😺 ChatGPT became your tutor

PLUS: NotebookLM has video now & GPT 4o-level AI runs on laptop +

Welcome, humans.

According to Bloomberg, Microsoft and OpenAI are in heated negotiations to restructure their ~$14B partnership—and it's getting messy.

Here's the drama: their current deal has a bizarre clause where Microsoft loses access to OpenAI's technology once the startup hits “artificial general intelligence” (AGI). That trigger could happen when OpenAI reaches $100B in total profits OR when OpenAI's board simply decides they've cracked AGI.

Microsoft is basically saying “uh, no thanks” to potentially losing access to the crown jewel AI models they've been bankrolling for years.

The stakes couldn't be higher. These former best friends now compete for the same customers, OpenAI wants more independence (and revenue), and there's a year-end deadline looming that could affect additional funding from SoftBank.

The talks could wrap up in weeks, but if Microsoft doesn't like the terms? They're perfectly happy to stick with the current contract until 2030 and watch this whole thing play out in court.

Meanwhile, somewhere on TikTok… this giant AI dog is taking up space in 250M people’s minds rent free…

Here’s what you need to know about AI today:

ChatGPT launched tutor mode that refuses to give direct answers.
Anthropic is raising $3-5B at $170B valuation.
Apple lost fourth AI researcher to Meta in one month.
NotebookLM has video overviews now.

Advertise in The Neuron here.

ChatGPT just got a tutor mode that refuses to do your homework for you.

ChatGPT is officially DONE being that friend who always just gives you the answer instead of helping you figure it out.

OpenAI just launched “study mode” in ChatGPT: a new learning experience that guides you step-by-step through problems instead of spitting out quick solutions. It's available right now for all logged-in users (Free, Plus, Pro, and Team), with ChatGPT Edu (ironically) getting access in the coming weeks.

This reminds us of those “In mother Russia” memes from awhile back… “In Study Mode, ChatGPT prompts YOU.”

Here's how it works: instead of asking “What's 2+2?” and getting “4,” study mode asks questions like “What do you think happens when you add these numbers?” and “Can you walk me through your thinking?” It's like having a patient tutor who won't let you off the hook that easily.

The key features include:

Socratic questioning: It guides you with hints and follow-up questions rather than direct answers.
Scaffolded responses: Information broken into digestible chunks that build on each other.
Personalized support: Adjusts difficulty based on your skill level and previous conversations.
Knowledge checks: Built-in quizzes and feedback to make sure concepts actually stick.
Toggle flexibility: Switch study mode on and off mid-conversation depending on your goals.

Try study mode yourself by selecting "Study and learn" from tools in ChatGPT and asking a question.

OpenAI built this with actual teachers, scientists, and education experts after realizing ChatGPT had become one of the world's most-used learning tools. The goal? Make sure students actually learn something instead of just copying answers.

We wrote a whole DEEP DIVE on this issue here, in case you missed it.

College students who tested it early gave rave reviews. One called it “a live, 24/7, all-knowing office hours.” Another said it helped them finally understand a concept they'd struggled with for months after a 3-hour session.

Why this matters: This is OpenAI's first real attempt to tackle the “cheatGPT” problem that's plagued the education system since ChatGPT launched. Instead of trying to detect and block academic misuse, they're making the tool actively better for learning. It's a smart pivot that could reshape how we think about AI in education.

The company hints at bigger features coming: better visualizations, goal tracking across conversations, and deeper personalization. They're also partnering with Stanford's SCALE Initiative to run longer-term studies on how AI affects actual learning outcomes.

On the learning feature front, OpenAI is definitely behind. Google’s NotebookLM is the tool to beat here, and it just got Video Overviews that create videos out of your study materials in addition to its podcasts. Not only that, Google has LearnLM too.

Ethan Mollick, the world’s #1 AI and education promoter, highlighted that Study Mode is not a perfect system, but a step in the right direction. And Simon Willison shared the supposed system prompt for Study Mode, so you can see how it works under the hood. And if you want to create Study Mode in Gemini or Anthropic, just copy the prompt into a new project folder’s custom instructions, and it’ll functionally work similar.

Our take: We think this tool could be even more prescriptive by building around HOW we learn: that is, hard-coding some level of retrieval, spacing, interleaving, and generation directly into the model with notifications for periodic “pop-quizzes” (or something similar) that encourages long term learning and growth.

With even more refinement, this could be very powerful. BUT… left to their own devices, will students actually choose this, or will they just default to regular GPT or NotebookLM? It depends on if they actually want to learn or not…

FROM OUR PARTNERS

CodeRabbit’s Free Context-Aware AI Code Reviews in your IDE

Code Rabbit is the AI code review platform purpose built to speed up code reviews and improve code quality. Integrating seamlessly into git workflows and existing toolchains, it delivers context-aware reviews and supports all programming languages.

Now they are excited to deliver AI code reviews directly in VS Code, Cursor, and Windsurf–letting you and your team (vibe) code with confidence and review while keeping in flow.

Even better: code reviews in the IDE are free! That’s right: line-by-line reviews and one-click fixes, all in your IDE at no cost to you.

Install the extension and start vibe checking your code today!

Prompt Tip of the Day

Wikipedia editors created a master list of AI writing red flags—and you can use it to fix your prompts (and your outputs) to avoid these annoying quirks.

The guide reveals telltale AI phrases like “serves as a testament,” “rich cultural heritage,” “moreover,” and “it's important to note” (plus so much more).

You can actually take this list, copy it into a smart reasoning AI like ChatGPT-o3 or Google’s Gemini 2.5 Pro and ask it to turn each unique item on the list into an “editor rule” to remove as part of your editing process.

“Turn each negative into an active verb positive—instead of 'avoid serves as a testament,' write 'replace importance-inflating phrases with direct statements.’”

Apply these rules to any AI content—emails, reports, creative writing—either when editing your draft or built into your original prompt.

Check out all of our Prompt Tips of the Day from July here.

P.S: Completely new to AI? Start here!

Treats To Try.

*Asterisk = from our partners. Advertise in The Neuron here.

*Incogni erases your sensitive data—like addresses and phone numbers—from the web. The internet remembers, but you can delete. Get 55% off with code NEURON.
Qwen3 30B A3B 2507 delivers GPT-4o-level conversations that run completely offline on your laptop—meaning no internet required, no data sent to external servers, and no per-use costs for enterprise-quality AI that previously only existed in the cloud.
1. To run it, download LM Studio, search for “Qwen3 30B A3B 2507”, then load the model in the chat tab with Temperature=0.7/TopP=0.8 settings. Oh, and if you want to test drive it on the cloud, try this.
Qwen3-235B-A22B-Thinking is Alibaba’s thinking AI that solves your most complex reasoning problems with transparent thinking steps, supporting a 256K token context window for deep analysis of documents and code.
Runway Aleph lets you edit videos by simply describing what you want—remove cars, swap backgrounds, or change camera angles—maintaining perfect consistency across frames.
Flyde lets you visually build complex AI workflows right inside your code repository, so you can quickly prototype and iterate on backend logic while letting non-technical teammates contribute to AI systems—free to try.
Doco streamlines your document creation in Microsoft Word with prompt-based editing, source tracking, and security features designed for sensitive business documents—launching in 2025.
Nitrode helps you build playable 3D games in a day using visual tools and instant AI-generated assets, all without coding experience.
Toolplay unifies all your AI creation needs—images, videos, voiceovers, and writing—under one account using flexible credits instead of multiple subscriptions.
Claude Code Router sends your coding requests to any AI model you choose, not just Claude, so you can code with Claude Code even without using Claude.
No-hype AI is a resource for learning about how to code with AI and the top tools without any… you guessed it… hype.

P.S: After Monday’s latest meme, M.W brought to our attention the existence of GPT-powered Clippy (which we knew about, but low key forgot)… still think it’s a major miss Microsoft hasn’t legit done this themselves though. Where y’all at, Microsoft? Memes are the new moat!

See our top 51 AI Tools for Business here!

Around the Horn.

The internet’s favorite unofficial video benchmark, Will Smith eating spaghetti, is basically 100% solved.

Google enhanced AI Search with Canvas for organizing research across sessions and Search Live for real-time visual understanding.
Anthropic is raising $3-$5B at a $170B valuation, nearly 3x its March valuation.
AI video startups Luma and Runway pivoted toward robotics and autonomous vehicles, expecting these markets to become major revenue drivers for their “world models” that generate training data for robots.
- Related: Disney and Netflix are experimenting with Runway, and Runway's 2025 AI Film Festival expanded to include IMAX screenings in multiple cities, offered substantial cash prizes up to $15K, and featured a jury of prominent industry figures for short films created using generative AI tools.
Yelp launched AI-stitched videos that automatically create promotional narratives from user content for local businesses.
Apple lost its fourth key AI researcher, Bowen Zhang, to Meta's superintelligence team within a month.
DOGE created an AI tool targeting elimination of 100K federal regulations within Trump's first year in office.
New research shows how simply adding irrelevant cat trivia to math problems can increase AI error rates by up to 700% in models like DeepSeek V3.