Welcome back! Reddit is now using LLMs to fight spam. Great! Except for the fact that the spam was largely created by LLMs in the first place … Still, Reddit says it blocks 23 million spam views per day and catches about 25,000 new spam posts and comments daily. It also reduced users’ exposure to spam by 20% from January to March.
AI made the internet messier, and now platforms are using AI to clean it back up.
What Happens When the Regulator Becomes a Shareholder?
OpenAI just floated an idea that sounds great in theory but has me a little uneasy in practice.
According to a Financial Times report, OpenAI has proposed giving the US government a 5% ownership stake in the company worth roughly $42.6 billion based on OpenAI's most recent $852 billion valuation. Sam Altman's pitch is that if AI creates massive economic upside, the public should share in it, modeled loosely on the Alaska Permanent Fund (where the state's oil revenue gets invested and pays annual dividends to every resident).
Altman also wants other major AI labs—Anthropic, Google, Meta, xAI—to each contribute a similar 5% stake to the same public wealth fund.
A few context points that matter here: This isn't happening in a vacuum. The government has been ramping up pressure on AI companies for months. OpenAI's GPT-5.6 release was delayed at the administration's request, and as you’ll remember, Anthropic's Mythos and Fable models got export-controlled a few weeks ago.
This proposal reads a lot like OpenAI trying to trade equity for political goodwill.
My POV: I agree with the principle. If AI generates trillions of dollars of value across the economy, ordinary people should share in that upside. That's a good instinct.
But this specific version of the deal feels off.
The government is supposed to keep AI companies in check on behalf of the public. If Washington owns 5% of OpenAI, does it still have the same incentive to enforce safety rules when those rules might dent the company's valuation? Or does it now have a financial reason to look the other way, because faster growth means a bigger return on its stake?
— Matt
Mercor Hits $2B Revenue Run Rate
Via TechCrunch
Mercor hit more than $2 billion in gross revenue run rate in June, doubling its pace from earlier this year. This is the company that helps AI firms and enterprises hire contractors with domain expertise in fields like physics, finance, and medicine. Those experts answer questions, evaluate outputs, and create specialized data used to train and fine-tune AI models.
Why it’s growing: Talent is in demand from AI app developers and Fortune 500 companies that want to build new models or improve existing ones. The more companies try to make AI work in specialized fields, the more they need high-quality human data from people who actually know what they’re talking about.
Worth noting: Mercor reportedly pays 60% to 70% of its topline revenue to contractors, meaning its net revenue is likely closer to $600 million to $800 million. Still, the company is profitable on a free cash flow basis. CEO Brendan Foody said last year that Mercor was “the fastest growing company of all time” after scaling from a $1 million to $500 million run rate in 17 months. A month later, the company raised $350 million at a $10 billion valuation led by Felicis.
Why it matters: The AI boom is still pulling huge amounts of money into the picks-and-shovels layer. Models get the attention, but training data, expert labeling, evaluation, and fine-tuning are becoming massive businesses of their own. Mercor’s growth is a reminder that value still comes from humans teaching the machines.
UK Official: Britain Should Consider Regulating AI Models
A senior UK financial regulator said Britain should review whether large language models like ChatGPT, Claude, and Gemini should be regulated as general-purpose AI tools.
Consumer risk:
Financial Conduct Authority executive Sheldon Mills said more than a quarter of UK consumers trust tools like ChatGPT, Claude, and Gemini for financial advice, even though many do not realize those tools lack the same protections as regulated financial services.
Mills recommended that the FCA consider over the next three to six months whether to “secure and adapt” its regulatory perimeter by reviewing the scale, nature, and impact of general-purpose AI models that currently sit outside it.
The bigger picture: Regulators around the world are trying to figure out where AI fits inside existing rulebooks. The challenge is that general-purpose models are not banks, doctors, lawyers, or financial advisers—but consumers are increasingly using them that way. The hard part is deciding how much to regulate AI before it causes real harm, without slowing down the parts that could actually make services better.
AI Can Run Your Entire Life—But Should It?
With the explosion of AI technologies available, nearly every part of your professional—and personal—life can be automated. But some stuff still requires a human touch. The key is deciding which stuff.
Come to my webinar in partnership with Teachable, and you'll walk away with practical frameworks for:
Identifying high-impact AI use cases
Integrating AI into your existing workflows
Increasing productivity without sacrificing creativity or critical thinking
Ready to incorporate AI into your life without going overboard? I’ll show you how.
Compare Multiple AI Models at Once
Via Talkory.ai
Talkory.ai is a multi-LLM comparison platform that queries GPT, Claude, Gemini, Grok, and Perplexity at the same time, then shows side-by-side responses and generates a ranked consensus answer with confidence scores.
How you can use it
Compare answers across multiple AI models
Reduce hallucinations by cross-checking outputs
See which model performs best for research, writing, or coding
Generate consensus answers for teams or clients
Pricing: Free and paid plans available
Screenshots for Coding Agents
Via AgentScreenshots
AgentScreenshots is an AI-ready screenshot capture tool that lets coding agents like Claude Code and Codex see precise parts of a UI through a single CLI command. It can render localhost or live sites, capture specific components with CSS selectors, simulate clicks, hover, and scroll, and choose different viewports.
How you can use it
Give coding agents visual context while they work
Capture specific UI components instead of full noisy pages
Verify frontend edits with screenshots
Test hover, scroll, and click states
Pricing: Paid
Jobs, announcements, and big ideas
Microsoft replaces OpenAI and Anthropic models with its own in-house AI across several Office apps.
Meta launches Muse, a new image generation model, directly inside the Meta AI assistant.
Beijing weighs new curbs on overseas access to DeepSeek and other top Chinese AI models.
Anthropic expands Claude Cowork to the web and mobile, bringing its agentic workspace to more devices.
Apple locks its new Home AI features behind a 2TB iCloud+ or Apple One Premier subscription.
Google DeepMind teams with studio A24 on a new AI research partnership for filmmaking.
Picture this: Imagine typing "edit this for me" and letting AI handle your entire video timeline. As a creator, I have a lot of thoughts on this one!
That’s a wrap! See you this Friday for more.
