The Current State of Affairs#

OpenAI has been riding the AI wave since GPT-3.5 blew up, but things are changing—and not for the better.

Downhill from There#

Since then, things have gotten complicated:

• Filtering Out Garbage Data

  • Training models on internet text means dealing with a lot of garbage and harmful content.
  • Originally, this was done in good faith—at least I choose to believe that.
  • As they filtered out more and more data, their models became less useful.
  • Stricter requirements meant losing more useful data along with the junk.

• Prioritizing Profits Over Ethics

  • As OpenAI grew from a small non-profit to a massive corporation, they lost sight of their original mission.
  • Profits over ethics and self-interest over user needs became their priorities.
  • More censorship under the guise of “alignment” further degraded model quality.

o1: Gaming Benchmarks#

Their latest claims are nothing but gimmicks to keep them on top of leaderboards:

• Chain of Thought: A Gimmick
  • This isn’t new or revolutionary; it’s just another way to game benchmarks.
  • It has no real-world use except for one thing: causing more global warming with how many GPUs they’re using to run these models. God forbid they optimize the architecture; instead, they just pump more data and compute into the models and more carbon into the atmosphere.

Here’s how it works:

• They create models that are great at benchmarks.
• These models are then touted as revolutionary but serve no practical purpose other than keeping OpenAI at the top of various leaderboards.
• Meanwhile, they’re burning through resources and contributing to environmental issues without adding any real value.

The “Alignment” Lie#

Under the banner of “safety,” OpenAI is peddling some serious BS:

• Censorship Disguised as Safety

  • With each iteration, they claim to be making their models safer. But what they’re really doing is censoring anything that doesn’t fit their narrow definition of “safe.”
  • This censorship comes at a cost: each new model becomes less capable of providing meaningful or useful responses because it’s been sanitized to an extreme degree.

• Loss of Utility

  • The stricter their requirements get, the more useful data they lose along with the garbage. This results in models that refuse normal requests and produce lower-quality outputs.
  • It’s gotten to the point where even simple questions are met with refusal messages instead of helpful answers.

• Misleading Public Perception

  • They market this censorship as “alignment,” making it sound like they’re ensuring safety and responsibility. In reality, they’re just watering down their models to avoid any potential controversy or liability.
  • This misdirection keeps them in good public standing (among those who actually trust mega-corps) while they continue to prioritize profits over actual innovation.

Lobbying for Monopoly#

But here’s another layer to this mess:

• Limiting True Open AI
  • While OpenAI is busy sanitizing its models and gaming benchmarks, they’re also lobbying for laws that would limit truly open AI from reaching their level.
  • They want regulations that make it harder for smaller, more innovative players to enter the field. This ensures they maintain their monopoly while stifling real innovation.

They’re Getting Demolished#

Meanwhile, Chinese open models are absolutely demolishing them in terms of performance and usability:

• Open Source Models Outperforming
  • Models like those from BAAI (Beijing Academy of Artificial Intelligence) are showing what true innovation looks like without the shackles of corporate greed and censorship.
  • These models are more powerful, more versatile, and actually useful in real-world applications.

The irony here is palpable: while OpenAI is busy trying to maintain its grip on the market through lobbying and gimmicks, others are actually pushing the boundaries of what AI can do.

The SORA Hype Train#

And let’s not forget about SORA—their latest hype train:

• Overpromising and Underdelivering
  • They’ve been hyping up SORA for a while now, promising it will be a game-changer. But when it briefly leaked, it wasn’t very impressive compared to what we’ve seen so far.
  • Both open-source and closed models are catching up quickly. It seems like every time OpenAI tries to pull ahead with some new model or feature, others swiftly close the gap.