AWS can control your browser & more (April 1, 2025 edition)
David Pawlan
Co-Founder
Happy April 1st — a day when the tech world walks a tightrope between real launches and really elaborate jokes.
But here’s the twist: this week, the AI news is so wild, it’s hard to tell the difference. Let's dive in and you can decide what’s real and what’s just playing 4D chess with your trust.
🕵️♂️ Amazon’s Agent Can Now... Use a Browser?
Amazon dropped Nova Act, a browser-controlling AI agent that quietly crushes OpenAI and Anthropic in performance tests. It fills out forms, clicks through websites, and manages calendars like a caffeinated intern. And yes—it’s coming to Alexa. So if your Echo starts applying for jobs on your behalf, maybe don’t be surprised.
🤖 Not a joke: This might be Amazon’s most consumer-facing AI play yet.
🎞️ Runway’s Gen-4 Video Model: Too Good to Be Real?
Runway’s new video model holds scene consistency, generates cinematic motion, and even got a shout-out in Madonna’s tour visuals. It’s basically a filmmaker's dream—or deepfake nightmare—depending on your POV. They’re calling it “GVFX” (Generative Visual Effects), and early adopters are already integrating it into pro workflows.
🎬 Joke potential: Still waiting for the model that generates your Oscar speech.
🧠 OpenAI’s Thinking Sliders: Serious Feature or Saturday Night Sketch?
OpenAI is testing “thinking sliders” in ChatGPT. The idea? Let users dial up or down the model’s depth of reasoning—because nothing says trustworthy AI like giving it a mood ring. Also in the works: desktop dictation, deep research sharing, and a notifications tab (finally, ChatGPT pings like Slack).
🧪 Real-ish: OpenAI raised $40B, so at this point, we’re all living in their simulation.
🧬 Google Gives Away Gemini 2.5 for Free — No Foolin’?
Gemini 2.5 Pro (experimental) just dropped into free accounts, making it the most powerful model you can access without opening your wallet. It tops the LMArena board and has reasoning baked in. Oh, and there’s now a Canvas tool for building projects visually.
🚨 Possibly the biggest April Fools twist: It’s actually real.
📡 OpenAI Is Going “Open” Again. Sort Of.
Sam Altman says a new open-weight model is on the way—the first since GPT-2. Open...ish. Details are fuzzy. Will it be usable? Transparent? Priced for humans? TBD. But hey, they’re accepting feedback, so if you’ve ever wanted to tell OpenAI how to do their job, now’s your moment.
🔓 Joke probability: 70%. True openness might be the biggest prank of all.
⚡ Quick Hits (No Joke... We Think)
- 🧠 UC Berkeley built a brain-to-speech system with only a 1-second delay.
- 💊 Google’s Isomorphic Labs raised $600M to design drugs with AI.
- 🎥 Google Gemini can now drop your product into any scene and turn it into a product video.
- 📚 NotebookLM makes mind maps from your PDFs. Finally, your thoughts have a UI.
Takeaway?
This week’s AI drops feel like they were pitched in an April Fools’ group chat and somehow made it to production. Browser agents? AI Oscar-worthy video models? Brainwave-to-speech tech?
But here’s the thing—most of it’s real. Or at least real enough that if you blink, your fridge might start giving TED Talks.
Let me know if you want a follow-up breakdown of which of these are already being used in real-world workflows (vs. just being tweeted into existence). Might do a roundup next week.
– David
P.S. If you made it this far, congrats: you passed the Turing Test... for newsletter readers. 🧠💌