Nicole Hennig
LightNews LightNews
banner
nic221.bsky.social
Nicole Hennig
@nic221.bsky.social
4.9K followers 400 following 3.8K posts
E-learning dev & AI educator at U of Arizona Libraries. Former head of UX at MIT Libraries. Winner of MIT Excellence Award. nicolehennig.com. Digital nomad from 2013-17, locationflexiblelife.com.
- vegetarian
- car-free
- universal basic income: yes
Posts Replies Media Videos
nic221.bsky.social
Vibe Check: Opus 4.5 Is the Coding Model We've Been Waiting For every.to/vibe-check/vib… #AI #coding #Anthropic
Text Shot: We’ve been testing Opus 4.5 over the last few days on everything from vibe coded iOS apps to production codebases. It manages to be both great at planning—producing readable, intuitive, and user-focused plans—and coding. It’s highly technical and also human. We haven’t been this enthusiastic about a coding model since Anthropic’s Sonnet 3.5 dropped in June 2024. The most significant thing about Opus 4.5 is that it extends the horizon of what you can realistically vibe code.
November 30, 2025 at 4:22 PM
nic221.bsky.social
When Familiarity Breeds Trust: How Experience Turns AI Skeptics into Believers www.edelman.com/insights/famil… #AI #trust #opinions
Text Shot: Familiarity breeds contempt,” the saying goes, except when it comes to artificial intelligence. According to the Fall 2025 Edelman Trust Barometer Flash Poll: Trust and Artificial Intelligence at a Crossroads, hands-on experience is the fastest route to trust. The more people use AI, the more they trust it, and the more they trust it, the faster they embrace it. Across five major markets surveyed—Brazil, China, Germany, the UK, and the United States—trust has emerged as one of the most powerful forces determining whether AI adoption accelerates or stalls. In the poll, trust in AI ranged from 87 percent in China and 67 percent in Brazil to just 39 percent in Germany, 36 percent in the UK, and 32 percent in the United States.
November 30, 2025 at 3:44 PM
nic221.bsky.social
Going Bananas 🍌 www.digitalnative.tech/p/going-banana… #AI #NanoBanana #images #society
Text Shot: much of the brewing AI backlash stems from general discontent with societal problems that go far beyond tech. Inflation, stagnant wages, tariff-induced high prices. People aren’t feeling great about things. Policy is hugely to blame, but tech is a convenient villain, a villain that embodies the billionaire class and the gulf between haves and have-nots. I think this anti-tech sentiment is going to be one of the defining narratives of the next decade, and it’s only going to get worse: AI will drive major productivity gains, which will boost the stock market; but most Americans don’t participate in the capital markets, meaning most people won’t going to share in the gains. The Gini coefficient will go up. I haven’t seen a new exciting technology be met with such vitriol. The internet and mobile eras both sparked hand-wringing, sure, but mostly from parents worried about screen-time and the highlight reel of social media. This is another level. Everything that’s happening…
November 30, 2025 at 3:28 PM
nic221.bsky.social
Rare look at how Hollywood is already harnessing AI abcnews.go.com/Technology/rar… #AI #filmmaking #video
Text Shot: "There are things that I wasn't able to achieve before, because I have a very big vision of creating the cosmos, and outer space, and having these images that are really difficult to achieve otherwise. And so AI was really integral to the process," McNitt told ABC News. Using a special version of Veo 3, McNitt decided she would train the AI on images of her as a child to eliminate the need for a real baby on set. "I felt it was actually quite unethical to shoot with a newborn baby," McNitt said. "We wanted to find a new innovative solution to that that felt very realistic. So in the movie, what you're seeing is actually me as a newborn baby."
November 30, 2025 at 3:16 PM
nic221.bsky.social
📈 Why everyone missed solar’s exponential growth www.exponentialview.co/p/the-forecast… #SolarEnergy #forecasts #predictions
Despite this new energy paradigm, many influential forecasters continue to apply commodity-based thinking to technology improvements. This has led to a systematic overestimation of the cost of renewables. Imposing constraints and cost floors, while potentially appropriate for commodities, fails to capture the technological learning demonstrated by solar, wind and batteries. The result has been a persistent undervaluation of renewable energy's potential and an overestimation of transition costs.
November 29, 2025 at 3:43 PM
nic221.bsky.social
Nano Banana Pro is the “ChatGPT Moment” for Visual Communication - Jakob Nielsen jakobnielsenphd.substack.com/p/nano-banana-… (lots of details and ideas about use of these types of images) #AI #UX #NanoBanana #images #VisualCommunication
A smiling person in a full banana costume uses a pencil to point at a whiteboard filled with colorful drawings and a flowchart about illustration methods, including digital painting, vector graphics, hand-drawn sketches, concept art, and data visualization.
November 28, 2025 at 4:55 PM
nic221.bsky.social
Google TPUv7: The 900lb Gorilla In the Room newsletter.semianalysis.com/p/tpuv7-google… (lots of details about TPUs) #AI #chips #Google
Text Shot: For Google to quietly muscle in and establish a performance lead in one of the most challenging hardware problems is a truly impressive feat for a company whose core business isn’t, or should we say, wasn’t in hardware business
November 28, 2025 at 4:39 PM
nic221.bsky.social
🟡 Semafor Tech: AI Glossary www.semafor.com/newsletter/11/… (of paradoxes) #AI #terminology
Text Shot: Amara’s Law: We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run. The Bitter Lesson: Building knowledge into AI agents only limits their knowledge and capabilities. The Black Box Paradox (or fallacy): We have to choose between powerful AI we can’t trust and trustworthy AI that isn’t powerful. Jevons paradox: The cheaper AI becomes, the more demand increases. Moravec’s Paradox: AI is good at the hard things and bad at the easy things. The productivity paradox: AI is everywhere but hasn’t made a dent in productivity. Tesler’s Theorem: AI is whatever hasn’t been done yet.
November 28, 2025 at 4:20 PM
nic221.bsky.social
Google rolls back NotebookLM's new features, sort of www.xda-developers.com/google-rolls-b… #AI #NotebookLM
Text Shot: Last week, Google rolled out NotebookLM’s new Slide Decks and Infographics features. Google has now temporarily restricted access for free users and added limits for paid users due to overwhelming demand. The specific limits for these new features remain unclear, as support articles do not provide concrete numbers for either free or Pro users.
November 28, 2025 at 3:19 PM
nic221.bsky.social
Anthropic reveals new Opus 4.5 model, brings Claude Code to the Mac app - 9to5Mac 9to5mac.com/2025/11/24/ant… #AI #Claude #Mac
Text Shot: Additionally, Claude Code has been added to Anthropic’s desktop apps including the Mac version. Claude Code was previously limited to mobile apps and the web. It allows software engineers to code, research, and update work with multiple local and remote sessions running at the same time.
November 28, 2025 at 3:17 PM
nic221.bsky.social
Debunking AI’s Environmental Panic - with Andy Masley pca.st/episode/54e370… #AI #environment (good episode)
PRESENTED BY Galileo CHAIN THOUGHT PODCAST Hosted by Conor Bronsdon Debunking Al's Environmental Panic | Andy Masley Chain of Thought
November 28, 2025 at 2:27 AM
nic221.bsky.social
A weekend ‘vibe code’ hack by Andrej Karpathy quietly sketches the missing layer of enterprise AI orchestration venturebeat.com/ai/a-weekend-v… #AI #coding
To the casual observer, the LLM Council web application looks almost identical to ChatGPT. A user types a query into a chat box. But behind the scenes, the application triggers a sophisticated, three-stage workflow that mirrors how human decision-making bodies operate. First, the system dispatches the user's query to a panel of frontier models. In Karpathy's default configuration, this includes OpenAl's GPT-5.1, Google's Gemini 3.0 Pro, Anthropic's Claude Sonnet 4.5, and xAl's Grok 4. These models generate their initial responses in parallel. In the second stage, the software performs a peer review. Each model is fed the anonymized responses of its counterparts and asked to evaluate them based on accuracy and insight. This step transforms the Al from a generator into a critic, forcing a layer of quality control that is rare in standard chatbot interactions. Finally, a designated "Chairman LLM" — currently configured as Google's Gemini 3 - receives the original query, the individual
November 27, 2025 at 7:18 PM
nic221.bsky.social
Gemini 3 Solves Handwriting Recognition and it’s a Bitter Lesson generativehistory.substack.com/p/gemini-3-sol… #AI #Gemini #ComputerVision #handwriting
For the historical community, as we gradually become accustomed to this new reality, it will radically alter how historians, genealogists, archivists, governments, and researchers relate to our documentary past. It's also a harbinger of the larger changes in our relationship with information that are now certain to come from scaling.
November 27, 2025 at 3:42 PM
nic221.bsky.social
[Guest post] Lost in (incidental) memorization: When the (case) law mistakes AI training for copying ipkitten.blogspot.com/2025/11/guest-… #AI #cooyright
Text Shot: . The InfoSoc Directive permits the incidental inclusion of a work “in other material” under Article 5(3)(i). However, the Munich Court rejected this argument, stating that the law requires the larger material to be a work. Nevertheless, the Directive itself refers to “other material” and not “other work”. Different language versions (French: “dans un autre produit”, German: “in anderes Material”) confirms that the scope is broader than in the Munich judgment. If a model’s memorization is unintentional and rare, and the model’s primary function is not to reproduce protected works, then it could qualify as incidental inclusion. This could be a key argument in OpenAI’s appeal. Training such models involves exposing them to many examples (e.g., images or texts) so they can learn statistical correlations in the data that are useful for performing a specific task. During training, the model makes predictions, measures the resulting error, and adjusts its internal parameters,…
November 27, 2025 at 3:36 PM
nic221.bsky.social
Microsoft’s Fairwater datacenter will use more power than Los Angeles epochai.substack.com/p/microsofts-f… (read beyond the headline) #AI #DataCenters
It should be noted that today's chatbots do not use large amounts of energy per query. Earlier this year, we estimated a typical GPT-40 ChatGPT query uses about as much energy as using a microwave for one second. Energy consumption per ChatGPT query is small compared to everyday electricity use Typical ChatGPT query - 0.3 Running a 10W lightbulb for 5 minutes Using a laptop for 5 minutes - Long-input query - Microwaving for 30 seconds - Average US household consumption per minute Maximum-input query ⅖ EPOCH AI 10 30 40 Energy cost (watt-hours) Pessimistic estimates of the energy usage of ChatGPT with GPT-4o across for different query lengths: typical (<100 words), long (~7,500 words), and maximum context length (~75,000 words), with an average response length of 400 words. CC-BY epoch.ai It's unclear how energy-intensive future AI systems will be to run. Increasing use of inference-time compute scaling may push energy usage per query upwards, while algorithmic and hardware efficiency
November 27, 2025 at 3:03 AM
nic221.bsky.social
Warner Music Group strikes ‘landmark’ deal with Suno; settles copyright lawsuit against AI music generator - Music Business Worldwide www.musicbusinessworldwide.com/warner-music-g… #AI #music #copyright (similar to Udio deal)
Text Shot: Artists and songwriters, according to the companies, “will have full control over whether and how their names, images, likenesses, voices, and compositions are used in new AI-generated music”. In 2026, according to the press release, Suno will make “several changes to the platform, including launching new, more advanced and licensed models”. The release added: “When the new models launch in 2026, the current models will be deprecated. Moving forward, downloading audio will require a paid account”. Suno also says that it will introduce download restrictions in certain scenarios: “specifically, in the future, songs made on the free tier will not be downloadable and will instead be playable and shareable”.
November 26, 2025 at 3:47 PM
nic221.bsky.social
The Munich Mirage: Why the "GEMA vs. OpenAI" Verdict is Less About AI and More About Geography www.linkedin.com/pulse/munich-m… #AI #copyright
The Appeal Reality: Why You Shouldn't Panic Here is the data point that the alarmist headlines are missing: The Reversal Rate. The "notoriety" of courts like Munich comes with a catch. Because they often take extreme, plaintiff-friendly positions to attract cases, their rulings are frequently overturned by higher courts. Empirical observations of the German IP system reveal a massive "pendulum effect." Aggressive first-instance judgments often face a buzzsaw when they reach the Federal Court of Justice (BGH) or get referred to the Court of Justice of the European Union (CJEU). In patent nullity cases, for example, reversal or modification rates can hover near 80%. Even in general IP litigation from "magnet courts," it is estimated that a third of these aggressive judgments are overturned on appeal. Remember the "Wallpaper" insanity from Cologne? The Federal Court of Justice recently stepped in and effectively told them to get a grip, overturning those rulings and restoring common sense
November 26, 2025 at 3:19 PM
nic221.bsky.social
AI’s Environmental Impact www.library20.com/ai-s-environme… (I’m repeating this webinar on Dec 5) #AI #environment #climate #webinars
Text Shot: This webinar cuts through the confusion to help you make informed choices about sustainability in your classrooms, libraries, and communities. We’ll examine independent estimates of AI’s energy and water use and put them in context in ways that are easy to understand. We’ll include an introduction to how data centers work and what they are used for. We’ll clarify what we know and what’s still uncertain about AI’s carbon footprint (both in the present and in future projections). We’ll compare individual AI use to other digital activities, and we’ll also look at global use of data centers with statistics from the International Energy Agency.
November 26, 2025 at 2:56 PM
nic221.bsky.social
And now for a break from AI news to see the best cat in the world - our cat, Max. Super-smart & affectionate. #cats #TabbyCats
A tabby cat with dark stripes sits alert on an orange and blue blanket, looking toward the camera with a gold heart-shaped ID tag on its collar.
November 26, 2025 at 2:53 PM
nic221.bsky.social
HypeDetector: Based on my linguistic analysis of this WIRED article, here are the key instances of misleading language and rhetorical techniques: ## **Framing and Loaded Language** poe.com/s/X9TL4axvvGD3… (my Hype Detector bot on a story from Wired) #AI #reporting #misleading
Structural Bias Techniques • Headline-to-content gap - Headline promises "new forever chemicals" but article shows existing chemicals getting faster review • Buried counterarguments - EPA's detailed response about maintaining review standards appears deep in article • Conclusion placement - Ends with critic's "safety implications" warning rather than balanced assessment
November 26, 2025 at 2:45 PM
nic221.bsky.social
The Case for AI as Accommodation (opinion) www.insidehighered.com/opinion/views/… #AI #education #accessibility
Text Shot: I encouraged this student to share his story at our next class meeting, which he did. Inspired by his bravery and honesty, soon other students began opening up. The stories students have shared across my classes have been revelatory to me. A student with aphantasia uses AI to generate diagrams and mnemonics in organic chemistry, compensating for the fact that she cannot form mental images. Another battling anxiety builds confidence by generating practice tests with ChatGPT, rehearsing so she no longer freezes at exam time. A commuter student turns dense readings into audio so she can review them while driving. One student identifying as on the autism spectrum told me she uses AI for social insights, such as how she could contribute to a group project without sounding “pushy.” A first-generation student told me she relies on AI to figure out what she needs to do in college: How to email a professor, what office hours are for, what an annotated bibliography is. A…
November 26, 2025 at 2:23 PM
nic221.bsky.social
Paper Notes: Nested Learning itcanthink.substack.com/p/paper-notes-… #AI #research #Google
As usual though, with a theory paper like this, it's worth noting that these seem fairly far from use on anything more than toy tasks — and that inference with transformers continues to improve, in part because they scale extremely well. Titans and HOPE won't be replacing GPT next year or anything. But the idea here — that we're thinking about neural networks wrong, that "optimizers" and "modules" are not actually discrete entities — seems extremely interesting and has a ton of potential for the future.
November 26, 2025 at 6:46 AM
nic221.bsky.social
Six reasons to think there’s an AI bubble — and six reasons not to www.understandingai.org/p/six-reasons-… #AI #bubble #boom
Last week, we were discussing how one could capture the entire Al-bubble debate in about 12 statistics that people just keep citing and reciting - on CNBC, on tech podcasts, in Goldman Sachs Research documents, and at San Francisco AI parties. Since everybody seems to be reading and quoting from the same skinny playbook, we thought: What the hell, let's just publish the whole playbook! If you read this article, we think you'll be prepared for just about every conversation about Al, whether you find yourself at a Bay Area gathering with accelerationists or a Thanksgiving debate with Luddite cousins. We think some of these arguments are compelling. We think others are less persuasive. So, throughout the article, we'll explain both why each argument belongs in the discussion and why some arguments don't prove as much as they claim. Read to the end 1, and you'll see where each of us comes down on the debate.
November 26, 2025 at 4:10 AM
nic221.bsky.social
How GPT-5 helped mathematician Ernest Ryu solve a 40-year-old open problem openai.com/index/gpt-5-ma… #AI #math #GPT5
Text Shot: It was a surreal feeling,” he said, equating the experience to working with a competent student who would come up with ideas, ask questions, and brainstorm. “It was really intellectually exciting to work with GPT‑5,” Ryu said. “I had to be vigilant in checking the calculation steps, but it was still much faster than doing the calculation myself.” While the process was significantly faster, it also required careful verification. GPT‑5 often produced arguments that looked plausible but did not hold up on inspection. Ryu reviewed each suggestion, discarded the ones that failed, and developed the ones that contained a mathematically meaningful idea. The workflow depended on his judgment—both to identify promising directions and to recognize when a line of reasoning had been fully explored. Ryu used GPT‑5 like a tool, and he took great care to understand what it was doing so he could discern what it was good at and what it wasn’t. When asking GPT‑5 to check the work, Ryu found…
November 26, 2025 at 2:24 AM
nic221.bsky.social
Introducing shopping research in ChatGPT openai.com/index/chatgpt-… #AI #ChatGPT #shopping
Text Shot: Shopping research is built for that deeper kind of decision-making. It turns product discovery into a conversation: asking smart questions to understand what you care about, pulling accurate, up-to-date details from high-quality sources, and bringing options back to you to refine the results. It performs especially well in detail-heavy categories like electronics, beauty, home and garden, kitchen and appliances, and sports and outdoor.
November 25, 2025 at 11:49 PM