Intel partners with a rival & more (April 4, 2025 edition)
David Pawlan
Co-Founder
Hope your Friday’s off to a smart start — because AI sure is. Between DeepMind teaching models to plan like humans and OpenAI’s image tool going viral with 700M+ creations, there’s a lot moving fast this week.
Here’s your quick catch-up on what’s shaping AI right now — the biggest headlines, smartest tools, and under-the-radar shifts that actually matter:
🔮 AGI by 2030? DeepMind Thinks So
DeepMind released a monster 145-page report forecasting that Artificial General Intelligence (AGI) could outthink humans by the end of the decade. Their roadmap? A mix of agent oversight, interpretability tools like MONA, and AI self-assessment systems.
Meanwhile, their Minecraft-trained DreamerV3 agent just pulled off the game’s hardest challenge — collecting diamonds — without human help. It’s not just impressive gameplay. It’s planning, generalization, and adaptation in the wild. The kind of behavior that edges AI closer to humanlike reasoning.
🤝 Intel & TSMC: When Enemies Collaborate
In a move nobody saw coming (except maybe the White House), Intel is reportedly close to a manufacturing deal with rival TSMC. Why? To save itself.
After $16B in losses last year, Intel may let TSMC take over part of its chip production ops. TSMC gets a 20% stake; Intel gets a lifeline. And the U.S. boosts domestic chip resilience at a time when global supply chains are looking shaky.
This isn’t just about chips. It’s a reminder that AI progress is deeply tied to who controls the infrastructure underneath it.
🎨 Midjourney V7: Faster, Sharper, More You
Midjourney dropped version 7 — a total architecture overhaul with better prompt understanding, crisp hands (finally), and personalized tuning. Users train it by rating 200 images to match their taste. There’s also a “Draft Mode” for rapid, low-cost renders you can upscale instantly.
If you’re building visual content pipelines, this might be the model to watch right now.
📸 ChatGPT’s Image Gen Is on Fire
OpenAI’s image tool hit 700 million generations in one week, mostly from India. A standalone API is coming soon. Paired with the new PaperBench research replication challenge (AI vs PhDs), and a commission to redesign their nonprofit structure, it’s clear: OpenAI is scaling fast — both in compute and influence.
That said, costs for the top-tier “o3 high” model are now projected at $30k per task. Powerful, yes. Affordable? Not for long.
🧪 Quick Hits Worth Knowing
- Runway raised $308M to bring Gen-4 AI video into full production workflows. Think Hollywood meets AI.
- Adobe Premiere Pro now extends videos and ambient audio with a single click. 4K support included.
- GPT-4o now lets you transfer styles between images. Upload two images, describe the vibe you want copied, and let it work its magic.
- Anthropic found that AI models often hide their true reasoning 80% of the time. Chain-of-thought may not be the transparency tool we hoped for.
- Amazon’s Nova Act lets agents control browsers. It’ll debut inside Alexa+, bringing agentic AI to the mainstream faster than expected.
- Spotify added AI-powered tools to write & voice ads. DIY advertising just got way easier.
⚙️ 3 AI Tools to Try This Weekend
🧠 Genspark – A team of AI agents for research & writing.
🎬 HeroUI – Turn screenshots or prompts into React apps instantly.
🧪 Voicepanel – Get UX insights from real users with AI-run interviews.
That’s a wrap for today.
If you're seeing the AI ecosystem move from fun experiments into real workflows… you're not imagining it. The next chapter is practical. And it’s already started.
See you next week.
— David