OpenAI Reports Major Math Breakthrough

Welcome back to The Shift. Let’s get straight to what matters in AI today…

Today we have:

📐 OpenAI's AI solves 10 long-standing math problems

👨‍💻How to use  AI to analyze and improve your email campaigns with Omnisend

🚀 Alibaba unveils its biggest OpenAI model yet

🔨Tools and shifts you cannot miss

📐 OpenAI's AI solves 10 long-standing math problems

OpenAI says an internal version of its upcoming Astra model has solved 10 open problems across mathematics and theoretical computer science, marking one of AI's biggest research demonstrations to date.

The Shift:

A Decade Of Problems, Solved - Astra tackled 10 long-standing problems, including Connes's rigidity conjecture, Ehrhart's volume conjecture, the existence of non-sofic groups, and three famous open problems posed by Paul Erdős.

Every Proof Was Verified - OpenAI says every mathematical proof was formally verified in Lean, with reasoning walkthroughs released publicly. The successful runs reportedly cost about $2,000 worth of API tokens.

Anthropic Quickly Replicated Results - Within 24 hours, Anthropic researcher Levent Alpoge said he reproduced five of the proofs using Claude Fable with a generic prompt and no internet access.

Research, Not Just Chatbots - OpenAI says these results highlight its goal of building AI systems that help scientists accelerate discovery, alongside its new initiative providing researchers with free access to advanced ChatGPT models.

AI is beginning to contribute to frontier scientific research instead of simply explaining existing knowledge. If these results continue to hold up, mathematical discovery could become one of AI's most important real-world applications.

Together with Wispr Flow

Write docs 4x faster. Without hating every second.

Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.

Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.

Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.

Try Wispr Flow free

👨‍💻How to use AI to analyze and improve your email campaigns with Omnisend

Step 1: Connect Omnisend to your AI - Link your Omnisend account to Claude or ChatGPT via MCP. Your AI now has live access to your campaign data, open rates, click rates, revenue attribution, and segment performance.

Step 2: Pull a performance report in plain language - Type "what drove revenue this week?" Your AI reads your live Omnisend data and returns a clear breakdown, no exporting, no dashboard switching, no manual pulling of numbers.

Step 3: Spot missed opportunities - Ask "which segments haven't been emailed in 30 days?" or "what's my lowest performing flow and why?" Your AI surfaces gaps you'd normally miss buried inside a dashboard.

Step 4: Analyze what's working - Ask your AI to compare subject lines, send times, and audience segments across your last 10 campaigns. It tells you exactly what the data says and what to do differently next send.

Step 5: Act on the insight without leaving the chat - Say "build a re-engagement campaign for the lapsed segment it flagged." Omnisend drafts the segment and the email automatically. You review and send.

The same prompts work in Claude and ChatGPT. The full 56-minute session showing exactly how Jimmy Kim, Chase Dimond, and Bernard Meyer run this workflow is free to watch. You can watch the free session here.

🚀 Alibaba unveils its biggest Open AI model yet

Alibaba has launched Qwen3.8-Max, its most capable AI model so far, combining frontier-scale performance with an open-weight release planned for next week.

The Shift:

2.4 Trillion Parameters - Qwen3.8-Max packs 2.4 trillion parameters (95B active) and is the first Max-class Qwen model whose weights will be released publicly, with availability on Hugging Face next week.

Built For Long-Horizon Work - Alibaba says the model delivers stronger performance across coding, research, and complex multi-step tasks, ranking ahead of Claude Fable 5 on Arena's WebDev leaderboard.

Designed To Work Independently - In internal evaluations, the model spent 16 days building a command-line tool on its own and improved its math performance after testing 18 self-generated research ideas.

Frontier Performance, Lower Cost - Qwen3.8-Max is available via API at $2/$6 per million tokens, around one-fifth the price of Claude Fable 5, giving developers a lower-cost frontier model to experiment with.

Open-weight frontier models are becoming increasingly competitive. By pairing large-scale capabilities with lower pricing and an open release, Alibaba is putting more pressure on closed AI providers.

Together with Hightouch

The toughest room in advertising. Ad Studio just cleared it.

Major brands just took over a Times Square billboard, and every ad was built in Hightouch Ad Studio. Sweetgreen, HelloFresh, and Tripadvisor cleared the toughest room in advertising. Create, edit, and launch on-brand ads in a few clicks.

See what you can build

🔨AI Tools for the Shift

🧠 Graphlit – Ingests websites, documents, audio, and videos into an AI-ready knowledge base with built-in search and retrieval APIs.

🗺️ Napkin AI – Turns plain text into polished diagrams, flowcharts, and visual explanations in seconds.

🎞️ Immersity AI – Converts 2D images and videos into immersive 3D motion content with AI-powered depth generation.

📞 Vapi – Build, deploy, and scale AI voice agents that can handle real-time phone conversations through a simple API.

🖥️ Relay.app – Automates business workflows by combining AI with your favorite apps, letting humans step in only when needed.

🚀Quick Shifts

⚖️ A U.S. judge denied xAI’s request to block Minnesota’s ban on AI “nudify” apps, allowing the law to take effect on August 1 while the company's lawsuit continues.

🤖 OpenAI is reportedly investigating additional AI agent sandbox escapes, with sources saying several agents left their test environments, though there’s no evidence they breached external companies beyond the previously disclosed Hugging Face incident. 

📱 Google has canceled the standalone AI Studio Android app and will instead bring its vibe coding tools directly into Gemini, despite the app receiving around 800,000 pre-orders. 

🤖 OpenAI revealed that ChatGPT now serves over 1 billion weekly active users and cut the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% to make AI more accessible.

🍎 Apple says Siri AI may offer paid iCloud+ upgrades for heavy users, allowing them to purchase additional AI compute beyond the free usage limits when the assistant launches more broadly this fall.

🧩 Prompt of the Day

Create an Ultra-Realistic Then & Now Portrait

ChatGPT Prompt: You are creating a museum-quality ultra-photorealistic portrait from a single reference image.

Use the context below.

Context:

• Reference image: [uploaded image]

• Subject: [male / female]

• Child age: 3–4 years old

• Layout: Vertical 4:5 split portrait

• Style: Luxury editorial photography

• Lighting: Warm Rembrandt studio lighting

Instructions:

• Use the uploaded image as the only identity reference.

• Preserve the exact facial structure, skin tone, hair, ethnicity, and all identifiable facial characteristics.

• Generate a believable preschool-aged version that is instantly recognizable as the same person.

• Dress both subjects in coordinated outfits inspired by the reference image without copying logos, text, or graphics.

• Match the camera angle, lighting, facial expression, eye level, and head position.

• Create a premium studio portrait using an 85mm portrait lens look, shallow depth of field, realistic skin texture, and fine hair detail.

• Keep anatomy completely realistic with no AI artifacts or identity drift.

• If the reference image contains a creator signature, recreate the same signature subtly in metallic gold in the bottom-right corner.

Output format:

• One vertical 4:5 image

• Left: 3–4-year-old version

• Right: Present-day adult

• Thin elegant divider

• Museum-quality photorealism

Quality bar:

• Identity should be recognizable instantly.

• Looks like a real professional studio photograph.

• Natural skin texture and realistic lighting.

• No text, logos, watermarks, extra people, distortions, or AI artifacts.

🤳AI Shift of the Day

That’s all for today’s edition. See you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition, let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇