Good morning. It’s Wednesday, July 30th.

On this day in tech history: In 2001, Bernhard Schölkopf and collaborators popularized the kernel trick for support vector machines in their influential book “Learning with Kernels.” This mathematical technique let SVMs handle complex, nonlinear relationships by implicitly mapping data into high-dimensional spaces without ever computing those coordinates directly. This led to kernel PCA and kernel regression, tools that became widely used in machine learning before deep learning took over.

In today’s email:

• Runway’s Aleph AI Video FX

• OpenAI’s Study Mode

• NotebookLM’s Video Overview

• IdeoGram Nails Character Consistency

• 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 XerpaAI}

Skyrocket Your Growth with XerpaAI’s AI Revolution

XerpaAI, the world’s first AI Growth Agent (AGA), empowers businesses to scale rapidly with three standout features:

1. AI-Powered Content Creation: Craft targeted, high-quality content in minutes using advanced AI, streamlining marketing efforts and boosting engagement.

2. Influencer-Driven Distribution: Leverage a vast network of KOLs, KOCs, and community leaders for precise, conversion-focused campaign distribution.

3. Real-Time Data Feedback: Utilize comprehensive analytics to refine strategies and drive product iterations, ensuring agility and measurable results. These features deliver up to 3x higher conversion rates and 70% lower costs, replacing fragmented marketing with a unified, outcome-driven system.

Funding and Expansion

On July 9, 2025, XerpaAI raised $6M in seed funding from UFLY Capital, fueling innovation and global expansion across Silicon Valley, Tokyo, and Singapore. This investment cements XerpaAI’s role as a “Sherpa guide” for AI-driven growth. Web3 and Beyond. XerpaAI powers your projects and businesses, automating user acquisition to retention. Targeting a $600B–$1T market, it currently serves 80% blockchain and 20% AI startups, with plans for 2026 sales and compliance growth. Subscribe at http://xerpaai.com

^{Thank you for supporting our sponsors!}

Runway’s Aleph model upgrades real footage with AI precision

Runway’s new Aleph model flips AI video editing on its head: instead of generating clips from scratch, it lets filmmakers reshape real footage with text prompts. Want a new medium close-up? Aleph can create fresh camera angles. Runway calls it “endless coverage.” Need to add rain, remove smoke, or shift the time of day? Lighting adapts automatically to keep it seamless. Aleph digs even deeper: it can age characters, recolor props, generate green screen masks, and transfer motion from live video to stills. The result is faster, granular control over post-production that normally eats up budgets and time.

Currently limited to Enterprise and Creative Partners (with Lionsgate already testing it), Aleph reflects a bigger shift: AI tools evolving into multi-function post-production suites. Competitors like Google’s Veo 3 and OpenAI are following, but Runway’s strength lies in its polished interface and tight integration built for demanding film workflows. Read more.

OpenAI’s Study Mode turns ChatGPT into an adaptive tutor

OpenAI has launched Study Mode for ChatGPT, moving it from quick‑answer bot to interactive learning assistant. Instead of handing out direct answers, it guides users with targeted hints, reflection prompts, and adaptive quizzes, built on learning science principles like metacognition and curiosity‑driven exploration. Complex topics get broken into digestible steps, and the tool adjusts in real‑time to each user’s skill level. Study Mode is available now on ChatGPT Free, Plus, Pro, and Team.

Currently, the feature runs via external system instructions, meaning quality may vary, but OpenAI plans deeper model‑level integration, visual explanations, and personalized progress tracking. OpenAI is also partnering with Stanford’s SCALE Initiative to measure its real‑world educational impact.

Beyond structured tasks, OpenAI’s rumored GPT‑5 model briefly turned heads internally by piloting itself through Minecraft with near one‑shot competence. The agent’s short-lived experiment hinted at what might come next: large-scale models moving beyond static dialogue into creative, open‑world environments, combining reasoning, planning, and real‑time decision making. Read more.

Google’s NotebookLM adds ‘Video Overviews’ for multimodal research

Google is ramping up its generative AI portfolio with major updates across Vertex AI, Search, and NotebookLM. Veo 3 and Veo 3 Fast are now fully available on Vertex AI, giving enterprises high-definition video generation with native audio, precise lip-sync, and multilingual support. In August, both models will add image-to-video generation, turning static visuals into 8-second animated clips from simple text prompts.

Early adopters like Canva and eToro are using Veo to speed production, localize ads, and create cinematic content at scale - secured with SynthID watermarking and covered by Google’s AI indemnity.

In Search, Google’s AI Mode now accepts PDFs and images for context-aware answers, offers real-time camera-based queries via Search Live, and introduces Canvas to build living study plans that evolve across sessions. Chrome gains an “Ask Google about this page” shortcut for instant AI insights.

NotebookLM also steps up: its new Video Overviews modify user notes into narrated slideshows enriched with visuals, charts, and quotes, turning dense material into accessible, AI-generated video summaries. Read more.

Ideogram Character nails consistent AI characters from a single image

Ideogram has unveiled Ideogram Character, the first AI model to deliver character consistency from just a single reference image, now free on ideogram.ai and the iOS app. Creators can craft real or fictional characters that stay visually coherent across countless scenes, using curated templates or custom prompts.

You can blend it into existing tools for serious creative control: use Magic Fill to drop your character into any new scene, or tap Describe and Remix to capture and transfer specific styles from inspiration images. It even auto-generates a character mask (face, hair, clothing, or props) so you can fine-tune exactly what defines your character’s look.

Beyond just style, this solves a real headache in generative design: believable, repeatable character visuals that stay true shot after shot. Ideogram Character finally keeps your creations consistent. [Read more.]

• Alibaba’s Wan2.2 pushes open-source video AI to new speeds

• Grok readies “Imagine” for endless AI image & video generation

• Anthropic reins in Claude with fresh caps, cooks up Ghost and Beam chats

• Unitree drops ultra-budget humanoid R1

• Watch: China's humanoid robot nails human-like motion with 31 flexible joints

• Quantum computer error rate slashed to record‑low 0.000015%, paving way for smaller, faster machines

• Watch: Alibaba previews its first AI-powered glasses, joining China’s heated smart wearable race

• World's most advanced haptic gloves bring real-time touch into virtual reality

• Mira Murati's $2 billion seed round inspires female founders

• Researchers test the trustworthiness of AI by teaching it to play sudoku

• Harmonic, the Robinhood CEO's AI math startup, launches an AI chatbot app

• Chinese startup Z.ai releases cost-efficient GLM-4.5 reasoning model

• Lab team finds a new path toward quantum machine learning

• China tops global AI development with 1,509 large models

• Survey: 66% of devs battle “almost right” AI code that drains time

• Amazon-backed Skild AI unveils general-purpose AI model for multi-purpose robots

• Luma and Runway expect robotics to eventually be a big revenue driver for them

• Meta is going to let job candidates use AI during coding tests

• Photoshop’s new AI lets you drop in objects and sharpen shots in a snap

• New Super Agent from Writer tackles multi-step knowledge work using 600 tools

• Copilot now creates podcasts on demand from your queries

• Watch: Cutest Humanoid Robot Ready For Launch

• A profile of Luke Farritor, a 23-year-old Thiel Fellow who gained recognition for using AI to help decipher the Herculaneum papyri and later joined DOGE

• Moonvalley’s Marey model turns rough sketches into realistic videos

• Nvidia AI chip challenger Groq said to be nearing new fundraising at $6B valuation

📄 Deep Researcher with Test-Time Diffusion

📄 ARC-Hunyuan-Video-7B: Structured Video Comprehension of Real-World Shorts ​​

📄 ScenePainter: Semantically Consistent Perpetual 3D Scene Generation with Concept Relation Alignment

📄 Agentic Reinforced Policy Optimization

📄 SmallThinker: A Family of Efficient Large Language Models Natively Trained for Local Deployment

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!