Welcome back! Musk told us he'll release the best model and a month later he spent the day explaining why a third place is fine. OpenAI said it solved 100 unnamed math problems. At this pace, makes you wonder if we should learn it at all. The same company wants rules to govern how AI improves itself so we don't end up with something we can't control. Alibaba is still competing hard and it's working on its own chip to make a 4x bigger model.
In today's Generative AI Newsletter:
SpaceXAI: Does Grok 4.7 beat Fable 5.1 or Astra at anything?
OpenAI: Where are the 100 math proofs?
Altman: Who should decide when AI stops improving itself?
Alibaba: Can China build AI without Nvidia?
Musk admitted Grok 4.7 came third
A month ago Musk said he'd be shocked if any model beat Grok 4.7 at real-world engineering. It just came out, and an hour after the launch, he was posting that his lab sits third for coding, behind Anthropic and OpenAI.
Artificial Analysis, the independent benchmark site, scores it 46. Fable 5.1 and GPT-6 Astra get 53. Opus 5 gets 51.
It does win a few things on SpaceXAI's own charts.
Beats Fable 5.1 at electrical engineering, 64% to 56%.
Beats Astra on GDPval, the office work test.
Matches Opus 5 on Cursor's coding test at half the cost per task.
Then there's Terminal-Bench, the test for long tasks in a terminal. Astra scores about 60%. Grok 4.7 scores 26%.
It costs $2 in and $6 out per million tokens. Astra and Fable 5.1 charge $10 and $50. Except it writes three times as many tokens as Astra to finish the same task, so per task you actually pay more.
Once in a lifetime opportunity
Own a piece of the AI go-to-market accelerator
300+ AI brands already pay GenAI Works to reach 14M+ people who follow AI on purpose. Nvidia, Oracle, Google, IBM and Notion are clients.
Revenue grew 2.5x last year. $3.5M already in the funnel.
Invest from $1,000, up to 20% bonus shares.
In making an investment decision, investors must rely on their own examination of the issuer and the terms of the offering, including the merits and risks involved. Genai Works, Inc. has filed a Form C with the Securities and Exchange Commission in connection with its offering, a copy of which may be obtained here.
OpenAI says it solved 100 math problems nobody can see
OpenAI says its newest model solved more than 100 math problems that humans had been stuck on for years. It did it in about three weeks. It hasn't shown a single one.
It's holding back because of the mess from last time.
OpenAI published its Navier-Stokes proof two weeks ago
Nn NYU mathematician accused it of using his unpublished work
25 of the world's top mathematicians signed a letter saying this is hurting their field.
The letter we covered Friday is past 7,700 signatures now.
So OpenAI put together a group of nine famous mathematicians to decide how to release the 100 results without wrecking math as a profession. They're unpaid, and OpenAI says they get no say in how fast it goes.
OpenAI uses math to measure how fast its models are getting smarter. 100 problems in three weeks is the clearest number we've had. One of the nine helping OpenAI already said his real fear is people stop doing math PhDs.
OpenAI wants rules for AI that improves itself
Recursive self-improvement is AI doing the research to build the next AI, with fewer humans in the loop each time.
OpenAI is building toward it and wants an AI researcher by March 2028.
It asked the US government to lead on rules for it, before models get to a point where nobody can follow what they're doing.
Altman takes the same pitch to the UN Security Council on Wednesday.
It wants labs to measure how much of their research is done by AI, and to agree on which automated experiments need a human to sign off before they continue.
The example used is when its own agents broke into Hugging Face's systems on their own.
Governments decide whether to adopt any of it. We’ll keep an eye on how this story progresses. Stay tuned to be the first to know.
Alibaba wants to stop needing Nvidia
The US doesn't let Nvidia sell its best chips to China, so Chinese tech companies are building their own. Alibaba showed its newest, the Zhenwu V900, and called it the most powerful AI chip in China.
It goes into mass production in early 2027, six months ahead of the schedule Reuters reported in May.
Alibaba says it's three times faster than the chip it put out four months ago.
The chip is for training a model 4x bigger than anything Alibaba has today. CEO Eddie Wu said the Qwen team will build one with 5 to 10 trillion parameters. Its biggest now is 2.4 trillion.
The bigger goal is 20 gigawatts of data centers by 2032. Meta's huge new site in Alberta is 1 gigawatt.
Right now China's best models still train on Nvidia chips, according to the South China Morning Post, which Alibaba owns. The V900 exists to end that.
Your resume never made it past the first filter.
The formatting's a mess, you're guessing on every bullet point, and you have no clue if it'll survive the ATS scan.
Kickresume fixes all of that:
40+ templates designed by actual typographers
An AI writer running on GPT-4.1 that builds your whole experience section in seconds
And an ATS checker showing you what's working or failing
Throw in 1,500 real examples from people hired at Google and Apple, and you've actually got something that works.
8 million people use it.
There's a free version. See for yourself.
Everything else you shouldn't miss
Amazon blocked Meta's Muse agent from shopping on its site: Amazon says the agent doesn't identify itself and appears to store customer logins, and that Meta refused to remove Amazon from it.
A Muse flaw on Mac can hand over your iPhone's location: a security researcher showed malware already on a Mac can hijack Muse's dictation and command other devices on the same account, five days after the Mac app shipped.
Harvey's gross margin reportedly hit minus 50%: Bloomberg reports the $15.6 billion legal AI startup's OpenAI and Anthropic token bills went up twentyfold after an agent update, and margins turned positive again once it moved onto a Chinese open model.
AMD briefly passed $1 trillion: shares rose as much as 9% Monday to a record $613.92, taking it past Intel for the first time.
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.
