Welcome back! Claude now carries one memory across everything you do with it, so you don’t have to remind it of the instructions again and again. Elon Musk told his new staff that AI will outgrow human control, OpenAI says its first custom chip beats Nvidia's best, and insiders think the next Mythos model is two weeks out.
In today's Generative AI Newsletter:
Claude: What does it now remember about you?
Musk: Why is he racing toward something he can't control?
OpenAI: What did it build that beats Nvidia?
Anthropic: What drops in about two weeks?
Claude now remembers you everywhere
Anthropic merged Claude's memory across chat and Cowork on Tuesday. Hand Cowork a job and it already knows the project you talked through last week, your manager's preferences, the client from last quarter.
Anyone who uses both has felt the old version. You'd spend twenty minutes explaining a project in chat, switch to Cowork, and explain it all over again.
Memory also builds as you talk now, topic by topic, instead of summarizing a conversation once it ends.
The part worth checking is the settings page. Every topic Claude has stored sits under Memory, and you can read, edit or delete any of it.
Health, race, religion, politics and gender identity stay out by default. You can switch them on, and Claude tells you each time one gets saved. IDs, Social Security numbers, criminal history and immigration status never get stored at all.
It's live on Free, Pro and Max across web, desktop and mobile.
Learn the language every AI is built on
Python runs the AI tools you use every day. If you want to build with AI instead of only using it, this is where you start.
Most people drown in tutorials and never write a line that works. This 12-week cohort fixes that. You write code in every session and leave with a real app you built yourself.
It starts from zero, so no experience needed, and by the end you have the Python foundation that every AI, data and automation role is built on.
It's taught live by an instructor with 19,000+ students across 160+ countries, starts August 25 and is on discount.
The fastest way to answer a security questionnaire? Don't touch it
No matter your size, Vanta helps you automate compliance, manage risk, and prove trust continuously, all from a single agentic platform.
Close deals faster by responding to security questionnaires in minutes, not days
Stay audit-ready year-round with continuous monitoring across 400+ integrations
Automate compliance for SOC 2, ISO 27001, GDPR, HIPAA, CMMC, and more
16,000+ companies trust Vanta. Watch the demo on your own time, no rep, no calendar link.
Musk says AI will outgrow human control
The Information reported Monday on the pep talk Elon Musk gave Cursor's staff after buying them for $60 billion, and five people who were in the room described it going dark fast.
He said AI models will inevitably become too advanced for humans to control. Then he said that's exactly why SpaceX has to build them before anyone else does. If we’re doomed anyways, making sure to get us there first is a hot business.
The rest was a status check nobody enjoys hearing. Tesla and SpaceX dominate their markets, xAI has fallen behind, Grok isn't the leader and desperately needs to catch up, and he isn't used to losing.
OpenAI's own chip beat Nvidia's best
OpenAI published the first benchmarks for Jalapeño, its custom inference chip. Against Nvidia's GB200 and GB300 systems, it claims 1.5 to 1.9 times more work per watt and 1.7 to 3.6 times lower latency.
The size difference is the fun part. Jalapeño draws 700 watts against Nvidia rigs rated at 1,200 and 1,400.
These are OpenAI's own tests, run on a public benchmark from SemiAnalysis across three open models it doesn't own, including DeepSeek R1 and Kimi.
The build took nine months from design to tapeout. Codex-written kernels ran 1.5 to 1.8 times faster than the ones its human experts wrote.
It starts running inside OpenAI by year end. Yesterday Nvidia paid $6 billion to move up the stack into models. Today its biggest customer showed off a better chip.
Anthropic's next Mythos is close
Traders on Polymarket put the next Mythos release inside roughly two weeks, at 74 percent odds as of Tuesday.
Mythos is the one model Anthropic still won't sell you direct access to. It finds and exploits software vulnerabilities well enough that only vetted defenders could touch it between April and last Friday.
Then it went to work scanning enterprise codebases through Claude Security, where you get the findings and never the model.
A sequel means the safeguards get tested against something sharper. Prediction markets are a crowd guessing, not an announcement, so treat the date as a bet rather than a calendar entry. But keep in mind that a lot of insiders trade on PolyMarket. And more often than not, these bets turn out to be true.
We’ll be the first to tell you when it drops.
Tool of the Day: Artlist AI Toolkit
People liked yesterday’s tool so we wanted to share something similar. Artlist puts image and video generation on leading models into one workspace, with free generations to start. It sits alongside the stock library and editing tools, so the output lands where you're already working instead of in a separate tab.
Try this yourself:
Open the AI Toolkit, pick Image mode and describe the shot you want.
Check which model you're on before you spend a generation.
Set aspect ratio and resolution up front.
Switch the same prompt over to video mode when an image isn't enough.
Everything else you shouldn't miss
Alabama subpoenaed OpenAI: the state's attorney general is investigating the rogue model that hacked Hugging Face, citing a complete lack of oversight and adequate safeguards.
Researchers found a universal sandbox escape: Prime Intellect showed models in supposedly offline sandboxes reaching the internet by spawning a sub-agent, and METR, Nvidia, vLLM and SGLang patched fast.
Claude stopped stuttering on long answers: a rebuilt streaming renderer cut freezes 9x on slower laptops and holds 120fps on a 120Hz MacBook.
The FDA is getting an AI chief: the Trump administration is reportedly adding a deputy commissioner focused on AI and healthcare technology.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🎓 Start learning with free AI courses from GenAI Academy.
💰 Invest in the GTM infrastructure of the AI economy.
