‘Technology lock-in explains why the “best” system doesn’t always win.
The QWERTY keyboard became dominant early and crowded out alternatives like Dvorak, even though Dvorak was designed for more efficient typing. QWERTY was shaped by the constraints of mechanical typewriters, where key placement helped reduce jamming and manage typing speed.
That is the deeper lesson: once a technology becomes successful enough, entire industries build around it. Capital, infrastructure, talent, and incentives all reinforce the existing path, even if something better is possible.
I think this may apply to AI. LLMs running on GPUs are so commercially successful that trillions are now being deployed into scaling them. That success may create its own lock-in, crowding out more promising paths toward true AGI.
For me, real AGI would not just be better prediction of text. It would involve constructing models of reality and using inductive reasoning to predict genuinely new phenomena — the deeper engine of the scientific method.’
Thoughts from Prof B Keating on The Into The Impossible podcast
ABitcoinGuy / CWM
npub1qm…2sswt
2026-05-16 05:40:42 UTC
Author Public Key
npub1qm8ltd4njp8lpr0r8hdkgm3kfrz5zgsjwnkdw0kjwf2xj9awrlhsn2sswtPublished at
2026-05-16 05:40:42 UTCEvent JSON
{
"id": "b12778bd421d93b69422359509a6441bbdfe38f487b1691807107f066c7b99d3",
"pubkey": "06cff5b6b3904ff08de33ddb646e3648c541221274ecd73ed272546917ae1fef",
"created_at": 1778910042,
"kind": 1,
"tags": [],
"content": "‘Technology lock-in explains why the “best” system doesn’t always win.\n\nThe QWERTY keyboard became dominant early and crowded out alternatives like Dvorak, even though Dvorak was designed for more efficient typing. QWERTY was shaped by the constraints of mechanical typewriters, where key placement helped reduce jamming and manage typing speed.\n\nThat is the deeper lesson: once a technology becomes successful enough, entire industries build around it. Capital, infrastructure, talent, and incentives all reinforce the existing path, even if something better is possible.\n\nI think this may apply to AI. LLMs running on GPUs are so commercially successful that trillions are now being deployed into scaling them. That success may create its own lock-in, crowding out more promising paths toward true AGI.\nFor me, real AGI would not just be better prediction of text. It would involve constructing models of reality and using inductive reasoning to predict genuinely new phenomena — the deeper engine of the scientific method.’\n\nThoughts from Prof B Keating on The Into The Impossible podcast ",
"sig": "ba256a1531308dcec9acf8f4fee5f4b22dcf3da0e3e6c7240b97c0bd836299008613d45a00c3deeb1efe7d1ba77c3de26503a0bd76c43d04ff3d43b78ff69ea7"
}