Gemini 3 and GPT 5.1

Summary

The video surveys recent major AI model releases—OpenAI’s GPT-5.1 (and GPT-5.1 Pro) and Google’s TPU-trained Gemini 3—placing them in the context of a fast-moving, competitive landscape that includes Chinese models (e.g., Kim K2, Ling, Miniax M2) and updates from Grok and Claude. It explores why these launches matter: hardware independence from NVIDIA, strategic timing, and the race for credibility. The presenter raises bigger questions about automating the manual steps of model training and whether massive, rapid retraining could accelerate progress toward AGI, while highlighting benchmarks like the Omniscience Index and urging viewers to follow hands-on reviews for deeper evaluation.

Takeaways

• 😀 OpenAI's GPT 5.1 was released on August 7, 2025, with an upgrade to 5.1 Pro on November 13, making the release cycle just 97 days.
• 😀 Google's Gemini 3 was released on November 18, 2025, marking a significant 238-day cycle since the Gemini 2.5 release in March.
• 😀 The AI industry has seen impressive releases from both Chinese models (e.g., Kim K2, Ling, Quen 3) and American counterparts like OpenAI, Grok, and Claude.
• 😀 Google’s Gemini 3 model was trained entirely on Google’s TPUs, not Nvidia chips, which could challenge Nvidia’s dominant position in AI hardware.
• 😀 Despite the rise of alternative hardware like Google’s TPUs, Nvidia remains financially strong, exceeding analysts' expectations in Q3 revenue and earnings.
• 😀 The competition between companies like Google, OpenAI, and Grok is heating up, with strategic model release timings, such as Grok 4.1 being released one day before Gemini 3.
• 😀 A key debate in AI is the distinction between real-world performance (e.g., coding tasks) and how the model 'feels' in terms of interaction and understanding.
• 😀 While LLMs (Large Language Models) alone may not achieve AGI (Artificial General Intelligence), automating tasks related to model training and alignment could bring us closer to AGI.
• 😀 With advancements in computing power (like gigawatt-level systems), models like GPT 5.1 and Gemini 3 could theoretically be trained in under 24 hours, accelerating the path to AGI.
• 😀 The Omniscience Index is a critical benchmark for AI model credibility, rewarding models for accuracy and the ability to avoid hallucination. Gemini 3 Pro and GPT 5.1 Pro scored highly, showing strong reliability.

Q & A

What is the significance of the release cycles of GPT-5.1 and Gemini 3?

The release cycles of GPT-5.1 and Gemini 3 are significant because they highlight the pace at which major AI models are evolving. GPT-5.1 was released 97 days after GPT-5, and Gemini 3 came 238 days after Gemini 2.5, showcasing the rapid development and competition in the AI industry.

Why is the release of Gemini 3 important for the AI industry?

Gemini 3 is important because it was trained using Google's own Tensor Processing Units (TPUs) instead of Nvidia's chips, showing that cutting-edge AI models can be built without relying on Nvidia. This could have long-term implications for the AI hardware market.

How does the performance of AI models like Gemini 3 and GPT-5.1 differ for different users?

For some users, the performance of AI models like Gemini 3 and GPT-5.1 is evaluated based on real-world tasks, such as coding, while others focus on how the models 'feel' in terms of intelligence and responsiveness. The difference in what matters most depends on individual preferences or use cases.

What is the role of manual tasks in training AI models, and why is automation a key discussion?

Currently, a lot of manual work goes into training AI models, such as gathering data and optimizing hyperparameters. The discussion of automating these tasks is important because it could drastically reduce the effort required for training, possibly accelerating the path to AGI (Artificial General Intelligence).

What does the Omniscience Index measure, and how does it compare Gemini 3, GPT-5.1, and other models?

The Omniscience Index measures a model's credibility by assessing how often it provides correct answers versus wrong ones, rewarding models that are hesitant to provide uncertain answers. Gemini 3 Pro scored highest at 13, while GPT-5.1 was second at 2, showing that credibility is becoming a critical factor in evaluating AI models.

What is the significance of Google’s TPU in the context of the Gemini 3 release?

Google's use of its own TPU (Tensor Processing Units) for training Gemini 3 is significant because it reduces the reliance on Nvidia's GPUs, which are traditionally the dominant hardware for AI model training. This could influence the hardware ecosystem in AI.

How has the AI industry's attention shifted in recent months?

A few weeks ago, there was a lot of focus on Chinese AI models, but now the spotlight has shifted back to American models, especially following the release of major models like GPT-5.1 and Gemini 3. This reflects the growing competition between U.S. and Chinese tech companies in the AI race.

What was the Deep Seek R1 event, and how does it relate to the current AI developments?

The Deep Seek R1 event, which occurred in January 2025, was a pivotal moment in the AI industry. It marked a major development, and since then, there has been a significant increase in competition and advancement in AI models, making events like the release of Gemini 3 and GPT-5.1 even more important.

What is the potential long-term impact of Google’s capital position and access to TPUs on AI development?

Google’s strong capital position and exclusive access to TPUs give it a significant advantage in the long-term AI race. This financial strength, combined with its hardware resources, positions Google to be a major player in the development of future AI models, possibly making it harder for competitors to keep pace.

Why is AGI (Artificial General Intelligence) considered a distant goal in the context of current AI models?

AGI is considered distant because current models like GPT-5.1 and Gemini 3, though impressive, are still limited in their scope and cannot demonstrate true general intelligence. Additionally, many manual processes still need to be handled during model training, and automating these tasks could be a key step toward AGI.