This author is taking the analogy to silos too literally, and constructs a weird argument around it that makes absolutely no sense.
The silo analogy is not about protecting academic integrity (in the way grain silos are meant to protect grain), it's about isolating teams from each other. If the physics department never spoke to the math department, that would be an example of what people mean by "silo" and a bad thing. It weakens both departments to be so severely isolated.
I have seen this in real academic disciplines, not just hypotheticals. CS academics have done a lot around modeling formal systems. Then you go over to systems engineers and they have done the same thing (especially around critical systems and safety properties of systems). Both have good ideas, but both domains operate largely in isolation from each other. This impedes their work, it's a pair of bad (though naturally occurring, rather than forced) silos.
If the author actually understood what people meant when they said silos need to be torn down, they wouldn't have written this bizarre blog post.
It's possible for the locking up of knowledge to be bad, and for the preservation of the integrity of knowledge to be good at the same time.
The mistake the article makes is of modeling the flow as bidirectional. And since she doesn't want integrity to suffer, she says siloing is good. You want information to get out without getting diluting. It is never good for the left hand not to know the right hand. Information should flow freely, subject to ingress monitoring; don't admit bad information, but don't close the border.
The question is whether you have the ability to police your side of the border.
There are too many blanket statements in this article that aren't well argued or even explained. This for example is just a series of assumptions:
"""
The attack on “silos” usually comes from people outside of a silo, generalists who don’t have deep disciplinary knowledge or focused training. These people don’t want their ideas validated by a community of experts. They find expertise to be inconvenient. The image of the silo as narrow, contained, a kind of ivory tower, seems to support the claim that those in them are narrow, out of touch, or secluded.
"""
The links in the first paragraph actually do quite a good job of explaining what people mean when they say Silos are bad. No one claims that disciplines and departments and teams should become a single blob. This article might be defending something that needs no defense and is not under any attack.
The conclusion to this essay doesn't make a convincing case. How do we know that academics are doing things correct when "sparks of AGI" is two years old and we don't have serious agentic software? How can you assert that isolated knowledge improves AI cognition when LLM performance directly correlates with the volume of training data? How could anyone actually prove that "academic silos" are challenging old paradigms when we aren't allowed to see what's inside them? What heuristics are you using? Why can't we see them?
This author is taking the analogy to silos too literally, and constructs a weird argument around it that makes absolutely no sense.
The silo analogy is not about protecting academic integrity (in the way grain silos are meant to protect grain), it's about isolating teams from each other. If the physics department never spoke to the math department, that would be an example of what people mean by "silo" and a bad thing. It weakens both departments to be so severely isolated.
I have seen this in real academic disciplines, not just hypotheticals. CS academics have done a lot around modeling formal systems. Then you go over to systems engineers and they have done the same thing (especially around critical systems and safety properties of systems). Both have good ideas, but both domains operate largely in isolation from each other. This impedes their work, it's a pair of bad (though naturally occurring, rather than forced) silos.
If the author actually understood what people meant when they said silos need to be torn down, they wouldn't have written this bizarre blog post.
Another example of silos is that Doctor who tried to claim the trapezoid rule and name it after herself.
https://news.ycombinator.com/item?id=26384357
I agree, she is conflating things.
It's possible for the locking up of knowledge to be bad, and for the preservation of the integrity of knowledge to be good at the same time.
The mistake the article makes is of modeling the flow as bidirectional. And since she doesn't want integrity to suffer, she says siloing is good. You want information to get out without getting diluting. It is never good for the left hand not to know the right hand. Information should flow freely, subject to ingress monitoring; don't admit bad information, but don't close the border.
The question is whether you have the ability to police your side of the border.
There are too many blanket statements in this article that aren't well argued or even explained. This for example is just a series of assumptions:
""" The attack on “silos” usually comes from people outside of a silo, generalists who don’t have deep disciplinary knowledge or focused training. These people don’t want their ideas validated by a community of experts. They find expertise to be inconvenient. The image of the silo as narrow, contained, a kind of ivory tower, seems to support the claim that those in them are narrow, out of touch, or secluded. """
The links in the first paragraph actually do quite a good job of explaining what people mean when they say Silos are bad. No one claims that disciplines and departments and teams should become a single blob. This article might be defending something that needs no defense and is not under any attack.
I did enjoy reading about the history of Silos :)
At my company siloing yourself is the only way to get anything done.
This is my experience.
The conclusion to this essay doesn't make a convincing case. How do we know that academics are doing things correct when "sparks of AGI" is two years old and we don't have serious agentic software? How can you assert that isolated knowledge improves AI cognition when LLM performance directly correlates with the volume of training data? How could anyone actually prove that "academic silos" are challenging old paradigms when we aren't allowed to see what's inside them? What heuristics are you using? Why can't we see them?