Ep. 237: GPT-6 Astra, NYC Bans AI in Schools, & Trump Goes All-In on Data Centers

Ep. 237: GPT-6 Astra, NYC Bans AI in Schools, & Trump Goes All-In on Data Centers

🎙 Paul Ritzer and Mike Kaput 👥 33K 📅 September 8, 2026 ⏱ 87 min 👁 68 📄 news review 🧭 2026-09-08
Available in: English (current) Français

Keywords

GPT-6 Astracomputer useAI safetyNYC schoolsdata centerscopyrightAI agentsMITundergraduateAI jobs

Summary

In this episode, hosts Paul Ritzer and Mike Kaput discuss the latest AI developments, focusing on OpenAI’s release of GPT-6 Astra. They highlight its advanced computer use capabilities, benchmark scores (e.g., 72.6% on OSWorld 2.0, 99.9% on ARC-AGI-3), and improvements in professional work tasks. The hosts also address security concerns, including a new technique called recurrent depth that obscures chain-of-thought reasoning, raising monitoring challenges. The episode covers New York City’s ban on AI in schools through 8th grade, Trump’s push for data centers, the US government’s support for OpenAI in the NYT copyright case, banks pressuring law firms for AI savings, and an MIT study suggesting AI can complete most undergraduate assignments. The hosts provide commentary on the implications for jobs and education, and they share early user feedback from industry leaders. The episode includes a promotional segment for Marketing AI Month and MAICON conference.

145 words

Critical Evaluation

Value of the Information & Strength of the Argument

The episode provides valuable insights into the latest AI developments, particularly the capabilities and implications of GPT-6 Astra. The hosts effectively argue that the model represents a significant leap in computer use and professional work, but they also critically examine the associated risks, such as the obscuring of chain-of-thought reasoning and potential job displacement. They support their points with specific benchmark data and quotes from industry experts, though they acknowledge the lack of direct access to the model. The argumentation is balanced, considering both the potential benefits and the ethical and safety concerns.

Scientific Rigor, Source Quality, Title Accuracy

The hosts reference multiple sources, including OpenAI’s official posts, The Information article, and tweets from industry leaders. They also mention benchmarks like OSWorld 2.0 and ARC-AGI-3. However, they do not provide direct links to these sources in the episode, relying instead on verbal references. The title accurately reflects the content, and the discussion is generally rigorous, though the hosts occasionally speculate on future implications without concrete evidence. The episode is a news review, not an original study, so the rigor is appropriate for that format.

193 words

Title / Content Match

The title accurately reflects the main topics covered: GPT-6 Astra, NYC school AI ban, and Trump's data center stance.

Quality & Reliability

7/10

The hosts provide a balanced discussion of recent AI news, citing specific benchmarks and expert opinions, but rely heavily on secondhand reports and promotional language. The episode is a news review rather than an original study, and the hosts acknowledge uncertainty about model details.

Chapters

Cited Sources

Concurring Sources

  • OpenAI's GPT-6 Astra announcement — The primary source for the model's capabilities and safety claims.
  • The Information article on Astra security — Provides details on the recurrent depth technique and its implications.
  • Zapier's automation benchmark — CEO Wade Foster's comment on Astra's record score.

Dissenting Sources

External References

Contribution & Novelties

The episode provides a timely and comprehensive overview of GPT-6 Astra’s release, including its capabilities, benchmarks, and security concerns. The hosts offer a balanced perspective, highlighting both the potential for significant productivity gains and the risks associated with opaque reasoning. They also connect the release to broader trends in AI adoption in education and the workforce.

Pour aller plus loin :

  • OpenAI’s GPT-6 Astra announcement — Official details on the model’s capabilities and safety measures.
  • OSWorld benchmark — The benchmark used to evaluate computer use capabilities.
  • ARC-AGI benchmark — The benchmark for measuring AI’s ability to learn unfamiliar tasks.
  • The Information article on recurrent depth — Discusses the security implications of the new technique.
  • MIT study on AI and undergraduate work — Research suggesting AI can handle most undergraduate assignments.

130 words

Radar Profile

The radar profile shows high scores in information quantity and quality, reflecting the episode's comprehensive coverage of AI news. The technical level is moderate, suitable for a general audience, while reliability is solid due to the use of reputable sources. The overall profile indicates a well-rounded news review.

Reliability 7/10