GPT- 4.5 Gossip Crushed But a 100T Transformer Model Coming? Plus ByteDance + the Mixtral Price Drop

1. GPT-4.5 rumors were denied by OpenAI employees.

🥈85 00:00

Multiple OpenAI employees have denied the rumors of GPT-4.5, calling it a hallucination and stating that there is no such model.

• The denials came from senior members of OpenAI, including Will Dew and Samman.
• The employees emphasized that people's expectations for GPT-4.5 are low and that there is no significant difference between GPT-4 and GPT-4.5.

2. A new company claims to have developed a 100T Transformer supercomputer.

🥇92 02:12

A new company has claimed to have designed a Transformer supercomputer, called Soo, that is optimized for running large language models like GPT-5.

• The company has burned the Transformer architecture onto a chip, which outperforms Nvidia's H100 in terms of tokens per second inference.
• The chip is expected to be available in 2024 and could potentially enable real-time interaction with large language models.

3. The Mixtral model matches or beats GPT-3.5 in benchmarks and is significantly cheaper.

🥈88 06:46

The Mixtral model, an 8.7 billion parameter model, has been open-sourced and has shown superior performance to GPT-3.5 in benchmarks.

• The pricing for the Mixtral model has dropped significantly, with providers offering it for free or at reduced prices.
• The combination of improved performance and lower cost is driving the development of more advanced language models.