gpt-6VIDEO

OpenAI reportedly shifts 80-90% of research to GPT-7 and GPT-8

An OpenAI applied research lead is said to have described small model iterations as short-sighted, with the real plan being to distill small models like Luna from a stronger teacher model, Astra.

SuSu_酥酥👅

@NFT_ChenArchived 2026.09.28Originally posted 2026.09.27

SuSu_酥酥👅 @NFT_ChenOriginally posted 2026.09.27

Open original on X

OpenAI应用研究负责人Boris Power刚刚甩出重磅内部真相：公司80-90%的研究，已经在干GPT-7、GPT-8 以及更往后的东西。现在这些小版本迭代，内部直接被看成“极度短视”，只是为了今天学得快一点，根本不是长期正道。真正的打法更狠：先砸出超强大师模型Astra，再把它当老师，蒸馏出小而精的Luna。最好的小模型，从来不是从小模型训出来的，而是从最强的老师身上偷来的。 GPT-6 才刚上场，他们已经在下一盘更大的棋。 #OpenAI #GPT7 #GPT8 #Astra #Luna #AGI #SamAltman #Tibo https://x.com/nft_chen/status/2104208980302315617?s=46 https://x.com/firesidealpha/status/2103821387307651085/video/1

This post quotes another post: x.com/nft_chen/status/2104208980302315617

x.com/nft_chen/status/2104211526660079816Archived 2026.09.28 10:02