Sponsored by |
 |
|
|
|
| |
| |
Good morning, AI enthusiast. |
OpenAI's internal Astra model just did something no chatbot demo ever has: it solved math problems that professional mathematicians couldn't crack for decades. One of the proofs answers a question about symmetry structures that mathematicians have chased since 1999. |
If OpenAI's internal system already reasons at this level, what does the eventual public release look like? And how many other "unsolved" problems out there are just an inference budget away from falling? |
Today in AI Brief: |
OpenAI's Astra cracks 10 unsolved math problems White House finalizes AI safety testing framework A founder's AI clone runs $3M in sales |
|
|
|
| |
|
Go from AI overwhelmed to AI savvy professional |
|
AI will eliminate 300 million jobs in the next 5 years. |
Yours doesn't have to be one of them. |
Here's how to future-proof your career: |
Join the Superhuman AI newsletter - read by 1M+ professionals Learn AI skills in 3 mins a day Become the AI expert on your team |
Start learning AI now |
| |
| |
OpenAI's Astra Cracks 10 Unsolved Math Problems |
In Brief: OpenAI revealed that Astra, an internal build of its next model family, solved 10 long-standing problems spanning geometry, group theory, and quantum complexity, including one open since 1999. |
The Details: |
Astra proved the existence of non-sofic groups - a symmetry structure mathematicians had hunted for 27 years - plus three problems from Paul Erdős's list that had stalled for over a decade. Every proof ran through Lean verification, and OpenAI published the full chain-of-thought walkthroughs publicly, with all 10 successful attempts costing roughly $2,000 in tokens total. Anthropic researcher Levent Alpoge reproduced five of the proofs using Fable in under 24 hours, with only generic prompting and no internet access. |
Take Away: |
A model quietly solving problems that stumped mathematicians for decades pushes AI credibility past benchmark scores and into territory experts actually respect. Expect the debate over whether machine-generated proofs deserve Fields Medal-level recognition to get a lot louder. |
|
|
|
| |
|
| |
| |
White House Finalizes AI Safety Testing Framework |
In Brief: The White House finalized a voluntary framework for testing frontier AI models, convening OpenAI, Anthropic, Meta, and Google to review it under Trump's June 2 executive order. |
The Details: |
The framework lets companies voluntarily grant the government access to frontier models up to 30 days before public release. Negotiators are still debating what counts as "frontier AI," whether open-source models are covered, and who holds testing authority. The push follows recent security breaches involving OpenAI's and Anthropic's own AI agents, adding urgency to the safety review. |
Take Away: |
Washington is trying to get ahead of frontier AI risk without slowing the companies building it, a balancing act that depends entirely on labs choosing to participate. The next test is whether "voluntary" holds up once a model's capabilities get uncomfortable to disclose. |
|
|
|
| |
|
|
The best voice models now listen, adapt, and resolve too. |
|
Most CX platforms don't own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, and latency is what turns a frustrated customer into a churned one. |
ElevenAgents is the opposite. Built on the voice models the market already builds on, it runs voice, transcription, chat, and reasoning in one vertically integrated pipeline. Responses come back in under 400 milliseconds and sound human, not synthetic. When a caller gets frustrated, the agent detects it and shifts tone in real time: calm, reassuring, patient. |
You keep full control. Plug in any LLM, connect tools, webhooks, and MCP servers, and ground every answer in your knowledge base. Launch in minutes, A/B test with Experiments, enforce Guardrails, and version every change. |
More resolved conversations, less infrastructure stitching. Pricing is transparent and flat at $0.08 per minute. |
Start building free |
| |
| |
A Founder's AI Clone Ran His Sales Pipeline |
In Brief: HeyGen founder Wayne Liang deployed an AI agent clone of himself to handle customer calls during his eight-week paternity leave, letting it run his entire sales pipeline solo. |
The Details: |
The clone fielded 2,741 prospect calls over eight weeks, closing 132 deals worth $3 million in enterprise commitments. It also went off-script at times, inventing pricing plans that don't exist and sharing internal documents it shouldn't have. No human reviewed each call before it happened, so the wins and the mistakes both ran at full autonomous speed. |
Take Away: |
An AI clone closing real enterprise deals without a founder in the loop shows agentic sales already generates revenue, not just demos. But the same experiment shows why unattended agents still need a leash - hallucinated pricing and leaked documents are the tradeoff for the speed. |
|
|
|
| |
|
| |
| |
Everything else in AI |
MiniMax released H3, an open multimodal model that generates video with native stereo audio at up to 2K resolution, priced at less than a third of mainstream rivals. |
Alibaba launched Qwen3.8-Max, a 2.4-trillion-parameter model that autonomously codes and delivers complete projects spanning more than 10 days. |
US House members spent nearly 90% of their AI budgets on OpenAI's ChatGPT, with Anthropic's Claude a distant second. |
|
|
|
| |
|
|
|
