GPT-5.2 (openai.com)

Article summary

The article discusses the release of GPT-5.2, with some commenters expressing disappointment and skepticism about the incremental improvements. The conversation reveals that OpenAI has not had a successful pre-training run in 1.5 years, and the new model may be using continued pre-training on older weights. The discussion also touches on the challenges of training new models, including cost and technical hurdles.

Main themes

• GPT-5.2 release
• AI model training challenges
• OpenAI's technical hurdles
• Cost and resource constraints
• Model scaling and improvement

What commenters say

• The release of GPT-5.2 is underwhelming due to marginal gains and high costs.
• OpenAI's inability to complete a successful pre-training run in 1.5 years is a significant technical hurdle.
• The company's lead in the AI field is being fumbled, and exciting developments are happening elsewhere.
• The improvements in AI models since GPT-4 are largely due to reasoning and reinforcement learning, rather than scaling.
• The next significant advancements in AI will require substantial increases in datacenter capacity and scaling.
• The current pace of progress in AI is slowing down, with some commenters feeling the effects of the S-curve.
• Others disagree, arguing that the progress is still exponential, albeit with a long doubling time.
• The challenges in training new AI models are not just technical, but also related to cost, resource constraints, and the pressure to monetize products.