The Artificial Intelligence Show

Episode 170: "How ChatGPT Is Used at Work, New GDPval Benchmark, AI “Workslop,” ChatGPT Pulse, Meta Vibes & More AI Economy Warnings"

Hosts: Paul Roetzer and Mike Kaput
Date: September 30, 2025

Overview

In this jam-packed episode, Paul and Mike dissect the latest seismic shifts in AI technology and its effects on business, the workforce, and the economy. Major topics include OpenAI’s massive enterprise impact, a revolutionary evaluation benchmark (GDP VAL), the rise of “workslop” (low value AI-generated output), the new ChatGPT Pulse feature, Meta’s launch of Vibes, jaw-dropping numbers in AI infrastructure investment, and sobering warnings from business leaders about AI-induced job transformations.

The tone is candid, occasionally humorous, forward-looking, and oscillates between optimism about AI’s transformative potential and concern over its rapid, sometimes disruptive real-world impact.

Key Discussion Points and Insights

1. ChatGPT’s Explosive Workplace Adoption and Impact

[06:36-16:03]

• OpenAI Report: Detailed new data shows ChatGPT as the fastest adopted enterprise tech ever.

  • 28% of US workers use ChatGPT at work, rising to 45% for those with graduate degrees.
  • Usage is highest among IT/tech, followed by professional services, manufacturing, finance.
  • Healthcare lags due to privacy and compliance, but is showing signs of uptake.
  • Main tasks: Writing, research, programming, analysis.

• Adoption Patterns:

  • Technical teams focus on deep research and coding; go-to-market teams leverage AI for writing and ideation.
  • Most users stick to simple tasks; “power users” leverage advanced features like custom GPTs and deep research, but this is not yet mainstream.
  • There’s a significant knowledge gap between those on the AI frontier and the average worker.

• Emerging “AI Generalist” Role:

  • AI skills enable professionals to cross operational boundaries, increasing individual impact within organizations.
  • The shift toward generalists (who can leverage AI across domains) is only starting.

• Quotes:

  "If everyone else has that knowledge, then everything changes basically overnight." – Paul [00:00] "The average worker does not know how to build a custom GPT… that is not normal." – Paul [10:34] "You can really just establish yourself as such a leader in your organization." – Mike [13:58]

2. GDP VAL Benchmark: Measuring Real-World AI Job Performance

[16:03-29:16]

• OpenAI’s GDP VAL: A new framework that evaluates AI models on their ability to perform real-world knowledge work across 44 job roles (e.g., engineers, financial analysts, nurses).

  • Expert evaluators compared AI vs. human outputs blind; current models matched/exceeded human expert output nearly half the time.
  • AI completed tasks 100x faster and 100x cheaper than human experts.

• Automated Grader: AI trained to predict expert evaluations, potentially revolutionizing educational/classroom assessment.

• Industry Takeaways:

  • Significant implications for professional quality jobs.
  • AI performance is starting to rival—and even surpass—human experts, especially as reasoning and “agentic” capabilities increase.

• Prominent AI Researcher Weighs In:

  • Julian Schrittwieser (Anthropic cofounder) predicts:
    • By mid-2026, models will autonomously work full 8-hour days.
    • By end-2026, at least one model will match human expert performance widely.
    • By 2027, models will frequently exceed human experts.

• Economists Sound the Alarm:

  • New papers urge rapid research into transformative AI’s effects on growth, employment, and social structure.

• Quotes:

  "Frontier models can complete GDP VAL tasks roughly 100 times faster and 100 times cheaper than industry experts." – Paul [21:42]
  "It would be extremely surprising if these improvements suddenly stopped... 2026 will be a pivotal year for widespread integration of AI into the economy." – Julian Schrittwieser (paraphrased by Paul) [24:25]

3. AI “Workslop”: The Hidden Cost of Low-Value AI Output

[30:43-40:34]

• Definition (“Workslop”):
  AI-generated content that appears polished but is context-poor, inaccurate, or requires significant rework—creating hidden inefficiency.

• Research (Harvard Business Review/BetterUp Labs):

  • 40% of employees encountered “workslop” in the past month.
  • On average 15% of received work fits this category.
  • Each “workslop” case costs nearly 2 hours to untangle; damages trust in colleagues.

• Underlying Problem:

  • Not AI’s capabilities, but improper use—people often skip reviewing or refining outputs, leaving coworkers to detect and fix errors.

• Impact:

  • Colleagues see workslop senders as less capable/trustworthy.
  • Points to the urgent need for responsible AI training and effective change management, not just tool adoption.

• Memorable Moments:

  "Workslop is like my favorite phrase of 2025. I think I need that on T-shirts." – Paul [32:48]
  “Your deliverable is just something ChatGPT produced. What point do you have in the process?... Your job is what comes after ChatGPT.” – Mike [36:46]

• Anecdote:
  Paul recounts asking Gemini to generate an image for his editorial. The AI intentionally misspells “detected” on a “workslop detected” stamp. Was it a meta joke or emergent behavior?

  "I thought, holy… Did it only then realize it and provide the rational after the fact? My guess is no. I think it actually intentionally did it. And that's wild." – Paul [38:37]

4. ChatGPT Pulse: AI as a Proactive, Personalized Assistant

[40:37-47:54]

• Pulse Feature:
  ChatGPT now proactively provides personalized daily updates, curating visual cards based on your chat history, connected apps, and explicit interests.

• How It Works:

  • Scans your chat, calendar, Gmail, etc.
  • Suggests everything from dinner ideas to reminders and project updates.
  • Learns preferences through user feedback and specific topic requests.

• Strategic Implications:

  • Pulse marks the transition from reactive to proactive assistants, previewing the future “operating system” for personal and work life.
  • Potential ad model: "The business side of this is tremendous... the instant you start using it, you see it’s truly building a knowledge graph that predicts interests, behaviors, buying habits, fears, concerns, political leanings—like everything." – Paul [42:28]
  • Could upend search, SEO, and content discovery.

• Quote:

  "Go back and re-listen to those 150 words I just read from the thing or go read it yourself. That is a roadmap for where they're going." – Paul [45:10]

5. Staggering AI Infrastructure Build-Out: OpenAI x Nvidia & the Era of “Intelligence as a Utility”

[48:00-54:20, 76:04]

• OpenAI & Nvidia Announce 10GW Partnership:

  • Equivalent to 10 nuclear reactors, costing up to half a trillion dollars.
  • OpenAI’s Sam Altman targets 250GW of compute by 2033 (requiring 250 nuclear plants), costing ~$12.5 trillion.

• Rationale:

  • Anticipated exponential demand for “intelligence” in every workflow, device, and app—not just model training, but ongoing AI-powered services.

• New Utility Model:

  • Likely eventual pay-as-you-go system, with customers metered for actual compute/intelligence consumed—mirroring APIs, cloud billing, etc.
  • Drives the business case for replacing (more expensive) human labor with always-on, scalable AI.

• Quote:

  "When you give a CEO or a board... you can spend $50,000 a year for a customer support agent, or you can pay a human who needs to sleep... $120,000. Like, I'm sorry, but... they're taking the $50,000 agent seven days a week... It just seems too logical." – Paul [53:24]

6. Meta Launches Vibes: Fully Synthetic Short-Form Video Feed

[54:20-58:27]

• What is Vibes?
  TikTok-like scroll experience of AI-generated videos, powered by tech from Midjourney and Black Forest Labs and living inside Meta AI.

• Community Reaction:
  Overwhelming negative sentiment: "flooded with comments calling it things like 'AI slop' and asking who actually wanted an endless scroll of synthetic content." – Mike [55:28]

• Host Reaction:
  Paul is critical, seeing this as a weak debut for Meta’s new chief AI officer and a misalignment with their stated “personal superintelligence” vision.

  "Who wants this? Who needs this? Like how is this toward the good of humanity and like personal intelligence that benefits everyone? Just getting sucked into like scrolling non stop through AI generated stuff. Like I don't know, it, it just makes no sense." – Paul [57:00]

7. AI Safety and Jobs: Parental Controls, CEO Warnings

[58:27-68:34]

• OpenAI Adds Parental Controls:

  • Parents can link accounts to set time limits, manage features, and get safety alerts if AI chat detects serious risks.

• Companies Warning About Job Displacement (Walmart, Accenture, SAP):

  • Walmart CEO: “AI is going to change literally every job.” Headcount to remain flat as roles are transformed or eliminated.
  • Accenture is “exiting” untrainable employees, reinvesting in AI-literate talent.
  • SAP CFO: “I will be brutal… If you are left behind, you will have a problem for sure… AI lets SAP produce more software with fewer people.”

• Hosts Discuss Elite Blind Spots:

  • Many tech/AI leaders underestimate the societal costs of “transition periods.” Their optimism rings hollow for millions without job or retraining prospects.

  • “Transition is like the new go to word to describe the really painful part where nobody has jobs.” – Paul [66:26]

  • “Let’s also recall that the Great Depression was a period of transition and there was a decade of, of almost global disaster…” – Mike [68:14]

8. Expert Data Labeling & The Human-in-the-Loop Economy

[69:30-72:11]

• Merkur Case Study:

  • Interview with founder Brandon Foody (age 22): From $1M to $500M revenue in 17 months by connecting human experts to AI labs for high-precision data labeling and training.
  • Model of “reinforcement learning economy”—humans in the loop until AI surpasses their expertise.

• Recommendation:

  • Listen to 20VC podcast for venture capital perspective and more on this topic.

9. Rapid Product & Platform Updates

[72:11-75:49]

• Noteworthy launches:
  • Hux: Personalized AI audio briefings (similar to ChatGPT Pulse).
  • OpenAI adds team projects, expertise routing, and integration with Google Drive/Salesforce for enterprise users.
  • Apple quietly testing new LLM-powered ‘Siri-S’ internally.
  • Microsoft 365 Copilot now offers Claude 3.5 as an option and launches Analyst agent for data tasks.
  • Spotify takes down AI-generated music that clones artists without permission.
  • Clip Axiom (Palantir alumni) raises at $1.8B valuation for military/intelligence AI.
• Side note: Paul flags team projects as a future strategic move to challenge Office/Google Workspace.

Memorable Quotes and Moments

• On the transformational pace:

  “The knowledge we all have and are gaining every day is like unparalleled in human history. Like how quickly we can learn things and adapt.” – Paul [15:00]

• On the commercialization of AI:

  “The instant you start using it [Pulse], you see it’s truly building a knowledge graph that predicts interests, behaviors, buying habits, fears, concerns, political leanings—like everything.” – Paul [42:28]

• On challenges for average workers:

  “The disconnect, I think, that exists with the elite class... versus the reality of the worker, like the average working family.” – Paul [66:26]

• On responsible use:

  “Friends don’t let friends produce workslop. Like, consider this a policy for SmarterX. Like, we are not going to be doing this.” – Paul [36:31]

Timestamps for Major Segments

• [06:36] – OpenAI report: ChatGPT’s enterprise adoption
• [16:03] – GDP VAL: New benchmark for real-world AI job performance
• [30:43] – The emergence and risks of “Workslop”
• [40:37] – ChatGPT Pulse: personalized, proactive AI assistants
• [48:00] – OpenAI x Nvidia’s $500B AI infrastructure plan and Altman’s ambition
• [54:20] – Meta’s Vibes: all-AI short video feed launch
• [58:27] – OpenAI parental controls and weekly economy/job updates
• [69:30] – Interview insights: 20VC, expert data labeling economy
• [72:11] – Rapid fire product and platform news

Conclusion

This episode captures a pivotal moment as AI moves from novel tool to economic engine, with capabilities accelerating faster than businesses—and society—can fully absorb. The hosts stress the need for responsible, optimistic leadership, AI literacy, and proactive adaptation while cautioning against blind optimism and the human cost of “transition.” The future will demand new skills and roles, but also empathy and thoughtful guidance to navigate the upheaval ahead.

Stay curious and keep exploring AI.