This Day in AI Podcast – EP99.05-FLASH

Date: May 23, 2025
Hosts: Michael Sharkey & Chris Sharkey
Episode Title: Opus & Sonnet 4, Google I/O Recap, Microsoft BUILD & Sam Altman Has A New Friend

Overview

In this episode, the Sharkey brothers dig into a week overflowing with AI industry news and product launches. The main focus is on Google's dramatic comeback at Google I/O, OpenAI's headline-grabbing Jony Ive partnership, Anthropic’s new Claude Sonnet 4 and Opus 4 models, and rapid advances in AI-generated video exemplified by Google’s new VO3 model. The brothers bring their trademark irreverence, skepticism, and “proudly average” curiosity to assess what’s hype, what's practical, and what's just plain weird in the evolving world of AI.

Key Discussion Points & Insights

1. OpenAI’s Acquisition of Jony Ive’s “AI Device Company” IO

• Context: OpenAI announced the acquisition of IO, a startup led by famed designer Jony Ive, for a reported $6.4 billion.
• The launch featured a high-production-value video of Sam Altman and Jony Ive, which Mike and Chris found almost satirical in its self-importance and lack of product substance.
  • "It’s a nine-minute video of them literally blowing smoke up each other’s asses. It’s next level cringe." – Mike (07:55)
  • Both questioned whether this move was about real innovation or more about optics, calling the acquisition money-laundering-adjacent and comparing it to modern art-level fundraising.
• The video’s timing was seen as a strategic attempt to distract from Google I/O news.
• Memes & Social Media Reaction: Communities have compared the Altman/Ive collaboration to past failed hardware launches (“Humane AI Pin”) and poked fun at Jony Ive’s career arc post-Apple.

2. Explosion of New AI Models

• Google:
  • Gemini 2.5, Gemini Flash 2.5, Gemini Ultra Subscription ($250/month), and new feature parity with OpenAI.
    • Mike believes Google’s current model suite leads the industry:
      "I do think they have the best models right now... VO3, Imagen 4 is stunning." (05:49)
  • Imagen 4: Strong prompt adherence and image quality.
    • Chris: "The attention to detail… is just remarkable. The lighting as well is beautiful." (29:03)
  • VO3: Breakthrough in aligning audio and video for hyper-realistic AI-generated videos.
• Anthropic: Claude Sonnet 4 and Opus 4 arrived with “parallel test time compute” for better tool calling—esp. in coding-related tasks.
  • Chris noted Opus has been underwhelming (“the brand is tarred, tarnished” – 73:50), but is hopeful Sonnet 4 will impress over time.

3. Hands-On with Google VO3 Video Generation

• Demo Reactions:

  • The brothers played real VO3 clips (starting at 11:59) and marveled at the quality—even music synchronization and "streamer" game footage are shockingly authentic.
    • "If you played this and said that's AI-generated, I would not really necessarily believe." – Mike (13:58)
  • Flaws remain in longer sequences and physics-heavy scenes, but generative video is leaping ahead in usability for YouTubers and content creators.
  • Pricing is steep, reflecting the computing power needed.

• Societal Impact:

  • Chris: "Doesn't it to some degree fill you with profound sadness... People are going to spend their lives consuming content that isn't even real." (19:55)
  • Both discussed the quick path from AI creative technology to fake evidence, propaganda, and the erosion of any meaningful reality online.

4. Gemini Diffusion: The Most Exciting (But Under-Hyped) Release

• Gemini Diffusion introduces diffusion-based text generation: "Instead of predicting text directly, they learn to generate outputs by refining noise."

• Offers near-instant code and document generation—2.8 seconds for full websites, suggesting new productivity paradigms.

• Chris: "Speed helps in so many ways... it looks cool. And your examples... I was like, wow, that is astonishing." (41:34)

• Implications:

  • Real-time UI/UX generation, instant prototyping.
  • Could change workflows—less context switching, far more iteration.
  • Next bottleneck: not model speed, but tool orchestration.

5. Model Context Protocol (MCP) and the "Tool Stack" Future

• All major LLM providers now embrace Model Context Protocol—AI agents with parallel tool access.

• Anthropic’s implementation in Claude Sonnet 4 and Opus 4 is notable for allowing broader, deeper tool calls during "thinking" steps.

• The real innovation will be in orchestration, permissioning, and specialized “skills”—not just model power.

  "None of the model updates matter. What matters is the intelligent integration of the MCP explosion."
  — Chris (105:34)

6. Application vs. Model Layer

• Google (and others) release a dizzying scattergun of overlapping new applications and APIs.
• Gemini core experience/UI is still awkward and disjointed compared to ChatGPT.
• As models converge, the brothers argue that app design—and especially tailored, productivity-focused integrations—will differentiate the next phase.
• Mike: “It’s the orchestration and the next... AGI moment for me is seeing these tools... working on different things for me throughout the day in the background.” (106:39)

7. Microsoft BUILD, Copilot, GitHub Updates

• Microsoft is now “the Switzerland of models,” adding support for Grok and open-sourcing Copilot prompts/tools.

• GitHub Copilot’s new agentic features (auto-PRs, agents fixing issues) are viewed with skepticism: the time required for QA and real comprehension isn’t eliminated.

  “If someone just submitted a hundred PRs to SIM theory, for example, I’d be like, you’ve just ruined my life."
  — Chris (82:02)

8. Financial Realities for New AI Search Players

• Perplexity’s Leaked Financials:
  • Despite high valuation, margins are razor-thin; huge discounts, high ad spend, and massive AWS bills dominate their P&L.
  • The brothers are skeptical of their long-term prospects versus Google’s omnipresent, free AI search integration.

    "Can this company really challenge Google in light of Google just putting this capability in search ... ?"
    — Mike (101:55)

9. Miscellaneous Highlights

• Joking about prank-calling pet groomers at scale using MCP + AI agents as a practical chaos test.
• Reflections on Google’s corporate culture: Too many teams, too many names, UX still lags.
• Apple’s likely path: Wait, assimilate, and then release a polished app-layer experience.

Notable Quotes & Memorable Moments (with Timestamps)

• On OpenAI’s Altman/Ive Video:

  “It’s a nine-minute video of them literally blowing smoke up each other’s asses. It’s next level cringe.”
  — Mike, 07:55

• On Google’s AI Supremacy:

  “I do think they have the best models right now... VO3, Imagen 4 is stunning."
  — Mike, 05:49

• On AI Video's Societal Impact:

  "People are going to spend their lives consuming content that isn't even real."
  — Chris, 19:55

• On The Real Future: Orchestration, Not Models:

  "None of the model updates matter. What matters is the intelligent integration of the MCP explosion."
  — Chris, 105:34

• On Microsoft Copilot-Agent Skepticism:

  “If someone just submitted a hundred PRs to SIM theory, for example, I’d be like, you’ve just ruined my life."
  — Chris, 82:02

• Gallows Humor About Perplexity's Finances:

  "Why even list it? It's embarrassing. I don't... these were clearly leaked, but 27 million [in discounts/refunds]..."
  — Mike, 95:17

• On Google's Comeback:

  "It’s like the kid that got bullied in high school and now has come out and is super rich, has a hot wife, and, like, is really confident now."
  — Mike, 93:34

• On The Application Layer vs. Model Layer:

  “As models converge, the brothers argue that app design—and especially tailored, productivity-focused integrations—will differentiate the next phase.” (paraphrased summary, multiple points)

Timestamps for Major Segments

| Segment | Timestamp | |-----------------------------------------------|-------------| | Altman/Ive IO acquisition + launch video | 0:02–11:33 | | Google I/O Recap—Gemini family | 11:33–27:00 | | VO3 AI Video Generation—Hands-on Demos | 11:59–24:43 | | Imagen 4 & Image Prompting | 27:12–34:24 | | Gemini Diffusion Super-Speed | 36:27–44:47 | | Orchestration + Model Context Protocol (MCP) | 61:55–72:07 | | Application design vs. model arms race | 50:30–55:14 | | Microsoft BUILD, Copilot updates | 77:13–86:34 | | Perplexity financials dissected | 94:18–100:52 | | Agentic models—future, chaos with MCPs | 87:34–93:18 | | Broader reflections, closing thoughts | 104:09–109:28 |

Concluding Thoughts

The brothers close by urging listeners to focus not on chasing every hyped model release, but instead to explore how to practically combine these models and tools in ways that truly improve personal productivity and creativity. They call for better, simpler tools (like their own Sim Theory platform) to help average users experiment at the forefront, and suggest the coming year will be marked as much by breakthroughs in orchestration and specialized use-cases as the raw power of underlying AI models.

"We're all focused on the practical... It's fun to get all excited about the announcements, but it's really about what can we do with it."
— Chris (107:29)

Listener Homework:

• Watch the Altman/Ive IO launch video for a lesson in modern tech narcissism
• Try out Google Gemini and VO3 models (if you can get access)
• Reflect: What could you orchestrate in your daily work with AI-powered tool stacks?
• Join the Sim Theory AI club to test-drive all the weirdness the AI world can muster—maybe even join the great pet groomer prank experiment.

Support the show: Subscribe, average reviews heartily welcomed; try Sim Theory and look forward to more practical chaos in AI experimentation.