GPT-6 Not Yet Released, OpenAI May Have Already Trained Next-Generation 10 Trillion Parameter Base

GPT0%
AGI0%

Dynamic Beating AI News Flash: OpenAI has reportedly just completed a new large-scale pretraining, internally codenamed Bel, with a total of over 100 trillion parameters. Leaks suggest that OpenAI has positioned it beyond GPT-6 and may even serve as the foundation for a future "AGI threshold model."

Previously, Doug was considered the pretraining base for Astra and GPT-6; Bel is the next generation of Doug.

The same leak also suggests that Anthropic may struggle to produce a model on par with Astra this year, mainly due to computational power constraints.

Sumber: BlockBeats

Disclaimer: Konten ini berasal dari pihak lain atau diterjemahkan oleh AI dari pihak lain. CoinEx tidak menjamin konten ini benar, asli, atau akurat, dan tidak memberikan saran investasi. Harga aset kripto sangat tidak stabil, jadi harap berhati-hati terhadap risiko yang ada.

Kabar TerkaitSelengkapnya

OpenAI has claimed that its first self-developed inference chip, Jalapeño, has surpassed Nvidia's GB300: AI throughput per watt is 1.5 to 1.9 times higher, with latency reduced by up to 3.6 times.

Analysis: OpenAI's "Jalapeño" AI Chip Expansion May Be Affected by Samsung's HBM4 Supply Constraint

OpenAI's 「Spicy Chip」 Takes On NVIDIA: AI Inference Performance Highest, 3.6x Faster, Small-Scale Deployment by Year End

Pencarian Teratas

Kripto
Harga
Perubahan 24J