Some thoughts on OpenAI.

ChatGPT §

• OpenAI was surprised by its success¹
• in fact, the hype train it created was so big, that it single-handedly re-ignited the hype of the whole AI industry
• They reacted well and jumped on the train
  • Sam Altman took the chance and transformed OpenAI from a non-profit research lab into a full-blown product company
  • The transition was turbulent, shifting identity and goals
  • Ultimately, Sam Altman and his team succeeded, pushing out opposing parties
• ChatGPT is now a well-known product
• Consumers see it as a smart chatbot that boosts productivity
  • However, consumers don’t understand the technology behind it
  • To them, it’s a magical black box capable of creating images, writing code, answering questions, etc.
    • In reality, it’s many parts working together seamlessly
  • OpenAI intentionally obfuscates ChatGPT’s workings to:
    • Avoid giving competitors an advantage
    • Make ChatGPT seem like magic
      • This fuels fears about AGI
        • People worry about job security, the future, and AI dominance
      • i.e. it’s the best advertising they could wish for (see AGI)

Limits §

• OpenAI (and everyone else) is running out of training data for GPTs²
• Still, I don’t think that training data is the biggest problem, as there will probably be other ways of optimizing and improving models
• However, these improvements will take more time and effort than ‘just’ training a model with more data
• I also doubt one could reach any form of strong AI, which would pose the threat of Singularity using the GPT approach
  • Training GPT models with human artifacts will approximate performance, not exceed it
• don’t get me wrong, the current tech is indeed a major breakthrough
  • it single-handedly dropped the cost of (low-medium) intelligence
  • with that drop in cost, there will be major movements in the economy (and ones every day life)
  • I even built multiple projects myself using the OpenAI API myself, it’s great
  • however, its progress does not scale
    • they can’t just ‘train a new model longer, with more data’ to make it twice as smart forever
    • that’s their problem
• Energy consumption and economic viability will also limit GPTs’ success
  • Currently, VC funding covers costs
  • Investors will eventually expect returns
  • OpenAI’s current pricing and free offerings won’t last forever³
  • When costs rise, excitement and expectations for GPTs will decrease

AGI §

• OpenAI does not have a master plan
• It’s just a rapidly growing startup riding a wave of success from an experiment
• Initially, they aimed to build AGI
  • This goal remains distant (see Limits)
• Instead, they are currently riding the ChatGPT hype wave
  • They must maintain this momentum to avoid failure
    • They do this by:
      • Indirectly fearmongering about AGI
      • Obfuscating product workings to make them seem magical
      • Trying to monopolize popular AI terms to direct hype and energy towards them⁴
      • Generating revenue through investments and building successful products atop ChatGPT
• Furthermore, the pressure created by the ChatGPT hype squashed their (public) timeline of achieving AGI
  • This is evident in Sam Altman’s interviews, pre- and post-ChatGPT
  • Pre-ChatGPT, interviews were informative and targeted at experts, with AGI being a distant goal
  • Post-ChatGPT, interviews feel like pop-science
    • The focus is on grand visions of an AGI-driven world
    • These ‘interviews’ are disguised advertisements
      • Targeting well-educated audiences
      • Assuming AGI (created by OpenAI) as an inevitable future
      • Portraying OpenAI as an open research lab on the brink of achieving AGI, inviting public participation through prototype testing
      • In reality, they are a for-profit company focused on product creation and revenue
      • They rely on the research lab image to function
• Therefore, concerns about AGI or ethics are often just marketing
  • It’s similar to how Nike sells sportswear, making consumers feel like high-performance athletes through advertisements

Competition §

• OpenAI dislikes competition
• As noted in ChatGPT, they aim to create a monopoly
  • They want ‘OpenAI’ to be synonymous with ‘AI’
• They hinder smaller competitors by calling for AI regulations⁵
  • To make regulators move faster, they pull out their good old ‘AGI is coming’ trump card
  • These regulations would hurt OpenAI but devastate smaller competitors

Drama §

• OpenAI’s transformation has generated much noise
• This drama feels like a reality show for Zoomers
• It attracts media attention, creates discussions, and emotions
• This isn’t necessarily bad for OpenAI, as they benefit from the attention (and attention is all you need after all ;) )
• Important information remains inside the company through lifetime NDAs⁶

Links §

Other

• OpenAI researchers warned board of AI breakthrough ahead of CEO ouster | Hacker News (ycombinator.com)
• After OpenAI’s blowup, it seems pretty clear that ‘AI safety’ isn’t a real thing | Hacker News (ycombinator.com)
• OpenAI’s Long-Term AI Risk Team Has Disbanded | WIRED

Footnotes §

1. OpenAI Cofounder Surprised by ChatGPT’s Massive Success (businessinsider.com) ↩

2. CB Insights on X ↩

3. OpenAI is too cheap to beat | Hacker News (ycombinator.com) ↩

4. OpenAI – Application for US trademark “GPT” has failed | Hacker News (ycombinator.com) ↩

5. OpenAI Lobbied the E.U. To Water Down AI Regulation | Hacker News (ycombinator.com) ↩

6. OpenAI departures: Why can’t former employees talk? | Hacker News (ycombinator.com) ↩