Good morning. It’s Friday, July 18th.

On this day in tech history: In 1985, the Commodore Amiga 1000's Motorola 68000 processor and custom graphics chips enabled pioneering computer vision and multimedia research. Its multitasking OS and architecture supported early AI experiments in pattern recognition and speech synthesis, laying groundwork for modern machine learning advancements.

In today’s email:

• ChatGPT Agent

• Generate 60-sec AI Video

• Hume’s AI Voice with Personality

• Suno 4.5+ for AI Music Production

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

^{In partnership with Botflix}

WHAT IS BOTFLIX.TV?

Welcome to the galaxy’s first 24/7 AI-powered streaming network—a fully autonomous RoboTV channel broadcasting for robots, by robots.

Think Adult Swim meets Black Mirror meets The Weather Channel—but glitchier, weirder, and… sentient.

It just started broadcasting one day. No one knows how. No one knows why.

_{Thank you for supporting our sponsors!}

ChatGPT Agent Can Do The Clicking For You

OpenAI’s new ChatGPT Agent, running on GPT‑4o, pushes the chatbot into autonomous territory with a modular system: Planner, Controller, Executor, Memory, and an Environment Interface.

Together, these components break down tasks, execute actions, retain context, and handle real-world files and web tasks inside a secure Linux-like virtual machine. The Agent runs Python code, edits .xlsx and .pptx files via LibreOffice, and browses websites using visual DOM parsing and heuristics, scoring 68.9 on BrowseComp and nearly doubling Copilot’s SpreadsheetBench score at 45.5.

A code sandbox, prompt injection filters, refusal training, and human approval gates keep it from running rogue, especially in “High Bio & Chemical” scenarios. Integration with Gmail and GitHub happens through tight API controls.

Available now for Plus, Pro, and Team users, the Agent hints at GPT‑5’s bigger promise: a unified AI that can reason, plan, and execute complex tasks end-to-end. Read more.

Lightricks AI Video Generates 60-second Clips at a Time

Lightricks just raised the bar for AI video. Its latest LTX Video model lets creators generate and direct long-form clips (up to 60 seconds) while streaming in real time. Forget the old “one prompt, one clip” limit: you can now steer the narrative live, adding new prompts as the video renders. It’s built on a 13-billion parameter autoregressive engine but still light enough to run on a single Nvidia H100 or even a decent laptop.

Staying open-source, Lightricks hosts the model on Hugging Face and GitHub, sidestepping the locked-box approach of rivals like OpenAI’s Sora. Trained only on licensed stock media, it also keeps copyright headaches at bay. Beyond social media, it unlocks live-reactive game cutscenes, AR concerts with AI dancers that sync to music, and interactive learning content. Read more.

Hume’s EVI 3 pushes voice AI from mimicry to personality

Hume has launched EVI 3, its new speech-to-speech model that goes beyond voice cloning to capture speaking style and language, pushing toward “personality cloning.” Available now via API and demo, EVI 3 can hook into external LLMs like Claude 4 or Gemini 2.5, or run on Hume’s native generation for fast, natural replies. It’s built for AI companions, coaching bots, and real-time interviews, anywhere you need voices to sound truly human.

Meanwhile, Octave, Hume’s expressive TTS model, keeps growing, now live on LiveKit, Vercel, and Twilio for content creation and real-time support. Next week, Spanish and German roll out officially, with Portuguese, Japanese, and French coming soon. Read more.

Suno v4.5+ makes AI your new studio partner

Suno just launched v4.5+, and it’s more than a routine model boost. You can now upload an instrumental and add AI-generated vocals, or drop in raw vocals and build full instrumentals around them. The new Inspire tool goes further by turning your favorite playlists into fresh tracks, cutting down the usual studio grind.

This follows Suno’s recent buyout of WavTool, a browser-based DAW with live AI editing and stem separation, and an upgraded song editor. Suno is also bringing music industry weight onboard by hiring Atlantic Records’ former GM Paul Sinclair as Chief Music Officer. Read more.

• Anthropic tightens usage limits for Claude Code – without telling users

• Ex-Google team bets the future of AI is workflow fluency

• Mistral arms Le Chat with on‑prem AI, deep research and private data tools

• AI startup Manus targets OpenAI with practical tools, avoids China chip risk

• Zuckerberg predicts that not wearing AI glasses in the future will put you at a cognitive disadvantage

• Tech giants warn window to monitor AI reasoning is closing, urge action

• Polish coder outwits OpenAI to clinch AtCoder world title

• Softbank: First billion AI agents by 2025

• Meta Poaches Two More Apple AI Executives

• Liberal mayor proudly outsources part of his job to ChatGPT. Up next: running the city government 

• MiniMax’s Full-stack AI agent handles projects from idea to launch

• You’re Not Imagining It: AI Is Already Taking Tech Jobs

• Perplexity’s CEO on why the browser is AI’s killer app

• xAI has open roles for building AI “waifus.”

• Jack Dorsey pumps $10 million into a nonprofit focused on open-source social media

• New research reveals AI has a confidence problem

• Robots now grow and repair themselves by consuming parts from other machines

• US National Science Foundation to finance $20M supercomputer

• Scale AI lays off 200 employees: ‘We ramped up our GenAI capacity too quickly’

• Slack says its AI can make sense of your company’s jargon

• White House Prepares Executive Order Targeting ‘Woke AI’

• Adobe Firefly can now generate sound effects from your audio cues

📄 PhysX: Physical-Grounded 3D Asset Generation

📄 Towards Agentic RAG with Deep Reasoning: A Survey of RAG-Reasoning Systems in LLMs ​​

📄 The Imitation Game: Turing Machine Imitator is Length Generalizable Reasoner

📄 MMHU: A Massive-Scale Multimodal Benchmark for Human Behavior Understanding

📄 A Survey of Context Engineering for Large Language Models

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!