😺 Sam drops GPT-5 date

PLUS: OpenAI identified "toxic personas" in GPT?!

Welcome, humans.

Happy Juneteenth, everybody! Quick housekeeping note: no newsletter tomorrow, so we’re doing Intelligent Insights today instead of Trivia (we had A LOT of insights this week).

So, you probably saw the headlines by now, but buried in OpenAI CEO Sam Altman’s podcast bro down with his brother Jack, he admitted that Meta just tried to steal OpenAI's top engineers with $100M signing bonuses. Yes, you read that right—nine figures just to switch teams.

And yet, according to Sam, not a single (important) person took the bait. Either OpenAI's equity packages are absolutely insane, or they've built something even more valuable than money: a culture where people actually want to work.

Though let's be real, there’s a reason people SELL their companies for $100M… with that kinda money, who wants a job??

Why is Meta willing to pay so much? While most see OpenAI competing with Google, Sam says Zuckerberg fears you'd rather talk to helpful AI than doom-scroll angry feeds—a war for attention he's trying to win with cash.

The whole thing reminded us of that one scene from Mad Men…

Of course, real stans know Don does indeed think “at all” about Ginsberg…

ICYMI: Our revamped podcast is live! Grant and Corey tackle the question: Will AI wipe out half of white-collar jobs? We’ve got a great guest (Microsoft researcher Alexia Cambon), and cover 5 concrete strategies to future-proof your career. Watch on YouTube or Spotify.

Want to reach 550,000+ AI professionals through podcast sponsorship (or newsletter ads)? Just fill out this quick form.

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

Sam Altman puts a date on GPT-5.
MidJourney released a new video model.
The new Pope sees AI as a threat to human dignity.
OpenAI released a paper that identified “toxic personas” in ChatGPT.

Advertise in The Neuron here.

Sam Altman just revealed OpenAI’s roadmap, and it’s all about making ChatGPT simple again.

Sam Altman on AGI, GPT-5, and what’s next — the OpenAI Podcast Ep. 1

Man, Sam Altman’s really been on a podcast streak, hasn’t he? Today, OpenAI released a new, wide-ranging interview on all the company’s latest interests and issues.

In the interview, CEO Sam Altman laid out a roadmap focused on unifying the user experience, starting with the release of GPT-5 this summer.

What’s GPT-5? Have you used AI long enough to remember when ChatGPT was just… ChatGPT? No model picker, no juggling o4-mini-high versus GPT-o3 Pro, just one box to type in? Well, OpenAI is finally fixing its confusing product lineup, and GPT 5 will basically be a simplification (reunification?) of ChatGPT’s myriad models into one streamlined UI.

Here's the plan to end the model mess:

Altman said the next flagship model will “probably” arrive sometime this summer.
He admitted the current state of model choice is a “whole mess.” The goal is to get back to a simple progression (GPT-5, GPT-6) and ditch the complex variants (4, 4o, o4, oh for crying out loud!).
The future vision is a unified model that can handle everything seamlessly—from quick questions to complex, multi-step tasks using "reasoning" and agent-like tools like Deep Research. No more switching modes.

Altman explained this shift is happening because AI is moving beyond just giving instant answers. He's been surprised to find that for hard problems, users are “surprisingly willing to wait for a great answer.” This insight is driving the development of more thoughtful "reasoning models" that work more like a human expert, taking time to think before responding.

Our take: This simplification is a clear play to win over the mainstream user who doesn't want to learn a complex menu of AI options. By creating a single, smarter ChatGPT, they're making it easier to deliver on their bigger ambitions—like the agentic AI that can act as your personal assistant.

The long-term projects we've heard about before, like the massive Stargate supercomputer and the new hardware with Jony Ive, are all pieces of this puzzle, building the infrastructure and interface for this unified AI future.

Get ready for a simpler, more powerful ChatGPT that does the thinking for you—both in its answers and in choosing the right tool for the job.

Something to watch for: OpenAI says new, smarter models could increase “novice uplift”, or the ability for novices to do more dangerous things with high powered AI.

FROM OUR PARTNERS

Just Dropped: The INBOUND 2025 Agenda is Here

The INBOUND 2025 agenda is here and it's packed with innovation leaders. Join AI innovators like Dario Amoedi and Dharmesh Shah at INBOUND 2025 this September, where cutting-edge AI meets practical business applications.

Explore 200+ sessions including:

“Tech at the Edge: Strategies and Insights with MKBHD” with Marques Brownlee.
“Dwarkesh on AI's Future: Research-Backed Bold Predictions” with Dwarkesh Patel.

Through powerful keynotes, deep dives, breakout sessions, and strategic networking, INBOUND 2025 is your chance to build what’s next.

Prompt Tip of the Day

Try, try again (with a twist). If an approach isn’t working, rephrase your question or approach from a different angle. Sometimes asking "What should I have asked to get XYZ?" can even prompt the AI to give you the question and answer you need.

The wording and approach in a prompt can produce surprisingly different outputs, so a little experimentation can go a long way. Don’t be afraid to reword and retry in a new chat (or edit your prompt and rerun it). It's all part of the process!

Treats To Try.

*Anything marked with asterisks is sponsored content. Advertise in The Neuron here.

*Sesame is an online platform for brand expression. Graphics, motion, sound, and creative are templatized to accelerate ideation, production, and iteration. Increase speed-to-market and engagement. Plus, it’s fun!
MidJourney “V1” is the image generator’s first video model, and early reviews are blown away. Here’s how to use it.
Search Live turns Google Search into a voice conversation where you ask questions out loud and get spoken answers—try it here.
Arcee gives you tools to customize AI models for your apps and just released a 4.5B parameter model that runs on smartphones using only 3GB of RAM instead of expensive GPUs.
Liveblocks adds Figma-like collaboration and AI copilots to your web app with pre-built React components.
AgentX helps you build agent teams that handle your lead capture, customer support, and scheduling across WhatsApp, Slack, and your website.
Tila AI provides a visual workspace for multimodal projects with specialized AI agents for each media type
OpenAI released a “customer service agent” software kit for developers.

See our top 51 AI Tools for Business here!

Around the Horn.

OpenAI identified specific “toxic persona” features that control and even predict misaligned behavior (paper), and were able to toggle that feature up or down to steer models toward or away from misaligned behavior.
Pope Leo XIV made AI ethics central to his papacy, warning about threats to human dignity while advocating for a “third path” between self-regulation and government inaction.
California pushed forward multiple AI regulatory bills requiring transparency and bias testing, with employment regulations making both employers and AI vendors liable for discriminatory outcomes; meanwhile, the big tech giants lobbied for a 10-year federal ban on state AI regulations.

FROM OUR PARTNERS

RefineAI is Live!

RefineAI can.

No-code is history. React from scratch? That’s madness. Vibe coding is the future—but can generic AI builders really deliver full-scale internal enterprise apps?

Unfortunately, not yet. Unless it’s RefineAI.

Forget black-box outputs and zero maintainability. RefineAI uses domain knowledge and the open-source Refine framework (31K Stars) to generate production-grade internal React apps.

Start Building for Free

Intelligent Insights

Andrej Karpathy’s new “software is changing” talk is a must watch (or read—shout out to Donna Magi for sharing this!). The TL;DR is that “Software 3.0” (AI, a.k.a language models) is fundamentally changing software development from a 5-10 year learning curve to immediate accessibility for billions of people.
Dwarkesh Patel explains why AI needs continuous learning and context-stacking like humans to become truly useful at work.
Any AI asked for a random number will probably give you “27”, and there’s some compelling theories explaining this weird bias (like this video).
Check out bycloud's geeky deep-dive into how AI works locally and why “bitnets” could revolutionize AI development.
Apparently AI spontaneously develop object concepts that mirror our human brain regions, suggesting they discover how we understand the world.
Tech giants use AI as “intention extraction” tools to manipulate “what you want to want” through hyper-personalized sycophancy.
NYU's Leif Weatherby argues AI accidentally proves language comes from systematic relationships, not human cognition.
Gary Marcus rebuts the critics of Apple’s recent thinking paper who used ad hominem attacks while missing that today’s AI still can't handle the algorithmic complexity required for long context tasks.