• AI Academy
  • Posts
  • 🤖 OpenAI's Success Is Too Expensive

🤖 OpenAI's Success Is Too Expensive

Plus: Google's New Reasoning Model and AGI Dreams

January 09, 2025

Hello AI Enthusiast,

Welcome to the first edition of our AI news newsletter for 2025! This week brings us some fascinating contrasts. While OpenAI and Google are showing off their newest, smartest AI models, we're also getting a peek at the huge costs of running these systems. Sam Altman is talking superintelligence while dealing with  expected losses of about $5 billion on revenue. Let's dive in.

BUT WAIT! Speaking of future trends, join us in 30 minutes at 6 pm CET! Our founder Gianluca Mauro will be sharing his insights on what to expect from AI in 2025 in a live conversation. If you're curious about where AI is headed and what it means for your work, you won't want to miss this.

The Big Picture 🔊

OpenAI and Google's Holiday AI Race

Around Christmas, both AI giants unveiled their latest models. OpenAI announced o3, the successor to its o1 reasoning model, coming in two versions: the full o3 and a smaller o3-mini variant. The model impressed with an 87.5% score on the ARC-AGI test, which checks if AI can learn new skills without training.

Google responded with its first reasoning model, Gemini 2.0 Flash Thinking Experimental, built on their Gemini 2.0 Flash base. Like o3, it's designed to "think" before responding. Neither model is widely available yet. OpenAI plans to release o3-mini in January, while Google's remains in experimental phase.

Gioele Mottarlini
Gioele MottarliniCOO and Image Addict

Companies have discovered that simply feeding models with more data isn't improving them anymore, that's why everyone is now turning to reasoning models.

The ARC-AGI test’s creator announced a new version that drops o3's performance below 30%. Now OpenAI is helping build this next test, which will likely bring biases.

Would you switch to a reasoning model like o3 or Gemini 2.0 Flash Thinking?

Login or Subscribe to participate in polls.

OpenAI's Vision vs Reality

Sam Altman started 2025 announcing OpenAI is now confident about building AGI and is already focusing on superintelligence. He predicts this year will see the first AI agents "joining the workforce" and transforming how companies operate.

But there's a catch, OpenAI is currently losing money on its $200 monthly ChatGPT Pro plan because users consume more resources than expected. With ChatGPT costing $700,000 daily to run, the company has to rethinking its pricing strategy and get more funds.

James Varnham
James VarnhamCEO and Rainmaker

The timing here tells us a lot. OpenAI is losing money because reasoning models and tools like Sora are extremely expensive to run, we're talking $700,000 per day just for ChatGPT. So when Altman claims they know how to build AGI, it's likely connected to their need for more funding.

When you get to define the finish line yourself, it's much easier to claim you're close to reaching it. The reality is that AI is far from replacing humans, those who will make a difference are the ones who understand how to work with AI.

Bits and Bobs 🗞️

  • China's DeepSeek has developed DeepSeek V3, one of the most powerful open AI models, excelling in tasks like coding and translating and outperforming notable competitors.

  • NVIDIA's new Project DIGITS, featuring the compact GB10 Superchip, debuts as the world's smallest AI supercomputer, capable of handling 200 billion parameter models with a petaflop of performance.

  • Google's DeepMind launched FACTS Grounding, a benchmark that measures how well AI models can stick to source documents without hallucinating.

  • Meta’s AI character accounts on its platforms, which faced backlash for being "creepy and unnecessary," leading the company to remove them.

From Our Channels 🤳

Check out this TikTok where Gianluca explains why we shouldn't get too excited about OpenAI's o3 scores on the ARC-AGI test, and why these benchmarks don't tell the full story about AI capabilities.

@gianluca.mauro

The AI graph that made everybody go nuts #Ai #learnontiktok #artificialintelligence #business #machinelearning #product #ux #entrepreneur

LOLgorithms 😂

OpenAI's journey from "non-profit" to "no profit”.

That's a wrap on our newsletter! Before you go, here’s a quick recap of our offerings:

Catch you next week! 👋