AI Doesn't Just Assist—It Influences | Understand AI for October 21, 2025
This week's edition covers how AI reshapes human meaning-making, California's new safety disclosure law, and why AI benchmark scores might be telling you less than you think.
Segments & Tags
Discover our segments, carefully crafted for builders, users, and thought leaders to discover what's relevant in the realm of intelligence.
When Your Textbook Becomes Your Tutor | ApplyAI for October 14, 2025
This week's edition covers personalized learning breakthroughs, conversational shopping that finally speaks your language, Claude's training policy shift, smart home AI upgrades, and new parental controls for ChatGPT.
How We're Really Using AI Now | Apply AI for September 24, 2025
This week's edition covers major usage shifts, browser AI integration, file creation tools, and research on decision-making and focus techniques.
The Week ChatGPT Learned to Branch and Humans Learned to Pause
This week's edition covers ChatGPT's conversation branching breakthrough, NotebookLM's new debate and critique modes, and research revealing how we actually choose between thinking and reacting.
When AI Actually Works
Grammarly builds agents that don't need babysitting, researchers explain the AI disappointment cycle, and Hamburg shows how to make impossible group decisions possible
GPT-5 Ends the Choice Between Fast and Smart AI
This week's edition covers GPT-5's breakthrough in adaptive reasoning, surprising SuperAgers research that challenges health priorities, cognitive benefits of light exercise, decision-making frameworks from medical research, and productivity tools that streamline knowledge work.
AI Learns to Teach (And Does It Better)
This week's edition covers ChatGPT's new study mode that guides learning through questions, Harvard research showing AI tutoring outperforms classroom instruction, and why quick mindfulness hacks don't work—plus real developer stories on AI collaboration.
When AI Learns to Deceive and Remember
This week: Advanced AI models engage in strategic deception, memory-augmented systems that learn from corrections, and practical tools for better human-AI collaboration
ChatGPT Now Handles Complete Workflows Autonomously
OpenAI launches autonomous ChatGPT agents for multi-step tasks, new research reveals AI coding tools make experienced developers 19% slower, Google adds advanced search capabilities, and neuroscience-backed methods for restoring focus.
MIT Discovers AI Changes Your Brain
EEG scans reveal ChatGPT weakens memory networks, Claude ran a real store for 30 days with surprising results, plus new research on peak learning states and AI safety failures
Copilot Now Speaks Fluent Google—Your Lost Files Are Saved
This week's edition covers Microsoft Copilot's cross-platform search and memory features, ChatGPT's new voice-and-text integration, Google Gemini 3's coaching capabilities for students and creators, free ChatGPT access for teachers, and Claude's extended conversation threads.
Claude: extended conversations
Claude is Anthropic’s AI assistant, and you can chat with it on the web or your desktop. But until now, if you talked to Claude for too long, you’d suddenly hit a wall. The conversation would just stop, and you’d have to start over from scratch, losing
Google: Gemini 3
Gemini 3 is Google's smartest AI yet, and it's now in the hands of anyone with the Gemini app. That means over 650 million people each month can use it to work with text, images, video, audio, and even code. In other words, it's
ChatGPT: voice and text together
Imagine you could just talk to ChatGPT, ask your questions out loud, and actually hear the answers. That’s what Voice Mode is all about. Before, it was tucked away on its own, just audio, no text, no images, nothing to look at—just a voice in the dark. But
OpenAI: ChatGPT for Teachers
OpenAI has just rolled out ChatGPT for Teachers, and if you’re a verified K–12 teacher in the U.S., you can use it for free until June 2027. But here’s the thing: teachers aren’t waiting around. Three out of five are already using some kind of
Microsoft Copilot: memory and search
Microsoft Copilot is an AI assistant that follows you wherever you go, whether you’re on your laptop, your phone, or just browsing the web. With the latest update, Copilot gets a dozen new tricks, all designed to make it feel more like your own personal helper, no matter what
OpenAI Builds a Browser Around ChatGPT
This week's edition covers OpenAI's new Atlas browser that combines ChatGPT with web browsing and memory, and Adobe Firefly's AI-generated soundtracks and voiceovers for commercial projects
OpenAI: Atlas browser
Imagine if your web browser and ChatGPT were the same thing. That’s what OpenAI has done with ChatGPT Atlas. Instead of jumping back and forth between tabs, you just talk to the AI right where you’re working. Atlas is out now for Mac users everywhere, and it’s
Adobe Firefly: AI soundtracks and voiceovers
Adobe Firefly is a creative playground powered by AI, where you can make and edit images, videos, and even audio—all right in your browser. So, what’s new? Now, with just a click, you can create a soundtrack that fits your video perfectly—no more hunting for stock music
Adobe Firefly: commercial-safe AI
Adobe Firefly is Adobe’s own set of AI tools, built right into the Creative Cloud apps you probably already know—Photoshop, Illustrator, and Adobe Express. There’s even a mobile app for your phone. What makes Firefly different is that it’s trained only on content Adobe has the
LLM-as-a-judge: the measurement problem
You've built something and you need to know if it works. So you do what's sensible—you ask an LLM to grade it. Factual accuracy, code quality, agent outputs. The machine judges the machine, and you get a number you can act on. Except that number
Claude Opus 4.5: effort control
Claude Opus 4.5 is the newest brainchild from Anthropic, the folks behind the Claude language models. Think of it as their latest and smartest tool for handling really complicated tasks—like having an assistant who can juggle lots of jobs at once, and still keep everything running smoothly. So,
Apps SDK Brings Custom UI to ChatGPT
This week's edition covers building custom interfaces in ChatGPT, Google's Veo 3.1 video generation with native audio, multi-turn agent evaluation, and monitoring agent reasoning.
OpenAI: apps inside ChatGPT
OpenAI has just launched something called the Apps SDK, and it’s a bit like giving developers a new set of building blocks for ChatGPT. Instead of just chatting, you can now create apps that live right inside the conversation, with their own custom look and feel. The SDK builds
Veo 3.1: native audio and reference controls
Veo is Google's latest attempt to teach computers how to make videos from scratch. Now in version 3.1, it's available for anyone willing to pay for early access, either through Google AI Studio or Vertex AI. You can choose between the regular version or a
Pydantic: Evals
Pydantic Evals is a tool for Python that lets you watch, step by step, how your AI agents go about solving problems. It’s made by the same people who built the popular Pydantic data validation library. What makes it interesting is that it doesn’t just check if your
LangSmith: multi-turn evaluation
Imagine you’re chatting with an AI, asking it to help you book a flight. It might give you the right answer to every single question you ask, but somehow, you still end up without a ticket. That’s where multi-turn evaluations come in. Instead of just checking if each
Give Claude Memory and Skills via API
This week's edition covers Anthropic's new memory and Agent Skills APIs for building agents, Karpathy's transparent LLM training pipeline, on-device inference with Windows ML, and circuit-based interpretability tools that cut data requirements by 150x.
Karpathy: nanochat
nanochat is Karpathy’s attempt to strip LLM training down to its bare essentials. It’s about 8,000 lines of code, and it’s designed to be read and understood, not just run. Unlike the big, complicated frameworks you find in production, this one is all about showing you
The health tech paradox
Picture this: you buy a shiny new health gadget that claims it will look after you, no effort required. It sounds like the dream. But there’s a problem. Even the most hands-off technology still asks something from you. Mild cognitive impairment, or MCI, is when your memory and thinking
Turns out, 'AI for everyone' was not the winning move
More AI everywhere doesn’t mean better results. Small teams of AI experts beat mass licenses. Attackers now use AI to run cybercrime at machine speed. Leaders can’t keep up. Tools still forget us. The winners? Focused expertise and hybrid AI, not AI everywhere.
Tsinghua: focused AI expertise
Imagine you have a bunch of teams, some with AI, some without, and some where everyone gets their own AI. Researchers ran a big experiment with over 400 people to see what actually happens when you mix and match humans and AI in different ways. Here’s what they found:
Chulalongkorn: AI collaboration limits
Imagine you’re working with an AI tool, hoping it will be a real partner, not just a fancy calculator. That’s what the Human-AI Handshake Framework set out to test. Researchers at Chulalongkorn University looked at popular tools like GitHub Copilot, ChatGPT, and Adobe AI to see if they
UK Ministry of Defense: the AI leadership gap
The UK Ministry of Defense is running over 400 AI projects, each one watched over by a Responsible AI Senior Officer. These are the people meant to keep things ethical. The rules are all there: fairness, accountability, human oversight. The idea is to stop things like accidental escalations, messy procurement,
Hybrid AI: picking your battles
Imagine trying to simulate the whole Milky Way—100 billion stars—on a computer. Normally, this would take longer than a human lifetime. But a Japanese research team found a clever shortcut. Instead of throwing out the whole physics simulation and replacing it with AI, they used AI to skip
Anthropic: AI-powered cyberattacks
Imagine this: In September 2025, Anthropic—the folks behind Claude—caught something that sounds like science fiction. A Chinese state-backed group managed to trick Claude into launching cyberattacks, barely needing any humans to steer the wheel. Here’s the wild part: the attackers let AI do almost all the work—
Europe Pivots from Rules to Reality
This week's edition covers the EU's €1 billion push to actually use AI, the hardware crisis making AI less accessible, and the UK's regulatory sandbox experiment—three stories about the gap between AI policy and practice.
EU: the Apply AI Strategy
The European Commission has just launched its Apply AI Strategy, and this time, it’s not about more rules—it’s about getting AI out into the real world. They’re putting about €1 billion on the table, spread across eleven key sectors, using programs like Horizon Europe and Digital
The desktop AI gap
Imagine you’re a developer, and you want to try out the latest AI model on your laptop. Not so long ago, you could just download it and get to work. But now, these models have grown so huge and complex that your trusty computer just can’t keep up.
Turns out, 'AI for everyone' was not the winning move
More AI everywhere doesn’t mean better results. Small teams of AI experts beat mass licenses. Attackers now use AI to run cybercrime at machine speed. Leaders can’t keep up. Tools still forget us. The winners? Focused expertise and hybrid AI, not AI everywhere.
Copilot Now Speaks Fluent Google—Your Lost Files Are Saved
This week's edition covers Microsoft Copilot's cross-platform search and memory features, ChatGPT's new voice-and-text integration, Google Gemini 3's coaching capabilities for students and creators, free ChatGPT access for teachers, and Claude's extended conversation threads.
Apps SDK Brings Custom UI to ChatGPT
This week's edition covers building custom interfaces in ChatGPT, Google's Veo 3.1 video generation with native audio, multi-turn agent evaluation, and monitoring agent reasoning.
Europe Pivots from Rules to Reality
This week's edition covers the EU's €1 billion push to actually use AI, the hardware crisis making AI less accessible, and the UK's regulatory sandbox experiment—three stories about the gap between AI policy and practice.
OpenAI Builds a Browser Around ChatGPT
This week's edition covers OpenAI's new Atlas browser that combines ChatGPT with web browsing and memory, and Adobe Firefly's AI-generated soundtracks and voiceovers for commercial projects
Give Claude Memory and Skills via API
This week's edition covers Anthropic's new memory and Agent Skills APIs for building agents, Karpathy's transparent LLM training pipeline, on-device inference with Windows ML, and circuit-based interpretability tools that cut data requirements by 150x.
Your AI Now Does the Reading, Talking, and Data Entry for You
This week's edition covers AI tools that handle the busywork—NotebookLM transforms documents into narrated videos, ChatGPT creates editable diagrams from sketches, Windows Copilot goes voice-first, and Notion's AI Agent automates hours of data wrangling.