- One of the worst examples of this "psychosis", in terms of consequences, is the 1959-61 Great Chinese Famine which killed between 15 and 55 million people.
- We should heavily discount negative feelings related to criticism, instead of taking them at face value (as showing that something is wrong and should be fixed, e.g. by getting rid of the source of the criticism). I think this can often manifest not as "I hate this criticism" but more like "This person is so annoying and lack basic social skills."
- There's probably an effect where the less criticism we hear, the more sensitive we become to the remaining criticism, suggesting a slippery slope towards being surrounded by yes-men.
- Remember that most CEOs had to work their way up to that position, and have seen sycophancy from the bottom and understand that it's bad, but still fall prey to this problem.
Most organizational leaders have people they have to answer to and could be replaced if they perform badly. Why doesn't this fix the problem? Trying to answer this myself, I found this paper which says:
In a series of studies, Westphal and Stern (2006, 2007; Stern and Westphal, 2010) showed that ingratiation by a manager or director toward the chief executive officer (CEO) of a large company has a strong, positive effect on the likelihood of receiving the CEO's recommendation for a board appointment at another firm where the CEO serves as director.
I guess this is an instance of a more general phenomenon known as "board capture", which helps explain why having a board doesn't solve the whole problem. The paper also says:
Further analyses suggest that strategic persistence that results from high levels of flattery and opinion conformity directed at the CEO can result in the persistence of low firm performance and may ultimately increase the likelihood of the CEO's dismissal.
So I guess the CEO being fireable does solve the problem to some extent.
1
1
1...We should heavily discount negative feelings related to criticism, instead of taking them at face value (as showing that something is wrong and should be fixed, e.g. by getting rid of the source of the criticism). I think this can often manifest not as "I hate this criticism" but more like "This person is so annoying and lack basic social skills." There's probably an effect where the less criticism we hear, the more sensitive we become to the remaining criticism, suggesting a slippery slope towards being surrounded by yes-men. Remember that most CEOs had
2
So there’s this thing where someone talks to a large language model (LLM), and the LLM agrees with all of their ideas, tells them they’re brilliant, and generally gives positive feedback on everything they say. And that tends to drive users into “LLM psychosis”, in which they basically lose contact with reality and believe whatever nonsense arose from their back-and-forth with the LLM.
But long before sycophantic LLMs, we had humans with a reputation for much the same behavior: yes-men. The middle managers, advisors, and the like who surround CEOs and presidents and others at the top of large organizations. Given the similarity between yes-men and sycophantic LLMs, one might hypothesize a phenomenon analogous to LLM psychosis: yes-man psychosis.
The story: a CEO/president/etc, over the years, tends to retain the people who agree with them and say nice things, and tends to let go the people who disagree with them and say unkind things (just as many LLM users have a revealed preference for sycophantic models). Like a sycophantic LLM, the yes-men agree with everything the CEO/president says, tell them they’re brilliant, and generally give positive feedback on everything they say. And that tends to drive the CEO/president into “yes-man psychosis”, in which they basically lose contact with reality and believe whatever nonsense arose from their back-and-forth with the yes-men.
A central example: on my current models, yes-man psychosis was why Putin thought it was a good idea to invade Ukraine. Before the invasion, I remember reading e.g. RAND’s commentary (among others), which was basically “invading Ukraine would be peak idiocy, so presumably this is all sabre-rattling for diplomatic purposes”. Alas, in hindsight it seems Putin legitimately thought the invasion would be over in a few days, and Western powers wouldn’t do much about it. After all, the bureaucratic and advisory structures around him told him what he wanted to hear: that Russia’s military advantage would be utterly conclusive, Western leaders had no willingness to help, presumably nobody even mentioned the issues of endgame/exit strategy, etc.
The phenomenon of yes-man psychosis predates LLMs by millenia, so of course people have talked about it before, even if we didn’t have quite so crisp a name for it. I think I picked up the concept from John Boyd, the air force colonel who invented the term “OODA loop”. In Boyd’s frame, yes-man psychosis is a major way in which organizations-as-a-whole, or the people leading them, become de-facto unable to observe the world. Their perceptions are filtered through the yes-men, and the yes-men distort reality to the point where the leader can’t see it at all. (Think of a CEO surrounded by graphs and charts showing optimistic and coherent but basically false statistics about their company’s performance.) Without the ability to observe reality, the organization or its leaders cannot take different actions in response to different environment states. This, in Boyd’s frame, is a “dead” organization; like most dead matter and unlike living organisms, it just does its thing without particularly adjusting its responses to its environment in order to achieve particular outcomes.
Like rocks and logs, a dead organization can hang around for a long time if it’s not disturbed. But sooner or later the environment will change in such a way that the organization will collapse if it doesn’t do something different. And when that happens, the writing is on the wall for the organization.
I usually think of this as the default lifecycle of large organizations: they find a niche, grow for a while, then the leaders have the resources and time to select the people around them. And most leaders lack the awareness or discipline to avoid surrounding themselves with yes-men, so over time they lose the ability to accurately perceive their environment, and the organization “dies” in Boyd’s sense. Then the dead organization hangs around for some time - maybe a long time if it’s in a niche which doesn’t change much, or maybe a short time if it’s in a rapidly-changing environment. But it’s de-facto dead, just following its trajectory, and eventually it will collapse.