• AI Emergence
  • Posts
  • ⚔️ Iyo vs Altman – The Naming War Goes Public

⚔️ Iyo vs Altman – The Naming War Goes Public

Along with: Elon Wants Grok to Rewrite Human Knowledge

Hey there 👋

Sam Altman’s AI hardware brand “io” just walked into a naming war. The founder of another startup, Iyo, says his company’s had the name for years, owns the domain, and even pitched it to Altman’s team back in 2023. Then came radio silence, until “io” resurfaced as the name of OpenAI’s new device company. Now it’s turned into a full-on trademark dispute, complete with leaked emails, a lawsuit, and a very public airing out on X.

And if that wasn’t enough, Elon Musk casually claimed the next version of Grok will “rewrite the entire corpus of human knowledge.” The plan? Fix all the bad data foundation models were trained on, then retrain Grok on the cleaned-up version. No launch timeline, no product in sight, just Musk dropping future-plans mode on everyone, as usual.

Let’s see what else happened this week 👇

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Algorithm - Resources for learning

  • The Output - Our reflection

Table of Contents

💻 Google Open Sourced an AI Agent for Your Terminal

Google has open-sourced Gemini CLI, an AI agent that brings the power of Gemini 1.5 Flash directly into your terminal. It’s fast, customizable, and built for developers who want AI integrated into their actual workflow, not just tucked into a chat UI. The repository has over 20K GitHub stars in 24 hours, making it one of the fastest-growing AI repos right now.

What’s New:

  • Terminal-native experience: An open-source AI agent you can run entirely from your CLI. You can chat, execute tasks, and build workflows — all without switching windows.

  • Context-aware: Gemini CLI can read local files and understand your current project, making it useful for things like debugging, code explanations, or file generation.

  • Fully open source: Hosted on GitHub with a .geminirc config system. You can extend commands, change defaults, and build your own agents on top of it.

  • Powered by Gemini 1.5 Flash: Fast, low-latency responses optimized for dev environments.

  • Flexible backend: You can run it using Google’s API (free tier available via AI Studio) or hook it into a local model like Ollama.

Gemini CLI isn’t just another wrapper, it’s a lightweight, scriptable dev assistant that actually fits into real developer workflows. With local context, open customization, and strong early traction, it’s one of the more practical AI tools we’ve seen from Google so far. (Source)

🎙️ MiniMax Adds Audio – Now It’s Fully Multimodal

Last week we covered MiniMax’s new video model (Hailuo-02) and a big upgrade to its core language model (M1). Now, to close the loop, they’ve added Audio, giving them a full-stack setup across text, image, video, and voice.

What’s New:

  • Text-to-speech that actually sounds good: The new model supports realistic, expressive voices. Early demos suggest better emotion handling and tone control than many competitors.

  • Likely supports voice cloning: Based on MiniMax’s earlier “Talkie” system and hints in this release, cloning your own voice is probably part of the plan.

  • Richer voices than expected: Compared to the flat, robotic tone common in other models, these voices feel more fluid and natural, especially for character-style output.

  • Access still limited: No public UI for now. Available through platform/API access, mostly for developers and partners.

  • Completes the stack:  With M1 (text), Hailuo-02 (video), and image models already out, this audio launch officially puts MiniMax in the full multimodal club.


MiniMax isn’t slowing down. This voice model brings them up to full multimodal status — and early signs suggest the audio quality is more expressive and controllable than most. We’ll watch to see when public access opens up, but for now, they’re making a serious end-to-end AI play.

We’ve seen AI generate music before, think Suno, Udio, and Google’s own Lyria. But those tools are mostly about song creation, not real-time performance. Google’s latest release flips that script. Magenta RealTime is a lightweight, open-source music model built for live generation. It doesn’t just make songs, it responds to your input like an instrument, letting you improvise and perform with AI in the loop.

It’s one of the first music LLMs that’s fast, open, and designed for creators to actually jam with.

What’s New:

  • Magenta RT model: 800M-parameter autoregressive transformer trained on ~190k hours of instrumental audio. Small enough to run locally or on Colab, powerful enough to generate a clean 48kHz stereo.

  • Real-time generation: Produces 2-second blocks of audio in ~1.25 seconds. That’s fast enough to respond to prompts mid-performance, jazz solos, ambient transitions, even reactive soundtracks.

  • Open-source + permissive licensing: Released under Apache 2.0 + CC-BY-4.0. No paywalls, no tokens, no API keys. Code is on GitHub, weights on Hugging Face and Google Cloud.

  • Style control via text + audio: Prompt it with a phrase (“synthwave keys”) or drop in reference audio. You can blend both and shift styles midstream.

Magenta RealTime brings AI music generation into the live loop. You can tweak prompts on the fly, blend styles as you perform, and build new kinds of sound-driven experiences, all without waiting or paying.

It’s not perfect, it doesn’t do lyrics, leans heavily on Western instrumental styles, and works in short 10-second chunks with a bit of latency (~2s delay). But for artists who want hands-on, open, live AI tools, this is a massive step forward. (Source)

⚔️ Altman vs. IYO - A Trademark Tussle Over “io”

OpenAI CEO Sam Altman and Jony Ive’s new AI hardware project — called “io” — just ran into legal trouble from a voice-tech startup named Iyo. Turns out, Iyo says Altman’s team is using their name (just without the “y”), and they’ve taken it to court.

What Happened:

  • Iyo filed a lawsuit claiming “io” is too close to their name and will confuse users.

  • A federal judge agreed (for now), halting Altman’s team from using “io” in any promo or product material.

  • OpenAI pulled the branding, their new product site now says it’s down “due to court order.”

  • Altman clapped back on X sharing emails showing Iyo’s founder once pitched to partner with him. He called the lawsuit “silly and disappointing.”

  • Iyo’s founder replied: “Bad form” to leak emails, but fair game  “you just can’t use our name.”

What looked like a sleek product name is now a full-blown naming war. Altman’s on the defensive, Iyo’s holding the trademark, and a judge hits pause until October. Stay tuned!

🎙️ ElevenLabs Launches 11ai – A Voice-First AI That Actually Does Stuff

ElevenLabs just launched 11ai, and it's not your average voice assistant. It listens, talks back, uses your tools, and actually gets work done. You don’t have to type a thing.

What’s New:

  • Fully voice-first: Talk to it like you would a person. No keyboard needed.

  • Does more than chat: 11ai connects to tools like Gmail, Notion, Slack, and Zapier so it can take actions for you — from scheduling to summarizing to posting.

  • Powered by MCP: Uses ElevenLabs’ Multi-Component Prompting to combine tool use, memory, and reasoning behind the scenes.

  • Early version out now: It’s in free alpha, available to try at 11.ai. Just sign in, choose a voice(5000+ options or your own clone), connect tools, and start talking.

11ai is less “Alexa, what’s the weather?” and more “Do my Monday.” It’s still in early access, so don’t expect perfection but it’s already pointing toward a future where your voice controls everything.

✍️ Musk Says Grok Can Rewrite Human Knowledge

Elon Musk just casually claimed that the next version of xAI’s model, Grok 3.5 (or maybe Grok 4?), will feature “advanced reasoning” and can “rewrite the entire corpus of human knowledge.” The goal? Clean it up, then retrain on it. Because, as Musk put it, “far too much garbage in any foundation model trained on uncorrected data.”

If true, this points to a more editorial-style AI — one that not only consumes data but critiques, corrects, and restructures it. But for now, it’s all ambition and no receipts. No benchmarks, no release date, no public model. Musk’s tweet hints at a direction, but until we see it in action, Grok’s rewrite mission feels more under construction than deployment-ready.

🎥 Midjourney Launches Its First Video Model

Midjourney just launched V1 Video, its first take on animated AI. You start with a still image either one you’ve generated or uploaded  and it turns it into a 4–5 second clip with smooth, cinematic motion. All inside Discord.

What’s New:

  • Image-to-video animation: You can now animate images using the /animate command or by clicking “Animate” in your job history.

  • Built-in controls: Choose how much movement you want (low, medium, high), or go manual with keyframes if you're feeling fancy.

  • Longer videos for Pro users: You can stretch clips up to ~21 seconds if you're on the Pro plan.

  • Runs in Discord: Just like everything Midjourney. No extra installs or tools — it all happens in your usual channels.

  • Uses a lot more GPU time: Each video uses about 8× the GPU minutes of an image, so yeah, it’ll burn through credits quickly.

  • Starts at $10/month: Available on the Basic plan and up. No video access on the free trial.

  • DeepLearning.AI just dropped a short course that’s perfect for builders: Building an AI-Powered Game. You’ll learn how to generate game worlds and storylines using LLMs, structure outputs with JSON for gameplay features like inventories, and apply safety tools like Llama Guard for content control. It’s free, fast, and designed to get you shipping. Start here.

  • Google Research just shared how AI is tackling massive climate challenges - from predicting floods 7 days in advance for 700M+ people, to detecting wildfires from space with FireSat, and even reducing airline contrail emissions. Read the full breakdown from VP Yossi Matias here.

  • Vibe Coding with Replit: Just getting started with code? We made a free course that walks you through real projects right inside Replit. No setup, no stress. It’s super beginner friendly and perfect if you want to dip your toes into building with AI. Try it here

  • Analyzing Data with Power BI: If you’ve been meaning to learn Power BI but didn’t know where to start, this one’s for you. We’ll show you how to clean, explore, and visualize data step by step. No prior experience needed. Check it out

If last week was about agents and benchmarks, this one was all about sight and sound. Midjourney brought images to life with video. ElevenLabs made voice cloning mainstream. MiniMax rounded out its full-stack play by adding audio. Even Google’s Magenta popped up with a new real-time music model.

Everyone’s racing to make AI not just smart, but expressive — able to see, speak, and now even perform.

The chatbot era isn't over, but it's clear: the future sounds different. And it moves.

🎧 See you next week.

Reply

or to participate.