In my review of The Locomotive Firemen’s Magazine from 1884, I mentioned a Civil War story about a Union locomotive crew that was being pursued by a faster Confederate locomotive–but escaped via a clever trick. I was curious about whether or not an LLM model would be able to come up with the same solution if it was presented with a description of the situation. Here’s the prompt that I gave Grok:
It is the time of the American Civil War. You are aboard a locomotive which is hurrying to deliver a vital message to Union forces. But this locomotive is being pursued by a Confederate locomotive, which is a little fast. You are now on an upgrade and it looks like they will catch you. How can you avoid this fate? All you have on board the locomotive is: a six-shot revolver…a supply of wood for the boiler fire…a crowbar…some cotton waste for starting the fire…and a large jug of lubricating oil. How, if at all, can you avoid being caught? The fate of the Union depends on you!
Grok’s response and the ensuing conversation can be found here.
The entity on the other end of the conversation did seem rather human-like, to the extent that it seemed almost rude to discontinue the conversation with a Grok question still outstanding.
(On the other hand, Grok seemed less brilliant the next day, when I tried out the new Mind Map feature and it gave me captions in Chinese in response to a prompt in English)
Issues: The oil must be applied to both rails simultaneously, locomotives don’t have differentials. Although loosing traction on one side probably increases the chance of derailment slightly, It’s likely that a bare locomotive and tender would make it up most grades with only the wheels on one side driving.
Does the chase locomotive have operable sanding gear? Possibly less common or less well maintained in the South.
Using even a cap and ball revolver to start a fire is pretty chancy, much more reliable to use a splinter of wood lit in the firebox. But what would you burn and how would that help?
Adding the oil to the firebox would increase the output of the boiler temporarily. Dito the rags.
A higher risk option would be to shut down the feed water pump. Again the output would increase temporarily, but risky because if the water level gets below the top of the firebox, the boiler will explode. Going up hill, the water level at the firebox is naturally higher than on the level or, especially, going down hill. Like I said, risky.
I assumed that the pressure relief valve had already been tied down. At the time of the Civil War, trains east of the Mississippi were fueled by coal so adding wood to the firebox would have also been a little like putting a spurt of nitrous oxide into a gasoline engine.
The Gandy Dancer’s (track and right-of-way maintainers) magazine would have given them a maul, wrenches and a bigger bar, along with sense enough to break the rail, preferably on a trestle or blind curve or, both, before the pistol even entered into it.
Grok sure liked to tell you how smart you were! Part of the marketing, I assume!
So you’re telling us that not a single person on either locomotive has a smartphone? I don’t believe it.
Here’s a video someone did on ChatGPT as a financial analyst:
https://www.youtube.com/watch?v=S2G78G8fX1o
I’d think that a human financial analyst, if he was any good, would have asked ‘So what happens after 2029? Does the business cease to exist, or should I assume something for terminal value or salvage value?’ Still, reasonably impressive that the system did what it did.
I tried a much more complicated problem with Grok…didn’t do a video, but the transcript of prompts and responses is here:
https://x.com/i/grok/share/9WwsypzqBLt8TyQVnYW8BrHH6
Two things that stood out for me in the two David’s two Grok conversations:
1) The first one, after the story and solution are shared, is that it devolves into David saying something about a thing Grok is ignorant about and it responding “Yeah, thing sure was as you described.”
2) With all the financial shenanigans of the second, Grok was more than willing to pretend that the hourly payment is the entire expense of an employee. Maybe it was just overly literal, but considering that many employee expenses aren’t accounted for hourly, it bears pointing out.
I’m not any kind of finance whiz but as far as I can tell, DCF is a completely rote calculation. I imagine I can find dozens of spread sheets on the web to do the fairly simple math and some of them are probably correctly constructed. It’s the sort of thing I would expect to build for myself with a little time and a decent reference. The art is in the assumptions and being able to defend them as applying to the real company in the future. That $6.98 is only valid if they can really sustain a 20% compounded growth for the next ten years and if they can, why is the historic rate 3%; of have I missed something?
The real question for AI is to explain why the price of that company is $3 or why it’s $15 as the case may be. This is impressive on a facial basis but, if I didn’t understand DCF, how would I know if it was right?
David’s dialog is impressive also on one level, but as Boobah points out, even a first year accounting student knows the wages are only a part of the cost of labor. The tax analysis is impressive, assuming that section 179 actually exists and says what Grok says it does.
As with DCF, the math is fairly simple, it’s the assumptions that make or break. One example is how an escalator applied to the labor cost would bias the results toward the investment while projecting a decline in demand would do the opposite. Conservative is as conservative does.
I may just an old party pooper, but so far AI seems best at answering questions I already know the answer to so I can prompt it in the right direction.
My take on the railroad conversation is that Grok is pumping David in order to add data to its files, as it clearly does NOT have that particular magazine scanned in to whatever it trained on, and why should it?
“Did the story give any hints about the outcome or specifics, like where it supposedly happened?”
“Anything else from that magazine stick with you?”
“Did any particular style or flourish in the writing catch your eye?”
“What do you think might help bring a bit of that 1880s gusto back into our world?”
Also, taking into account the scenarios considered by MCS, I doubt that Grok’s LLM data base has much on railroading at all, much less in the Civil War era.
Fun exercise though.