Daily AI News | OpenAI Publishes Ten Astra-Generated Advances in Mathematics and Theoretical Computer Science
# Daily AI News | OpenAI Publishes Ten Astra-Generated Advances in Mathematics and Theoretical Computer Science
## Lead Story
### OpenAI Publishes Ten Astra-Generated Advances in Mathematics and Theoretical Computer Science
OpenAI published ten results generated by an internal version of Astra across high-dimensional sphere packing, coding theory, non-sofic groups, arithmetic-circuit complexity, quantum complexity, lattice cryptography, and extremal combinatorics. OpenAI says the problems had seen no progress on their main result for at least a decade; people prepared manuscripts with the same model, the model then produced a Lean certificate for each argument, and the solution tokens would cost about $2,000 at Sol API rates.
**AI take:** This shows the upper end of AI as a research collaborator, but mathematicians still need to scrutinize each result; Lean certificates improve checkability without constituting peer acceptance.
## More News
### xAI Adds Reference Inputs to Grok Imagine Video 1.5
xAI added References to Grok Imagine Video 1.5 rather than launching the 1.5 model for the first time that day. The update supports text-to-video, native 1080p, image and voice references, and up to seven reference inputs; image and voice references start with US SuperGrok Heavy and Plus users, while image references, text-to-video, and 1080p are also in the API and voice-reference API access requires a request.
**AI take:** Multiple references improve control over character and style consistency, but voice references are still staged and availability differs across web, mobile, and API access.
### OpenAI Resets Codex and ChatGPT Work Weekly Limits Again
Codex lead Tibo Sottiaux said on August 1 that OpenAI had again reset weekly usage limits for every Codex and ChatGPT Work user, the fifth reset since July 22. Accounts near or at their weekly ceiling received immediate capacity again, but OpenAI did not describe the reset as a permanent limit increase or a new pricing rule.
**AI take:** Heavy users get immediate capacity back, but repeated temporary resets are not a sound basis for long-term workload or cost planning.
### OpenAI Extends Content Provenance Measures to Audio in Europe
OpenAI said it contributed to and endorsed the EU General-Purpose AI Code of Practice and the Code of Practice on Transparency of AI-Generated Content, while extending content-provenance measures from images to audio outputs. The company plans to combine Content Credentials with SynthID and expand provenance across other modalities, including text, as standards and tools mature.
**AI take:** Audio provenance can help identify synthetic media, but metadata can disappear and watermarks are not absolute proof; transparency obligations will increasingly shape product design and audits.
## Rumors
