dario Amodei stood firm by his company's red lines, employees from OpenAI and Google signed onto a letter in solidarity on Thursday, pushing executives at their respective companies to resist "pressure" from the Pentagon.

Sam Altman says OpenAI shares Anthropic's red lines in Pentagon fight www.axios.com/2026/02/27/alt… #AI #OpenAI #Anthropic

Why Most Agent Harnesses Are Not Bitter Lesson Pilled [Article] Minh Pham The applied version and it surfaces a point people keep missing: agents are not humans. Most people building AI agent systems mirror human org charts — a Researcher agent, a Coder agent, a Writer agent. This makes intuitive sense because it's how we organize people. But a human-shaped amount of work and an agent-shaped amount of work look completely different. A human reading a thousand-page document needs days. An agent does it in seconds. A human can hold maybe seven things in working memory. An agent can cross-reference a hundred documents simultaneously. The flip side is also true. A human can walk into a room and instantly read the social dynamics and vibe.

Why Most Agent Harnesses Are Not Bitter Lesson Pilled [Article] - Minh Pham
threemagnolia.activehosted.com/index.php?acti… #AI #agents

Text Shot: This doesn’t mean native Mac developers are obsolete. I still used a whole bunch of my own accumulated technical knowledge (and the fact that I’d already installed Xcode and the like) to get this result, and someone who knew what they were doing could have built a far better solution in the same amount of time. It’s a neat illustration of how those of us with software engineering experience can expand our horizons in fun and interesting directions. I’m no longer afraid of Swift! Next time I need a small, personal macOS app I know that it’s achievable with our existing set of tools.

I vibe coded my dream macOS presentation app - Simon Willison simonwillison.net/2026/Feb/25/pr… #AI #VibeCoding

Deadline Day for Autonomous AI Weapons & Mass Surveillance www.youtube.com/watch?v=Cru804… #AI #Anthropic (interesting details)

In reality, businesses are risk-averse and consumers creatures of habit. Radiologists were supposed to lose their jobs to offshoring, and then to AI. They didn't, because patients and providers like having humans around to explain their medical images. Since Google Translate launched in 2006, the number of human translator and interpreter employees in the U.S. has risen 73%.

Tech Has Never Caused a Job Apocalypse. Don’t Bet on It Now. www.wsj.com/economy/jobs/t… #AI #jobs

With a projected $18 billion in 2026 revenue, Anthropic could easily afford to walk away from a $200 million contract. The Pentagon's leverage comes from the possibility that it could use a supply chain risk designation to force a bunch of other companies to choose between working with Anthropic or doing business with the federal government. But this would be a double-edged sword. Companies that do most of their business in the private sector might decide they'd rather drop the Pentagon as a customer than cut themselves off from a leading AI provider. The ultimate result might be that the Pentagon loses access to some of Silicon Valley's best technology.

The Pentagon is making a mistake by threatening Anthropic www.understandingai.org/p/the-pentagon… #AI #Anthropic

The OpenClaw-ification of AI www.youtube.com/watch?v=GR-j31… #AI #agents

Text Shot: The feature allows developers to initiate a complex task in their terminal and maintain full control of it from a phone or tablet, effectively decoupling the AI agent from the physical workstation. Currently, Remote Control is available as a Research Preview for subscribers on the Claude Max tier. While access for Claude Pro ($20/month) users is expected shortly, the feature remains a high-end tool for power users and is notably absent from Team or Enterprise plans during this initial phase.

Anthropic-just-released-a-mobile-version-of-claude-code-called-remote venturebeat.com/orchestration/… #AI #ClaudeCode #mobile

Text Shot: Perplexity launched a new product on Wednesday with an old name — “Computer” — that acts as a super agent that can create a website, write a report, or generate a dataset. It’s like the next level of vibe coding — all the user has to do is describe the end-product they want and Computer breaks down the assignment for task-specific sub-agents to do the work from start to finish. It even chooses the best model for the specific job, outsourcing to Nano Banana for images, Veo 3.1 for video, Gemini for deep research, and Grok for quick searches. It runs autonomously for hours, or even months, according to the company.

Perplexity launches ‘Computer’ super agent www.semafor.com/article/02/25/… #AI #Perplexity #agents

Text Shot: Anthropic introduced new prebuilt plugin templates spanning HR, design, engineering, operations, financial analysis, investment banking, equity research, private equity, and wealth management. The company also shipped new MCP connectors for Google Drive, Google Calendar, Gmail, DocuSign, Apollo, Clay, Outreach, SimilarWeb, MSCI, LegalZoom, FactSet, WordPress, and Harvey — dramatically extending Claude's reach into the software ecosystem that enterprises already use. And Claude can now pass context seamlessly between Cowork, Excel, and PowerPoint, including across multiple files, without requiring users to restart when switching applications.

Anthropic-says-claude-code-transformed-programming-now-claude-cowork-is venturebeat.com/orchestration/… #AI #Claude #Cowork