😸 Is ChatGPT Pulse OpenAI's first attempt at a social network??

PLUS: AI aces CFA finance exams?!

Welcome, humans.

Professor Ethan Mollick just casually asked GPT 5 Codex to recreate a lost 1990s video game for him… and it actually worked.

SimRefinery was a real Maxis simulation from 1993 that taught players how to run an oil refinery (because apparently SimCity wasn't stressful enough). The game disappeared into the digital void decades ago, but Mollick fed ChatGPT just an article and a screenshot, then sat back while AI built him a playable prototype.

Here's the wild part: Mollick has been building educational games and simulations for nearly 20 years with entire teams. Now he's getting the same results by just... asking nicely?

You can actually play his AI-generated version right now if you want to experience what it feels like to manage an oil refinery… without the day-to-day on-the-job stresses of Billy Bob Thornton’s character in Landman.

Here’s what happened in AI today:

ChatGPT Pulse and Meta Vibes are squaring off to capture your feed.
Meta released Code World Model, a 32B-parameter coding AI
Grok 4 expanded to 2 million token context window.
AI models passed the notoriously difficult CFA Level III exam.

Advertise in The Neuron here

ChatGPT now starts the conversation for you.

It’s like it already knows what we want!

NEWS BRIEF: Here’s Everything to Know About ChatGPT Pulse

OpenAI just flipped the script on how we interact with ChatGPT. Instead of you constantly prompting it for answers, it just launched Pulse, a new feature that wakes up every morning with personalized research it did for you overnight.

Think of it as your own AI research assistant who preps a daily briefing just for you (kinda like us newsletters! Sure, it’ll be customized just for you, but it won’t provide witty sardonic asides like this one… so you won’t replace us, right?? RIGHT?!).

Here’s how to make Pulse actually work for you:

Don't be passive. Early tester Olivia Moore (of A16z) found the real magic happens when you actively tell Pulse what you want. Use the “curate” button to give it specific instructions like, “focus on professional tennis updates,” or “find next steps for my triathlon training.”
Connect your apps. For the best results, you can connect your Gmail and Google Calendar. This allows Pulse to do things like draft a meeting agenda or remind you to buy a birthday gift. It's off by default, so you have to opt-in.
Give it feedback. Use the thumbs up/down on each card. OpenAI says this feedback only trains your version of Pulse, making it more useful over time.

The bigger picture is clear. This is OpenAI's play to become your first click of the day, competing directly with news apps and social media for that prime morning real estate. Uniquely, it’s a finite feed, designed to end, not for doom-scrolling; which conveniently also creates a perfect surface for future ads. Insiders have reported OpenAI is exploring ad-leadership hires, and Pulse looks like the ideal place to start.

This move also comes at the same time Meta launched Vibes, a new feed of AI-generated videos. Together, they signal the next big thing in AI: the “feed.” As Sam Altman and others have intimated, the “chat” form factor for AI is saturated. So now “the feed” form factor has begun in earnest.

But where Vibes is a public, creative, TikTok-style social feed, Pulse is a private, productive, anti-social feed designed to help you (not to get you scrolling).

What to do: If you're a ChatGPT Pro user, you should have access on mobile this week (Plus users get it later). Start by actively curating it with your professional and personal goals. This is the first step toward a future where AI agents proactively manage tasks and even purchases for you. The race for the “proactive assistant” is on, and OpenAI just made a huge move to the front; we’ll see if ppl like it! Read our full report here.

FROM OUR PARTNERS

Vibe, then verify

More code is being written than ever, but teams are not moving faster. With AI now producing 30% of the code at major tech companies*, engineering velocity has only risen by 10%. Why the gap? Because every line of AI-generated code still needs to be reviewed, fixed, and secured before it’s ready for production.

Sonar helps close that gap by automatically checking AI-generated code for quality, security, and maintainability. You get the speed of AI without drowning in bugs or technical debt, and your team can finally move as fast as it codes.

Learn how Sonar verifies AI code

Prompt Tip of the Day

Ryan Carr of Moodboard (a “vibe marketing” newsletter) just shared the ultimate subject line writer prompt, and TBH, it’s quite simple but effective. If you opened today’s email, you can thank Ryan! So if it worked on you, it’ll probably work FOR you. The whole thing is too long to include, but we’ll give a TL;DR of Ryan’s process:

The AI prompt that generates 8 different subject line options using proven psychological triggers.
It uses 8 Strategies: Curiosity gap (this is the big one), benefit-focused, urgency, question-based, data-driven, contrarian, story-based, direct statement
His guidelines: Keep it under 50 characters, avoid spam words, use power words, be conversational not corporate.

The key Insight: “Prompt for options”: instead of accepting one AI output, have it give you multiple angles to choose from (good practice for all prompting, btw).

Here’s our recommendation on how to implement this:

Step 1: Copy the prompt template from the article into the instructions of a project folder / gem on Gemini / customGPT (project folder is simplest!).
Step 2: Replace where it says “[Paste your newsletter/email content here]” with “the content to use will be pasted attached in the context window.”
Step 3: Paste the full text of your email content in the context window and hit enter.
Step 4: Pick your favorites, or ask for more and give it notes on what you want more / less of.

Treats to Try

*Asterisk = from our partners (only the first one!). Advertise in The Neuron here.

*Typing wastes time. Flow makes voice the faster, smarter option everywhere. Flow users save an average of 5–7 hours every week — that’s full workdays back every month. Give your hands a break ➜ start flowing for free today.
Doraverse unifies 15+ models and 50+ integrations so you can run office workflows in one workspace—starts at $15/mo.
Oreate AI helps you draft essays, reports, and slides with citations and charts.
Webhound builds web-sourced datasets from natural-language prompts and exports to CSV.
Burnt handles your food business's order emails and admin work so your team stops drowning in spreadsheets, and has already processed $10M+ in monthly transactions (raised $3.8M).
Uncommon generates 1K+ company name ideas for you, checks if their domains are available, and shows you the 10 best options, then builds your complete brand including logo, website, and pitch deck from your description (paid only rn @ $25/month).

Around the Horn

OpenRouter’s leaderboard of what models are being used most this week is fascinating for tracking dev usage: 4.37 trillion tokens used just this week, just through OpenRouter. The scale of this is WILD.

Grok 4 fast now has a 2M token context window (roughly equivalent to 5K-6K pages of text, several full-length novels, or 200K to 670K lines of code).
Meta FAIR released Code World Model (CWM), a 32-billion-parameter open-weights language model trained on Python execution traces and Docker environments that achieved 65.8% on SWE-bench Verified coding tasks.
Mindstate Design Labs completed the first human trial toward an AI-guided neurotech platform for mental states (put another way: its a psychadelic without the trip).
Inception Point AI apparently makes thousands of automated podcasts for $1, slotting programmatic ads inside, and even if only 20 ppl listen to a show, it makes money; oh, and they’ve already made 160K episodes…
Magical Toys sold 1K AI toys last year, targeting 100K this year; their toy Bondu chats with your kid by name, telling stories and playing conversation games instead of screen time (for ~$200).
Google released a new Chrome DevTools MCP that connects your AI coding assistant to Chrome's browser tools so it can debug your website, check performance, and fix issues by actually seeing what happens when code runs.
Simon Willison flagged a new cross-agent escalation issue where overlapping tools trigger unsafe permissions.
There is a benchmark for AI in finance, and as it turns out, frontier models like o4-mini can now pass the notoriously difficult CFA Level III in minutes with a 79% (paper).
Which is better, Codex or Claude Code for coding with AI? Steve from Builder who has become a bit of a viral AI coding influencer made the switch to Codex, arguing you get more value out of your $20 ChatGPT Plus sub w/ Codex.
Julian Schrittwieser writes that we’re failing to understand the exponential, comparing current dismissals of AI progress to how people missed COVID-19's exponential spread, while citing the GPDVal paper that showed Claude Opus 4.1 is actually performing above the predicted exponential trend and already nearly matching human expert performance across 44 different occupations.

FROM OUR PARTNERS

💼 Want to build a 6-figure AI Consulting career?

The AI consulting market is about to grow by a factor of 8X – from $6.9 billion today to $54.7 billion in 2032. But how do you turn your AI enthusiasm into marketable skills, clear services and a serious business?

Our friends at Innovating with AI have trained 1,000+ AI consultants – and their exclusive consulting directory has driven Fortune 500 leads to graduates.

Enrollment in The AI Consultancy Project is opening soon – and you’ll only hear about it if you apply for access now.

Click here to request access to The AI Consultancy Project