Yesterday likely marked the launch of GPT-5 by OpenAI. This update is based on extensive research, including public comments from OpenAI, an interview with a hardware CEO, and my analysis. Additionally, I’ll share a practical GPT tip and an interesting discovery about Darly.
The first hint of GPT-5’s training came from Greg Brockman, OpenAI’s co-founder. OpenAI often trains smaller models before a full-scale run. Brockman’s tweet suggested a shift towards maximizing computing resources for their largest model yet.
Another clue came from Jason Way, a prominent OpenAI researcher, who tweeted about the thrill of launching a massive GPU training run. This doesn’t mean GPT-5‘s immediate release; GPT-4’s training and safety testing took about three months.
Further evidence of GPT-5’s training includes OpenAI’s blog update about closing applications for their red teaming network, indicating readiness for new model safety testing.
GPT-4.2
OpenAI has previously released smaller, incremental updates (e.g., GPT-4.2) before major versions. Sam Altman hinted at GPT-5’s capabilities from earlier, smaller model versions, expressing excitement about pushing the boundaries of discovery.
GPT-5 aims to improve on reasoning steps and reliability, allowing it to lay out and internally check its reasoning before solving challenges. This was discussed by Sam Altman, emphasizing the model’s ability to articulate its reasoning process in natural language.
GPT-5’s parameter count might be 10 times that of GPT-4, based on an interview with Gavin Uberti of Etched AI. This suggests a significant increase in complexity, potentially enhancing the model’s reasoning and problem-solving abilities.
I also explored a quirky aspect of Dar 3 and provided a practical tip for using GPT: don’t worry about correcting minor typos, as the model can handle them.
Finally, I predict GPT-5‘s release towards the end of November 2024, considering training time, safety testing duration, and strategic timing post-U.S. election to avoid controversy.
This brief aims to inform about GPT-5’s development, potential capabilities, and anticipated impact, based on the most reliable sources and analysis.
Read related articles: