Stenobird

Episode

#241 - Opus 4.7, Muse Spark, GPT-5.4-Cyber, HY-World 2.0

Podcast
Last Week in AI
Published
Apr 23, 2026
Duration seconds
7188
Processing state
processed
Canonical source
https://rss.art19.com/episodes/9d913404-a06c-4f1e-94d4-e8008012aa65.mp3?rss_browser=BAhJIg90cmFuc2NyaWJyBjoGRVQ%3D--952c5701c84ad333c69d5faa668f8177091704f0
Audio
https://rss.art19.com/episodes/9d913404-a06c-4f1e-94d4-e8008012aa65.mp3?rss_browser=BAhJIg90cmFuc2NyaWJyBjoGRVQ%3D--952c5701c84ad333c69d5faa668f8177091704f0
JSON
/v1/public/podcasts/last-week-in-ai/episodes/241-opus-4-7-muse-spark-gpt-5-4-cyber-hy-world-2-0
Markdown
/podcast/last-week-in-ai/241-opus-4-7-muse-spark-gpt-5-4-cyber-hy-world-2-0.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/last-week-in-ai/episodes/241-opus-4-7-muse-spark-gpt-5-4-cyber-hy-world-2-0/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/last-week-in-ai/241-opus-4-7-muse-spark-gpt-5-4-cyber-hy-world-2-0.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

A deep dive into the rapid release cycle of frontier models, including Anthropic's Claude Opus 4.7 and Meta's Muse Spark. The discussion explores the shift from pure model scaling to agentic workflows, infrastructure expansion, and the geopolitical risks of AI-driven propaganda.

Topics

  • Anthropic
  • Meta AI
  • OpenAI
  • Large Language Models
  • AI Infrastructure
  • AI Agents
  • Cybersecurity
  • Machine Learning Research

Highlights

  • Main idea: The frontier model race has entered a new phase of 'test-time scaling' and specialized reasoning controls
  • Practical takeaway: Developers should monitor updated tokenizers in new models like Opus 4.7, as they can significantly impact token usage and costs
  • Failure mode: Automated weak-to-strong supervision research shows that current automated researchers still struggle to close the performance gap with ground truth labels
  • Business trend: Large-scale infrastructure projects, like Meta's multi-billion dollar data center plans, are becoming the primary bottleneck for AI scaling
  • Geopolitical risk: The rise of AI-generated propaganda and deepfakes is creating new asymmetric vulnerabilities in international cyber warfare

Chapters

  1. 10:05 Anthropic's Opus 4.7 Release: Analysis of Claude Opus 4.7's benchmark performance, new reasoning controls, and the implications of its updated tokenizer on usage costs.
  2. 28:25 Meta's Muse Spark and Hyperion: A look at Meta's new closed model, its 'contemplating mode,' and the massive 2-gigawatt infrastructure plans for the Hyperion data center.
  3. 38:10 OpenAI's Cyber and Codex Updates: Details on GPT-5.4-Cyber for security teams and the expansion of Codex capabilities into browser use and long-horizon task scheduling.
  4. 47:10 Market Consolidation and Mergers: Discussion on the potential Cohere and Aleph Alpha merger and the challenges facing European AI players in the scaling race.
  5. 56:30 Compute Infrastructure and CoreWeave: The growing importance of compute deals and the impact of high-stakes infrastructure investments on companies like CoreWeave.
  6. 1:24:30 AI Propaganda and Global Security: Examining the use of AI-generated video in geopolitical conflicts and the increasing risks of automated disinformation campaigns.
  7. 1:33:55 The Limits of Automated Supervision: Reviewing recent research on the difficulty of using automated agents to achieve human-level supervision performance.