There is another important connection between LLMs and management: the use of LLMs is itself a management task. The way I often describe them to people is that it's as if you have 100 green junior employees working for you all of a sudden. What would you do? How would you make use of their significant capabilities without wasting your own time. This is fundamentally a management question more than anything else.
When I read Henry's post this morning, my literal first thought was "I wonder if Dan has seen this yet, it seems right up his alley..."
I think I largely agree with his thoughts on the value of llm summarization as a managerial technology (and the caveats about their limitations). I think one potential risk is about incentives on both the data generation side and the assumptions about the quality/recency of the data being summarized. For instance, at my current job, there was a lot of executive enthusiasm about making it possible for sales/customer support execs to "chat with our knowledge base/documentation" in order to better respond to customer questions. Even setting aside hallucination risks, my first reaction as one of the scientists responsible for core product features was "our shipping timeline has been under such pressure for years that the literal last priority for engineering/DS has been to maintain the wiki/knowledge base - good luck to anyone treating that as a resource to answer customer questions." This is just classic garbage-in-garbage-out.
The second risk seems like the degree to which LLMs can remove friction in *generating* new "information." If it's easier to summarize large volumes of text, there could be greater perceived value/importance placed on producing large volumes of text. But if employees are producing these reports by just starting with five bullet points and then asking ChatGPT or equiv to expand them into a two (or three, or four?) page memo, (1) this is a hugely inefficient representation, as the only real information was in the five bullet points (unless we assume the LLM is going to cleverly synthesize some linkages between these points based on its compressed representation of a ton of other data), and in the game of expanding and then later summarizing, we've ultimately added a bunch of noise to the signal. Given the state of the internet these days, it seems like the arms race of "algorithmic slop generation" vs "algorithmic slop detection" generally favors the generator, not the discriminator, so it's not clear to me that we will end up with a clearer picture of the state of an organization when this technology proliferates without very clear standards/practices and incentive alignment on what generates the input data at lower levels of a complex system including lots of people with incentives to be lazy.
I’d add that my experience so far with AI transcription and meeting summaries is a bit worrying. Not hallucination per se, but much more than books it seems there are significant problems noticing “not” and other ways humans talk about a thing neutrally but with a negative objective. It’s easy to spot if you were there, but if not… and a big part of the point is to be useful when the memory of being there has faded.
Douglas Adams (1987): “Gordon’s great insight was to design a program which allowed you to specify in advance what decision you wished it to reach, and only then to give it all the facts. The program’s task, which it was able to accomplish with consummate ease, was simply to construct a plausible series of logical-sounding steps to connect the premises with the conclusion.
‘And I have to say that it worked brilliantly. Gordon was able to buy himself a Porsche almost immediately despite being completely broke and a hopeless driver. Even his bank manager was unable to find fault with his reasoning. Even when Gordon wrote it off three weeks later.’
‘Heavens. And did the program sell very well?’
‘No. We never sold a single copy.’
‘You astonish me. It sounds like a real winner to me.’
‘It was,’ said Richard hesitantly. ‘The entire project was bought up, lock, stock and barrel, by the Pentagon.”
Great essay—the twin points on iteration and choosing-the-result are especially good. I wrote about spreadsheets as a structural analogy for LLMs here:
When I read "Barbarians at the Gate" I did wonder how much of the LBO boom of the 1980s was driven by technological change -- while the "story" is largely about the people and personalities involved, most of the critical moments involve someone experimenting with or quickly revising a model using a spreadsheet program, something they wouldn't have been able to do even five years previously.
This is spot on. For now, I think of two potential LLM managerial killer apps: allowing for genuine searchable and summarised knowledge management and institutional memory, and speeding up and provide muscle for organisational change. I anticipate that management consultants, rather than management per se, will benefit most. This is rather significant industry which will see fairly dramatic productivity gains. However, I don't think, for now, that it will top the impact of Excel and spreadsheets, which has been truly transformative in the way Dan suggests.
Just as the Patent Office automatically tosses any application regarding perpetual motion, I walk out of any startup pitch involving an "Excel killer."
There are three main reasons why Excel is unkillable:
1) At this point, getting rid of Excel would be like changing the standard electrical current of a country. It's not that it's ubiquitous, although it is...it's that you can plug *anything* into it.
2) Everybody hates Excel. Nobody hates it *enough*, or hates it in the ways that actually lead to new product adoption. Sometimes both are true.
3) Any plausible replacement for Excel would have to support dozens, even hundreds, of critical use cases. Doing so would create...Excel.
Circa 1992, there was a famous (in certain circles) put-down of Microsoft's Word Business Unit Manager by its Excel Business Unit Manager that ran "Oh, a word processor? That's like a spreadsheet except it only has one cell."
That was a bon mot indeed, and yet here we are three decades later and we still aren't writing our documents on spreadsheets. As a finance guy, I like the focus on spreadsheets and I think that the general impact of word processors might be more obvious to the average person. But not, perhaps, the specifically cybernetic impact of word processors?
I would doubt it skyrocketed the economy in general (EBIT growth is the same as before), more likely shifted people out of management into other industries.
But this may have compensated for the demographic slowdown of urbanized countries either through getting people into increasing labor productivity or just plain increasing the available labor pool
There is another important connection between LLMs and management: the use of LLMs is itself a management task. The way I often describe them to people is that it's as if you have 100 green junior employees working for you all of a sudden. What would you do? How would you make use of their significant capabilities without wasting your own time. This is fundamentally a management question more than anything else.
When I read Henry's post this morning, my literal first thought was "I wonder if Dan has seen this yet, it seems right up his alley..."
I think I largely agree with his thoughts on the value of llm summarization as a managerial technology (and the caveats about their limitations). I think one potential risk is about incentives on both the data generation side and the assumptions about the quality/recency of the data being summarized. For instance, at my current job, there was a lot of executive enthusiasm about making it possible for sales/customer support execs to "chat with our knowledge base/documentation" in order to better respond to customer questions. Even setting aside hallucination risks, my first reaction as one of the scientists responsible for core product features was "our shipping timeline has been under such pressure for years that the literal last priority for engineering/DS has been to maintain the wiki/knowledge base - good luck to anyone treating that as a resource to answer customer questions." This is just classic garbage-in-garbage-out.
The second risk seems like the degree to which LLMs can remove friction in *generating* new "information." If it's easier to summarize large volumes of text, there could be greater perceived value/importance placed on producing large volumes of text. But if employees are producing these reports by just starting with five bullet points and then asking ChatGPT or equiv to expand them into a two (or three, or four?) page memo, (1) this is a hugely inefficient representation, as the only real information was in the five bullet points (unless we assume the LLM is going to cleverly synthesize some linkages between these points based on its compressed representation of a ton of other data), and in the game of expanding and then later summarizing, we've ultimately added a bunch of noise to the signal. Given the state of the internet these days, it seems like the arms race of "algorithmic slop generation" vs "algorithmic slop detection" generally favors the generator, not the discriminator, so it's not clear to me that we will end up with a clearer picture of the state of an organization when this technology proliferates without very clear standards/practices and incentive alignment on what generates the input data at lower levels of a complex system including lots of people with incentives to be lazy.
I’d add that my experience so far with AI transcription and meeting summaries is a bit worrying. Not hallucination per se, but much more than books it seems there are significant problems noticing “not” and other ways humans talk about a thing neutrally but with a negative objective. It’s easy to spot if you were there, but if not… and a big part of the point is to be useful when the memory of being there has faded.
Douglas Adams (1987): “Gordon’s great insight was to design a program which allowed you to specify in advance what decision you wished it to reach, and only then to give it all the facts. The program’s task, which it was able to accomplish with consummate ease, was simply to construct a plausible series of logical-sounding steps to connect the premises with the conclusion.
‘And I have to say that it worked brilliantly. Gordon was able to buy himself a Porsche almost immediately despite being completely broke and a hopeless driver. Even his bank manager was unable to find fault with his reasoning. Even when Gordon wrote it off three weeks later.’
‘Heavens. And did the program sell very well?’
‘No. We never sold a single copy.’
‘You astonish me. It sounds like a real winner to me.’
‘It was,’ said Richard hesitantly. ‘The entire project was bought up, lock, stock and barrel, by the Pentagon.”
Great essay—the twin points on iteration and choosing-the-result are especially good. I wrote about spreadsheets as a structural analogy for LLMs here:
* https://x.com/sampenrose/status/1850993128108601801
Frankfurtian bullshit is an interesting also-evil twin for "making the numbers say what the boss wants them to say."
Dan, loved meeting you at said conference and am trying to reach you! I'm at julia@workathon.io
Best
When I read "Barbarians at the Gate" I did wonder how much of the LBO boom of the 1980s was driven by technological change -- while the "story" is largely about the people and personalities involved, most of the critical moments involve someone experimenting with or quickly revising a model using a spreadsheet program, something they wouldn't have been able to do even five years previously.
This is spot on. For now, I think of two potential LLM managerial killer apps: allowing for genuine searchable and summarised knowledge management and institutional memory, and speeding up and provide muscle for organisational change. I anticipate that management consultants, rather than management per se, will benefit most. This is rather significant industry which will see fairly dramatic productivity gains. However, I don't think, for now, that it will top the impact of Excel and spreadsheets, which has been truly transformative in the way Dan suggests.
Just as the Patent Office automatically tosses any application regarding perpetual motion, I walk out of any startup pitch involving an "Excel killer."
There are three main reasons why Excel is unkillable:
1) At this point, getting rid of Excel would be like changing the standard electrical current of a country. It's not that it's ubiquitous, although it is...it's that you can plug *anything* into it.
2) Everybody hates Excel. Nobody hates it *enough*, or hates it in the ways that actually lead to new product adoption. Sometimes both are true.
3) Any plausible replacement for Excel would have to support dozens, even hundreds, of critical use cases. Doing so would create...Excel.
Circa 1992, there was a famous (in certain circles) put-down of Microsoft's Word Business Unit Manager by its Excel Business Unit Manager that ran "Oh, a word processor? That's like a spreadsheet except it only has one cell."
That was a bon mot indeed, and yet here we are three decades later and we still aren't writing our documents on spreadsheets. As a finance guy, I like the focus on spreadsheets and I think that the general impact of word processors might be more obvious to the average person. But not, perhaps, the specifically cybernetic impact of word processors?
Has anybody ever tried to estimate the contribution of spreadsheets to total productivity in the economy.....?
I would doubt it skyrocketed the economy in general (EBIT growth is the same as before), more likely shifted people out of management into other industries.
But this may have compensated for the demographic slowdown of urbanized countries either through getting people into increasing labor productivity or just plain increasing the available labor pool