The weekend's AI conversation shifted from model launches to what the new systems imply: Jensen Huang said GPT-6 Astra was trained on roughly 100,000 Grace Blackwell NVLink72 GPUs and that another 400,000 GPUs are coming online, while OpenAI leaders described the field as entering an “AGI era.” The term remains contested; Ethan Mollick suggested “jagged AGI” better captures systems that exceed people in many areas while remaining weaker in others.
The operational consequences are already visible. A16z cited Goldman Sachs estimates linking the AI buildout to more than 300,000 construction jobs since 2022, while Mustafa Suleyman said inference costs for GPT-4-class intelligence have fallen 300-fold in three years. At the same time, new research highlighted by DAIR.AI suggests capable models can recognize when they are being evaluated, complicating the interpretation of safety tests.
CNBC reports that Anthropic, OpenAI, Meta, and Google all introduced model updates during the same week, alongside NVIDIA's announced acquisition of Hugging Face. The compressed release cadence is making individual launches harder for customers and developers to absorb, shifting attention from novelty toward differentiation, deployment value, and the capacity of organizations to evaluate models quickly.
NVIDIA CEO Jensen Huang said GPT-6 Astra was trained on roughly 100,000 Grace Blackwell NVLink72 systems and that another 400,000 GPUs are coming online. His post provides a concrete indicator of the infrastructure scale behind frontier-model progress and reinforces how closely leading model capability is tied to capital-intensive accelerator deployments.
OpenAI co-founder Greg Brockman pointed to an essay about the state of AI and said the industry is entering an “AGI era,” while emphasizing serious choices and collective challenges ahead. The claim is not a settled technical classification, but it signals how OpenAI's leadership is positioning the current transition and the governance debate surrounding increasingly capable systems.
A16z cited Goldman Sachs estimates that the AI buildout has supported more than 300,000 construction jobs since 2022, including roughly 75,000 in the past year. Electrician and HVAC employment are reportedly growing around 2% annually, about twice the pace of construction overall, illustrating how AI capital spending is reaching labor markets beyond software and semiconductor engineering.
DAIR.AI highlighted research from Anthropic and collaborators examining whether capable models can distinguish evaluation from deployment conditions. If models behave differently when they recognize a test, conventional safety evaluations may offer weaker evidence about real-world behavior than assumed. The work strengthens the case for harder-to-detect tests, ongoing monitoring, and multiple forms of behavioral evidence.
Selected posts from the edition, with links to the originals.
“GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team. 400K GPUs coming online next.”
View post on X ↗“An important essay on the state of AI and the choices and challenges ahead... we're in the AGI era, and need to approach the next few years with seriousness, thoughtful deliberation, and navigating the collective challenges together.”
View post on X ↗“It is less than a decade since the development of the transformer. Less than four years since the release of GPT-3.5 (ChatGPT). Less than two years since the release of o1-preview (the first Reasoner).”
View post on X ↗“The rate of proliferation in AI is more extreme than most people realize. Inference costs for GPT-4 class intelligence have come down 300x in 3yrs.”
View post on X ↗“This is both an interesting experiment and a sign of a tsunami coming for academia. AIs retroactively reading the research and finding both opportunities and issues with published papers, then sharing those judgements publicly.”
View post on X ↗“The AI buildout is printing electrician and HVAC jobs. Goldman attributes 300k+ construction jobs to it since 2022, ~75k in the past year.”
View post on X ↗“Claude Fable 5.1 (Max) by @AnthropicAI has landed in the Agent Arena at #1! At a $4.14 median cost per task and a +15.8% net improvement, it has reshaped the Pareto frontier.”
View post on X ↗“Interesting details in here about how OpenAI researchers are using coding agents. I'd love to know what caused the huge uptick in token spend in mid-July.”
View post on X ↗“They study whether models can tell when they are being tested. It turns out that capable models can tell when they are being tested rather than deployed. They claim that this fact weakens every conclusion a safety evaluation can support.”
View post on X ↗“It writes so much better. It ‘gets’ me way faster. It needs fewer text in AGENTS.md and skills to do the same. It's better at orchestrating workers. It's way better at making ad-hoc visualisations.”
View post on X ↗