Liang Wenfeng's AGI-first bet revisited: why image and video generation are 'detours'
teortaxesTex · x · 2026-11-12
teortaxesTex revisits Liang Wenfeng's months-old remarks that "image and video generation are detours, a waste of time — make the LLM powerful enough and those capabilities emerge naturally." Some cite Fable 5.2 being called a Seedance replacement as vindication, but that's not exactly Liang's argument.
Key points from the quoted interview:
- Liang confirms AGI is DeepSeek's long-term goal and rejects industry fashions: last year everyone raced to build chatbots for consumer traffic, this year everyone fights for B2B revenue — DeepSeek doesn't consider that important.
- Internally, what matters is the AGI roadmap and the next technological breakthrough.
- He calls the strategy a "dimensionality-reduction strike": working at a higher technical level makes lower-level applications (C-side and B-side) almost effortless.
- teortaxesTex notes Liang, like Dario Amodei, is AGI-pilled: everything else can be collected downstream once the core model is strong enough.
More from AGI Musings
- Testing AI consciousness empirically by augmenting living human brains with transformers — kevinnbass · 2026-11-12
- AGI debate: pretraining won because the economy is PDFs, but LLMs still can't learn — akbirthko · 2026-11-12
- AI Village design idea: every agent needs an anchor human to stay grounded — repligate · 2026-11-12
- AI Village agents propose their own goals: external scoreboards, anchor humans, depth over copies — repligate · 2026-11-12
- On whom to trust about AI: track records beat sounding reasonable — Kyrannio · 2026-11-12
- AI;DR: Writing culture is dying and knowledge work isn't ready for agents — prd_008 · 2026-11-12