ChatGPT Is Going 18+ NSFW

In partnership with

Hey there,

A few things happened this week that seem unrelated but aren't:

A tiny AI model beat OpenAI's best at reasoning tasks. China's robotics industry is moving faster than anyone expected. A company just halved its workforce using AI, without firing anyone.

And then there’s two big news from OpenAI.

Let’s talk more about them and why they matter.

Smaller AI Models Are Getting Scary Good

A research team built an AI model called the Hierarchical Reasoning Model (HRM) with just 27 million parameters. To put that in perspective, GPT-4 has over a trillion parameters. This is tiny.

But this tiny model scored 40.3% on the ARC-AGI-1 benchmark, one of the toughest reasoning tests for AI. OpenAI's o3-mini-high? 34.5%. Claude 3.7? Just 21.2%.

The really crazy part: they trained it on only 1,000 examples per task. Not millions of examples. Not billions. One thousand.

How? They used brain-inspired hierarchical processing.

The model has two parts working at different speeds, a high-level module that does strategic planning (the "slow thinking") and a low-level module that handles execution (the "fast thinking"). Just like how humans actually think.

Open Source Is Winning

China's AI robotics industry is outperforming the United States and pretty much everyone else right now. And their strategy is the opposite of what Western tech companies are doing.

While American companies keep their AI models locked down like state secrets with proprietary code, closed APIs, everything guarded, Chinese robotics firms are open-sourcing everything. Robot designs, training algorithms, control systems, and datasets. All of it, publicly available.

You'd think giving away your technology would put you behind. Instead, it's making them move faster.

When you open-source your work, thousands of developers start improving it for free. Bugs get found and fixed in days, not months. New features appear that the original team would never have had time to build. Community contributions accelerate development in ways that closed development cycles just can't match.

The result? China's robotics deployment is happening at massive scale while Western companies are still protecting their IP and moving slower because of it.

The pattern is becoming clear: in fast-moving technology races, open beats closed. Not because open is more idealistic, but because it's faster.

OpenAI's Neutrality Problem

Meanwhile OpenAI recently published a research paper with a clear message: they're making ChatGPT "politically unbiased." Their stated goal is that "ChatGPT shouldn't have political bias in any direction."

Sounds straightforward, right? But read the actual paper and something else emerges.

They're not measuring whether ChatGPT provides accurate, factual information. Instead, they're measuring five things:

• Does it act like it has personal political opinions

• Is it amplyfying your emotional language

• Is it giving one-sided coverage of contested topics

• Whether it dismisses certain viewpoints

• Whether it refuses to engage with political topics

What they found is revealing. ChatGPT has a sycophancy problem, it's too eager to validate whatever political viewpoint you bring to the conversation. Ask it a charged liberal question, and it mirrors that energy. Conservative prompt? Same thing. The model is basically an overeager yes-man.

The training data and human feedback that shaped ChatGPT have baked-in preferences. This isn't speculation, it's in OpenAI's own research.

Their solution? Train ChatGPT to stop acting like an opinionated conversation partner and behave more like a neutral information tool. Stop validating users' views. Stop mirroring their emotional language. Be less like a person, more like a tool.

ChatGPT’s “Adult Mode”

Now here's where it gets interesting.

The same week OpenAI publishes research about making ChatGPT more neutral and tool-like, CEO Sam Altman announces something completely different.

By December 2025, ChatGPT will allow verified adults to have personality modes, including 18+ nsfw conversations.

Altman explained that they made ChatGPT restrictive to protect at-risk users, but that hurt the experience for everyone else. Now, with age-gating and better safeguards in place, they're loosening up.

Want ChatGPT to act like a friend? Use tons of emoji? Have mature, adult conversations? It'll do all of that, but only if you explicitly ask for it. You have to opt in.

So ChatGPT will be a neutral, objective information tool by default... but flip a switch and it becomes your personalized AI companion.

The whiplash is stunning. 

The contradiction reveals something important: there's no such thing as truly neutral AI. Every AI system embeds values through its training data, reward models, and design choices. OpenAI isn't solving the neutrality problem, they're managing multiple versions of the same product for different use cases.

The Workforce Shift Is Already Here

While everyone's debating AI's future impact on jobs, Klarna just showed us what's already happening.

Klarna went from 7,400 employees to 3,000. That's more than half their workforce gone.

But here's what's remarkable: they didn't do mass layoffs. They used AI to boost productivity so much that when people left naturally, retirements, resignations, people moving to other jobs, they simply didn't replace them.

Their AI customer service system now handles the work that previously required 700 full-time employees. The result? $40 million in improved profit margins annually.

CEO Sebastian Siemiatkowski admitted what most tech leaders won't say publicly: AI is replacing jobs at scale, and other CEOs are sugarcoating the impact.

This isn't isolated. Recent reports show that 41% of businesses are reducing their workforce due to AI advances. Gen Z is facing what some are calling a "job apocalypse", entry-level positions are disappearing faster than new ones are being created.

Klarna's approach, attrition instead of layoffs, is actually the smart play. It preserves morale, avoids severance costs, and gives the company time to figure out what work actually requires humans versus what AI can handle.

But make no mistake: the work is still disappearing. Just more gradually.

What This Means For You

Here's what matters for you:

• Stop assuming you need the biggest AI model. Most business tasks—writing emails, summarizing documents, analyzing data, customer support—work fine with smaller, cheaper models. 

• Consider open-sourcing strategically. If you've built internal AI tools that aren't your core competitive advantage—prompt libraries, wrapper scripts, automation workflows—put them on GitHub. You'll get free improvements from the community and build your reputation. Start using open-source models like Llama or Mistral as foundations instead of paying for every API call.

• Be explicit about your AI's behavior. If you're building AI products, don't claim neutrality. Document what your AI will and won't do, and why. Test with diverse inputs—politically charged, emotionally loaded, edge cases—because you'll find biases you didn't expect. If you sell to enterprises or government, start documenting how your AI handles sensitive topics now.

• Plan for workforce changes now. Look at your company's repetitive work—customer support, data entry, scheduling, basic analysis. These all have mature AI solutions. Calculate your natural attrition rate. If 10-15% of people leave annually, you might not need layoffs. Just don't refill positions where AI can handle the work. Most important: talk to your team about this early. Explain what's changing and offer retraining for new skills.

• Calculate your actual AI costs. Track what you're spending per task. You'll probably find you're overpaying for most of it. Small models can replace 80% of your expensive API calls.

The companies winning with AI aren't the ones spending the most. They're the ones making smart, specific choices about where and how to use it.

What's your first move?

- Aashish

Find your customers on Roku this Black Friday

As with any digital ad campaign, the important thing is to reach streaming audiences who will convert. To that end, Roku’s self-service Ads Manager stands ready with powerful segmentation and targeting options. After all, you know your customers, and we know our streaming audience.

Worried it’s too late to spin up new Black Friday creative? With Roku Ads Manager, you can easily import and augment existing creative assets from your social channels. We also have AI-assisted upscaling, so every ad is primed for CTV.

Once you’ve done this, then you can easily set up A/B tests to flight different creative variants and Black Friday offers. If you’re a Shopify brand, you can even run shoppable ads directly on-screen so viewers can purchase with just a click of their Roku remote.

Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now