Are breakthrough AI research ideas proprietary to individual labs or freely circulating among researchers?

Tom Blomfield · May 13, 2024 · hedged

Model innovations won't stay proprietary — talent mobility, helped by weaker non-compete enforcement in the US, means a discovery at one lab is replicated everywhere within weeks or months

Non-competes have become even less enforceable in the US, so talent jumps between labs quickly and carries innovations with it

Scope: unless raising a trillion dollars to train the biggest model really is the advantage

42:57 20VC: Behind the Scenes at Y Combinator: The Interview Process | What the Best & Worst Do in the Program | Do the Best All Raise Pre-Demo Day & YC's Fundraising Advice to Startups | Why the Value is in Application Layer AI with Tom Blomfield

Bret Taylor · Oct 2, 2024 · hedged

Breakthrough AI ideas are largely 'in the air' among researchers rather than proprietary to any one company, even though some knowledge is genuinely proprietary.

Many AI researchers come from an academic publishing ethos, and like the Wright brothers, multiple groups tend to converge on the same breakthrough at once.

Scope: some knowledge is proprietary; the publishing ethos has shifted somewhat in recent years; calls it a very complex question

29:16 20VC: Bret Taylor: The AI Bubble and What Happens Now | How the Cost of Chips and Models Will Change in AI | Will Companies Build Their Own Software | Why Pre-Training is for Morons | Leaderships Lessons from Mark Zuckerberg

Matt Clifford · Jul 1, 2024 · hedged

Secrecy norms inside the labs have tightened, so any idea with disproportionate impact on capabilities will be guarded and will produce short-term divergence between labs

Pretraining knowledge is now widely known, so the labs appreciate that the real edge lies in guarding high-leverage ideas

14:30 Secrecy norms have tightened so high leverage ideas are guarded short term

Noam Shazeer · Aug 31, 2023

The ability to experiment at small scale will produce far more published research even as large labs stop publishing to protect competitive advantage

Small-scale tinkering is widely accessible, so research output shifts to those who can still publish

21:03 Small scale experimentation democratizes published research even as large labs stop publishing