OpenAI's Core Experimentation team is significantly enhancing software maintenance efficiency by leveraging the GPT-6 Astra and GPT-6.1 Sol models. These AI agents automate tasks such as removing unused code, cleaning up stale experiments, and optimizing Continuous Integration (CI).

Specifically, GPT-6 Astra handles ambiguous requests by consulting past PRs, Slack threads, internal project trackers, and production profiling data to justify deletions. This has led to over 1,400 pull requests being merged, deleting more than 800,000 lines of code, and resulting in tangible benefits like a 48% reduction in one service's container image size.

For cleaning up stale experiments, GPT-6 Astra considers complex dependencies and production runtime conditions, identifying and proposing deletions for high-CPU-cost experiments. One example cited a 0.42% CPU saving.

Furthermore, GPT-6.1 Sol demonstrated high cost-effectiveness, achieving 92% of the findings of GPT-6 Astra Ultra at only 14% of the API cost. This has streamlined various maintenance tasks, including fixing flaky tests, improving agentic context, and reducing CI runtime, with median CI completion time falling from 10 minutes to 4.5 minutes.

👉 Read the full article on OpenAI Developers

Tesla's new Roadster prototype showcased astonishing capabilities at a SpaceX test site, demonstrating acceleration to 60 miles per hour in under a second using boosters, driving upside down, and performing aerial corkscrew maneuvers.

Elon Musk, Tesla's CEO, explained that the prototype is a collaboration between Tesla and SpaceX, utilizing an "advanced ultra-high pressure air tank" to power SpaceX cold-gas thrusters. The demo featured the Roadster launching upside down off a ramp, executing a corkscrew, and even jumping over an entire fleet of Tesla vehicles.

Musk stated that the car can "fly like a rocket and do rolls," with suction fans generating over 2 Gs of cornering grip for upside-down driving. Production is slated for 2028, though the practical purpose and legality of features like aerial corkscrews remain questionable.

👉 Read the full article on Gizmodo

The content of the provided URL was inaccessible. However, based on the article's title, it suggests that Meta has announced an open-weight AI model designed to run directly on Apple's Mac devices.

This announcement likely reflects Meta's strategy to democratize AI technology and expand its accessibility across a wider range of hardware. Local execution on Macs could offer benefits such as enhanced privacy, offline capabilities, and reduced latency compared to cloud-based solutions.

As an open-weight model, it is expected to accelerate innovation within the broader AI ecosystem by allowing developers and researchers to freely use, modify, and redistribute the model.

👉 Read the full article on TechCrunch