AI IMAGE GEN

Meta brings Midjourney's capabilities to 3.9 billion users

Meta struck a deal Friday to license Midjourney's "aesthetic technology" for future models and products, marking a shift toward external partnerships after internal AI development setbacks. The move comes as AI-generated content reaches a believability tipping point that's fooling millions.

Remember those viral trampoline bunnies that captivated over 200 million TikTok viewers? The AI-generated video was so convincing that most people couldn't tell it was fake until glitches were pointed out. That's the world Meta is betting on, where synthetic content becomes indistinguishable from reality.

For a company generating an estimated $200 million annually, Midjourney operates with remarkable stealth. While OpenAI, Anthropic and others dominate headlines, Midjourney has built an AI empire with just 40 employees and zero outside funding. Midjourney's $5 million revenue per employee dwarfs tech giants like Dropbox ($595K) and Slack ($240K).

Founder David Holz deliberately avoided venture capital after his previous company, Leap Motion, was sold for a fraction of its expected value despite raising over $100 million. This bootstrapped approach gives Midjourney complete creative control and independence from funding cycles that plague other AI companies. The company operates Discord's largest server with over 21 million members and only launched a standalone web interface in August 2024.

Meta's partnership signals its internal AI development hasn't kept pace with specialized competitors. While Meta has its own image generator (Imagine) and video tool (Movie Gen), neither has achieved Midjourney's aesthetic quality or cultural impact.

For Meta, the deal offers several strategic advantages:

• Access to Midjourney's superior aesthetic capabilities without acquisition costs

• Technical collaboration that could accelerate internal model development

• Association with a brand known for artistic quality rather than corporate utility

• Potential integration across Facebook, Instagram, and WhatsApp platforms

Midjourney faces major copyright lawsuits from Disney and Universal, who allege the company trained models on stolen intellectual property. Meta inherits this legal uncertainty as AI training practices face increasing scrutiny.

Companies are realizing they can't excel at everything and are choosing specialization over vertical integration.

This partnership accelerates that timeline, bringing Midjourney's capabilities to Meta's 3.9 billion users. The bunny video previewed a future where feeds are flooded with synthetic content designed to maximize engagement regardless of truth. We're watching the architecture of digital reality being rebuilt, one beautiful lie at a time.

TOGETHER WITH ZED

Code at the Speed of Thought

Zed is the fastest way to code with AI. Built from scratch for speed, it delivers instant completions that move at your pace. No bloat or latency, just precision and performance with every keystroke. Collaborate with agents and teammates in real time, all at 120fps. 

Choose from popular models, connect your own locally, or bring your own keys. We care about developer flexibility, not AI upsells.

Download today

FUNDING

OpenAI threatens to void unauthorized investments through SPVs

OpenAI fired a warning shot at the secondary market last week, threatening to void any equity purchases made through unauthorized channels, including special purpose vehicles (SPVs). The ChatGPT maker published a policy statement Saturday, making clear that any attempted transfer of its equity without written consent "is void" and "will not be recognized and carry no economic value to you."

SPVs have become a popular way for smaller investors to pool money and buy into hot AI companies that would otherwise be out of reach. Business Insider reports that Anthropic told Menlo Ventures it must use its own capital, not an SPV, to participate in an upcoming funding round.

The timing is particularly pointed given Robinhood CEO Vlad Tenev's aggressive push for tokenization. Earlier this year, Robinhood launched tokenized shares of OpenAI and SpaceX for European users, essentially creating crypto-wrapped exposure to private companies through SPVs. In a Washington Post op-ed in January, Tenev argued that tokenization could democratize access by offering:

• 24/7 trading capabilities for private shares

• Lower barriers to entry for retail investors

• More capital sources for high-growth companies

Companies like OpenAI clearly want all the benefits of massive valuations and investor hunger without surrendering control over their cap tables. When you’re as hot as OpenAI, I guess you can be very selective about whose money you take.

TOGETHER WITH FUEL IX

Secure your GenAI before it's too late

In the race to innovate with GenAI, don’t take unnecessary risks with AI safety. Join TELUS Digital, Fuel iX and guest Forrester for a discussion on how product teams can fast-track secure GenAI development without compromising on safety or speed.

This exclusive session will explore:

• Battle-tested security frameworks from leading enterprises

• Practical automation strategies for enterprise-wide protection

• Real implementation stories

Watch industry experts share their proven paths to secure, scalable GenAI deployment.

Watch on-demand

AI APPLICATIONS

AI takes over content moderation, struggles with the nuance

Social media platforms are aggressively replacing human content moderators with AI systems, despite mounting evidence that the technology isn't ready for the job. TikTok laid off around 150 content moderators in Berlin earlier this month, nearly 40% of the team responsible for moderating content for Germany's 32 million users. On Friday, TikTok announced plans to cut hundreds more moderators across the UK and Asia while investing in AI moderation technologies.

Human moderators are expensive, prone to psychological trauma from graphic content exposure, and companies have spent years outsourcing the work to poorly paid contractors. AI promises to handle the massive volume without needing therapy or breaks. But according to 13 professional moderators interviewed by Bloomberg, the technology consistently fails at the job's most critical aspects.

Kevin, a TikTok content moderator in Africa, estimates AI fails up to 70% of the time. Zhanerke Kadenova, who works for a content moderation firm in Kazakhstan, says AI suggestions don't match reality 80% of the time. The systems make bizarre errors:

• Highlighting low fuel gauges instead of dangerous speedometer readings

• Identifying children as 17-year-olds

• Missing contextual clues about violence or abuse

• Failing to understand regional dialects or cultural nuances

Child predators represent the most dangerous blind spot. They study platform automation tactics and evolve faster than AI can learn, using coded language like "Let's party" or "Meet me on the ghost app" to circumvent detection. When platforms catch on, predators simply put Xs between letters or invent new phrases.

Companies like Meta and Roblox continue facing scrutiny over child safety failures, yet they're doubling down on AI moderation to cut costs. The result will likely be platforms where coded hate speech, propaganda and predatory behavior persist while legitimate content gets incorrectly flagged and removed.

LINKS

• Startup Lightning AI launches GPU rental marketplace

• Google AI Mode expanding to 180 countries

• Amazon is betting on agents to win the AI race

• Meta poaches Apple AI executive even as it plans hiring slowdown

• OpenAI's chief people officer to depart

• OpenAI is challenging Google, while using its search data

• Waymo granted first permit to begin testing autonomous vehicles in NYC

• Tesla to integrate DeepSeek & Doubao AI voice controls in China, Grok in US

• Intel agrees to sell U.S. a 10% stake in its business

• Indirect prompt injection in Perplexity Comet

• Elon Musk says xAI has open-sourced Grok 2.5 with Grok 3 in 6 mo’s

• Netflix wants its partners to follow these rules when using gen AI

• Grok 2.5 (oss): xAI’s best model of 2024, now open source

• Recall: Chat with everything you’ve read, heard, watched, or noted

• Mocke: Predicts your reply rate to a cold email campaign before pressing send

• Gemini Live: A more helpful, natural and visual assistant

• Pylon: Software Engineer (AI Agents)

• Casca: Founder’s Associate

• Firecrawl: SEO & GEO Specialist

• Zed: BizOps

GAMES

POLL RESULTS

Do you care about chatbot energy usage?

• Yes (54%)

• No (35%)

• Other (share your thoughts) (11%)