The "given enough eyeballs, all bugs are shallow" line carries less weight once you ask who absorbs the loss when an eyeball misses. WR Berkley wrote an absolute AI exclusion into its liability policies, and AIG and Great American followed. That market did what no regulator managed: it made uncontrolled AI deployment uninsurable, and it drew no line between open weights and closed. Self-hosting an open model on your own infrastructure keeps your data in house, as Chesky notes, but it also means there is no vendor standing between you and that exclusion. Transparency answers the safety question. It does nothing for the indemnity one.
I wonder if the future ends up looking less like a choice between open and closed AI and more like a layered ecosystem.
The foundation may be open source, much as open source software already underpins much of the digital economy. But companies can still build proprietary workflows, specialized datasets, customer experiences, and domain expertise on top of that foundation.
In that sense, open source and commercial success may not be competing ideas. Open source may provide the platform while companies compete on how creatively and effectively they build upon it.
That possibility seems consistent with a great deal of technology history.
Given with what I have seen and heard about the way Fable/Mythos have been handled, the chaos in decision making and lack of deep technical expertise across key bureaucracies, anything is possible here...Dan Hendrycks has been calling for a ban on open weight models for sometime, and some are clearly listening now in a way different from last year...
Do you ever feel despair with how the government is acting?
Yes.
The "given enough eyeballs, all bugs are shallow" line carries less weight once you ask who absorbs the loss when an eyeball misses. WR Berkley wrote an absolute AI exclusion into its liability policies, and AIG and Great American followed. That market did what no regulator managed: it made uncontrolled AI deployment uninsurable, and it drew no line between open weights and closed. Self-hosting an open model on your own infrastructure keeps your data in house, as Chesky notes, but it also means there is no vendor standing between you and that exclusion. Transparency answers the safety question. It does nothing for the indemnity one.
I wonder if the future ends up looking less like a choice between open and closed AI and more like a layered ecosystem.
The foundation may be open source, much as open source software already underpins much of the digital economy. But companies can still build proprietary workflows, specialized datasets, customer experiences, and domain expertise on top of that foundation.
In that sense, open source and commercial success may not be competing ideas. Open source may provide the platform while companies compete on how creatively and effectively they build upon it.
That possibility seems consistent with a great deal of technology history.
Given with what I have seen and heard about the way Fable/Mythos have been handled, the chaos in decision making and lack of deep technical expertise across key bureaucracies, anything is possible here...Dan Hendrycks has been calling for a ban on open weight models for sometime, and some are clearly listening now in a way different from last year...