Summary:

• GPT-6 Astra scores 62.7% for $26K on ARC-AGI-3 Semi-Private with our Standard harness, and 99.9% for $19K with a Provider Adapter harness.
• GPT-6 Astra surpasses the human baseline in action efficiency on ARC-AGI-3. It used fewer actions than the median tested human on 96% of levels.
• A key behavior observed in GPT-6 Astra was its ability to turn unfamiliar environments into compact symbolic world models. It represented game mechanics as logical rules and developed its own domain-specific language shorthand to track state and plan actions.

For a cost comparison, during our controlled testing, human participants were paid $115 per 90-minute session, plus $5 per game completed. Participants attempted approximately nine games per session, roughly $12.78 per attempted game before bonuses.
Most of this fee pays for the participant’s time and willingness to take the test, rather than the energy their brain uses (a closer proxy to compare with AI). If we look at only the brain’s energy, and price it as electricity, the estimate drops to about 0.6 cents per session, or 0.067 cents per game attempted.1

Astra’s results are also a major milestone worth celebrating. From our perspective, Astra represents a noticeable step-function change in frontier model capabilities.

When we launched ARC-AGI-3, we made it clear that saturating the benchmark would not represent “proof of achieving AGI.” Therefore, while we believe Astra represents meaningful progress towards generalization, we are not claiming that it is AGI.

Summary:

• GPT-6 Astra scores 62.7% for $26K on ARC-AGI-3 Semi-Private with our Standard harness, and 99.9% for $19K with a Provider Adapter harness.
• GPT-6 Astra surpasses the human baseline in action efficiency on ARC-AGI-3. It used fewer actions than the median tested human on 96% of levels.
• A key behavior observed in GPT-6 Astra was its ability to turn unfamiliar environments into compact symbolic world models. It represented game mechanics as logical rules and developed its own domain-specific language shorthand to track state and plan actions.

For a cost comparison, during our controlled testing, human participants were paid $115 per 90-minute session, plus $5 per game completed. Participants attempted approximately nine games per session, roughly $12.78 per attempted game before bonuses.
Most of this fee pays for the participant’s time and willingness to take the test, rather than the energy their brain uses (a closer proxy to compare with AI). If we look at only the brain’s energy, and price it as electricity, the estimate drops to about 0.6 cents per session, or 0.067 cents per game attempted.1

Astra’s results are also a major milestone worth celebrating. From our perspective, Astra represents a noticeable step-function change in frontier model capabilities.

When we launched ARC-AGI-3, we made it clear that saturating the benchmark would not represent “proof of achieving AGI.” Therefore, while we believe Astra represents meaningful progress towards generalization, we are not claiming that it is AGI.

Lawyers representing victims and survivors of the Tumbler Ridge school shooting say they are filing 30 new lawsuits against OpenAI in U.S. federal court in California.

The lawsuits accuse OpenAI of "choosing profit over the lives of the children of Tumbler Ridge" as it prepares to become a publicly traded company.

[Jay Edelson] said OpenAI has so far refused calls to make the logs public, a choice he believes indicates their contents may not work in the tech company's favour.

Lawyers representing victims and survivors of the Tumbler Ridge school shooting say they are filing 30 new lawsuits against OpenAI in U.S. federal court in California.

The lawsuits accuse OpenAI of "choosing profit over the lives of the children of Tumbler Ridge" as it prepares to become a publicly traded company.

[Jay Edelson] said OpenAI has so far refused calls to make the logs public, a choice he believes indicates their contents may not work in the tech company's favour.

University’s move is part of trend that critics within academia say will lead to further staff cuts and the loss of ‘everything that makes the job worth doing’

One student, who asked to remain anonymous, said they felt “short-changed” after issues with their degree structure forced them to enrol in the online option.

“Instead of a group discussion or a tutor teaching you, it’s a class activity done with a robot,” the student said. “I feel like I’m teaching this robot how to take jobs away from my teachers.”

The StatsCan report: Perceptions of gender-based violence and gender equality, identity and expression in Canada, 2025

Excerpt:

A new Statistics Canada report says Canada is becoming less accepting of gender diversity.

In 2025, 77 per cent of women and 70 per cent of men surveyed said they agreed or strongly agreed that people should be free to express their gender however they choose.

That’s a drop from 2018, when 85 per cent of women and 78 per cent of men agreed.

In Saskatchewan, 69 per cent of women and 60 per cent of men surveyed last year said they agreed, making it the province with the lowest acceptance of gender diversity in the country.

Women in Saskatchewan specifically were down 14 percentage points from 2018, when 83 per cent agreed people should be free to express their gender however they choose.

Since 2018, Saskatchewan and Alberta's provincial governments have passed legislation targeting trans and gender diverse youth, such as Saskatchewan’s Parents’ Bill of Rights, enacted in 2023.

Here's the petition: https://www.ourcommons.ca/petitions/en/Petition/Details?Petition=e-7531

The top three solutions come from independent researchers. The best solution was built by a group of PhDs and professors, who released a corresponding paper. They all make use of some form of world-model.

I've generally been a skeptic, and I still am, but this news surprised me because I expected ARC-AGI-3 to remain difficult for a long while.

Note that the scores are self-reported and need to be independently verified. The solutions have not been tested against the larger private test set.

Primer on ARC-AGI-3:

ARC-AGI-3 is an interactive reasoning benchmark which challenges AI agents to explore novel environments, acquire goals on the fly, build adaptable world models, and learn continuously.

A 100% score means AI agents can beat every game as efficiently as humans.

Instead of solving static puzzles, agents must learn from experience inside each environment—perceiving what matters, selecting actions, and adapting their strategy without relying on natural-language instructions.

Source https://www.reddit.com/r/Wellthatsucks/comments/1v5elmq/forgetting_to_remove_the_ai_prompt_from_your/

The guy is Bill Oliver https://en.wikipedia.org/wiki/Bill_Oliver_(politician)

Amnesty International Canada is opposing the project by Meta, which is anticipated to take up an area larger than Vancouver’s Stanley Park in Sturgeon County.

“This is a highly unregulated industry that operates on a surveillance-based business model,” Tara Scurr, Amnesty International Canada’s Corporate Accountability and Climate Justice Campaigner.

Some families in Georgia are being forced to sell their homes or face government seizures to make way for powerlines. 70-80% of that power is needed for AI data centers.

In the past year, Canada's immigration rate has experienced a dramatic reversal. We explain how it happened.

Also on Nebula: https://nebula.tv/videos/tldrnewsglobal-canadas-insane-immigration-uturn-explained