- Future Tools
- Posts
- OpenAI's mental health plan, upacked
OpenAI's mental health plan, upacked
Plus: The future of human feedback
Welcome back! AI can fake almost anything these days…except accountability. OpenAI’s new Sora model is proving just how broken deepfake detection really is. Even though Sora’s videos are tagged with invisible “Content Credentials,” those labels rarely show up once clips hit TikTok, YouTube, or X.
Some food for thought today: If AI can make anyone say anything, who decides what’s real anymore?


OpenAI Faces Mental Health Issues

OpenAI CEO Sam Altman (Via Aaron Schwartz/Sipa/Bloomberg / Getty Images)
Over 1 million people per week talk to ChatGPT about suicide, according to new OpenAI data. That’s 0.15% of its 800 million weekly users—a small percentage, but still far too many people. And? OpenAI says a similar share of users show “heightened levels of emotional attachment to ChatGPT.”
Here’s how OpenAI is dealing with this stark reality ➡️
OpenAI says it’s working with 170 mental health experts to improve ChatGPT’s responses.
It says the latest version of GPT-5 is 65% better at handling sensitive conversations.
The company is also building systems to detect underage users and flag signs of emotional dependence.
A question of ethics: OpenAI’s efforts are under the microscope after 1) a lawsuit from the parents of a 16-year-old boy who shared suicidal thoughts with ChatGPT before his death by suicide 2) warnings from state regulators and 3) CEO Sam Altman’s recent decision to relax ChatGPT’s guardrails.
Why it matters: This challenge goes beyond detecting crises to deciding what responsibility an AI should bear for them. For rapidly growing AI companies like OpenAI, can safety keep up with scale?
Mercor’s $10B Bet on Human Feedback
Mercor just raised $350 million, hitting a $10 billion valuation. But it’s not building another chatbot—it’s hiring humans to train them.
From hiring to training: The Thiel Fellow–founded startup began as an AI-driven hiring tool before pivoting to what it now calls “human intelligence for machine learning.” Its 30,000+ contractors earn a collective $1.5 million per day refining AI models through expert feedback, context sharing, and annotation.
Timing is everything: After Meta’s $14.3 billion stake in Scale AI torpedoed its neutrality, Mercor swooped in to fill the gap. The new funding round, led by Felicis, will go toward expanding Mercor’s talent network and speeding delivery of training data to labs like Google and OpenAI.
Why it matters: Mercor’s 11-figure valuation makes one thing crystal clear: The future of AI still depends on human judgment.
Qualcomm’s Turning Phones Into AI Powerhouses
Qualcomm is going after Nvidia. The chipmaker announced two new processors—the AI200 and AI250—designed to run AI models instead of just training them.
Built for deployment, not labs: Using the same Hexagon architecture found in its smartphone chips, Qualcomm’s new AI processors can cluster up to 72 per rack, functioning like scaled-down data centers. The AI200 ships with 768GB of RAM, while the upcoming AI250 promises a leap in power efficiency—ideal for edge computing where speed and energy matter more than brute force.
Big picture: This isn’t just about catching up to Nvidia—it’s about decentralizing AI. Qualcomm’s approach pushes intelligence out of the cloud and closer to the devices we actually use, from factory sensors to automated cars.


Your Marketing Teammate That Never Sleeps

Toffu AI
Toffu AI automates campaign management, content creation, and data analysis across tools like Meta Ads, LinkedIn, and Google Analytics—all through natural-language chat.
How you can use it:
Get AI-generated performance insights in plain English
Automate repetitive marketing tasks and reporting
Orchestrate cross-platform campaigns in one place
Pricing: Free and paid

Turn Text Into Humanlike Voices

Fish Audio
Fish Audio is a voice generation tool that actually sounds natural. It offers 200,000+ voice models, fine-tuned expression tags, and fast cloning from short samples.
How you can use it:
Generate narration or ads with lifelike tone
Clone your own voice for videos or podcasts
Integrate realistic speech into chatbots or assistants
Pricing: Free and paid


Jobs, announcements, and big ideas
Nvidia CEO Jensen Huang’s message in his keynote speech at the GTC conference? Nvidia is angling to make its tech integral to everything, from cell phone towers to self-driving cars. The name of the game is ubiquity.
PayPal integrates ChatGPT for instant checkout and smarter shopping.
Adobe Firefly debuts new AI models for sound, video, and design at MAX 2025.
Senators move to ban AI chatbot companions for minors.
Microsoft and OpenAI extend their partnership to scale responsible AI.
OpenAI chair Bret Taylor shares a roadmap for AI designed to benefit everyone.
NVIDIA and Oracle will build America’s largest AI supercomputer for the DOE.
Google launches a Gemini Home assistant with 100 experimental features.
NVIDIA unveils IGX Thor for real-time AI in robotics and healthcare.


35 ways to supercharge your workflow with NotebookLM. Watch along as I break down my favorite free tips, tricks, and use cases.

That’s a wrap! See you Friday.
—Matt (FutureTools.io)
P.S. This newsletter is 100% written by a human. Okay, maybe 96%.
