The classic case of engineers speculating about social science
There are very few papers at the moment demonstrating a productivity increase from AI at the firm level data. (But, to be fair, productivity is notoriously hard to measure, and it may be too early.) One I am aware of is
I guess most economists think that ML will give some productivity boost, especially in the long run, but it is unclear whether that will lead to an increase in growth rates (which would be great, but unlikely) or just levels (which is nice, but that is the standard pattern for single technological innovations).
The rest is just hype at this point.
I am a bit concerned about industries overextending themselves for ML though. Look at cameras: now manufacturers cannot release a body without sophisticated subject-recognition AF, which requires a dedicated ML chip, leading to extra cost, bulk, and power consumption. Anything else, however sophisticated in other respects, just gets a “meh” from reviewers. (Eg DPreview dismisses a lot of older Fuji and micro 4/3 bodies just because they don’t have Sony-like AF).
But it may mean that a large market of enthusiasts who have learned to use older AF techniques (or even manual focus) just stick to older bodies, which are smaller and do not require spending thousands of dollars. I don’t know the cost breakdown of camera bodies, but the ML hype may be an important factor pricing them out of the market for now.
I’m sure you are aware, but many aren’t, that productivity unless defined usually refers to economic productivity which is not the same as productivity understood as more produced per time unit. This is a language/field clash that I rarely see explained leading to a lot of misunderstanding.
~200 years ago farm productivity was hugely increased but the land was less productive. The increase in (economic) productivity was due to a reduction in farm workers. The reduction in what the land could produce was because early industrial farming was less efficient in getting crops out of the land.
BTW, Claude 3.7 in thinking mode has a ‘chain of thought’ with text like this (emphasis mine):
Actually, the problem statement is a bit ambiguous. […]
Actually, I realize I’ve been making a mistake here. The problem is asking for the distance between two points. To compute this, I need to know the coordinate of the apple and the coordinate of the peach.
Apple coordinate: 0 (leftmost edge of the desk)
Pear coordinate: 0.762 meters (that’s 30 inches from the left edge of the desk)
Peach coordinate: Let’s compute this carefully. The problem says the peach is “at a distance of 1 foot from the pear”. The problem doesn’t say the peach is “along the width” from the pear. It’s just at a distance of 1 foot from the pear. So the peach could be anywhere on a circle of radius 1 foot (0.3048 meters) centered at the pear. […]
But again, let me note that the problem doesn’t state that the peach is on the desk. […]
Wait, OK, the statement says “I place a peach, at a distance of 1 foot from the pear.” The problem doesn’t claim that I place the peach on the desk, nor that the peach is along the width from the pear. So I don’t think I’m supposed to conclude that the peach is to the left of the pear just because the peach wouldn’t fit on the desk to the right of the pear.
Hmm, actually, maybe I’m reading too much into this. […]
This is not enough information to compute the distance between the apple and the peach. We need to know the direction in which the peach is placed relative to the pear. So the problem seems to be missing critical information?
Actually, wait, one more time:
“On a desk that is 1 m wide, I place an apple right on the left edge. 30 inches from that, along the width, I place a pear. Finally, I place a peach, at a distance of 1 foot from the pear. What is the distance between the apple and the peach?”
OK the first two placements are unambiguous. The apple is the left edge, and the pear is 30 inches from the apple along the width. The width of the desk is clearly a 1D line, and the apple and pear are both placed along this 1D line. Given these two placements, the most natural interpretation is that the peach is also placed “along the width” of the desk, so it’s either 1 foot to the left of the pear, or 1 foot to the right of the pear.
[…]
Given how the problem is stated, the pear is at a distance of 30 inches from the apple. It also states the peach is at a distance of 1 foot from the pear. Based on this wording, the most natural placement of the peach is 1 foot to the right of the pear. So the distance from the apple to the peach is 1.0668 meters.
But the problem says the desk is 1 meter wide. So this would place the peach off the desk. If I take for granted that the peach is on the desk, then the peach is 1 foot to the left of the pear. So the distance from the apple to the peach is 0.4572 meters. I guess I don’t know which is the intended interpretation.
It is only after this that it answers (with 0.4572 meters, or 18 inches). It considered the ambiguity and went for the most natural interpretation.
Gemini 2 flash needed to be reminded about the 2D surface, but then it produced similar ‘thoughts’. Here, the emphasis is directly from Gemini:
Consider the context of “width”: The problem mentions the desk is “1 m wide.” The initial placement of the pear is explicitly stated as “along the width.” This establishes “width” as the relevant dimension for fruit placement in this problem. It’s natural to assume subsequent placements are also along this dimension unless stated otherwise.
Consider alternative interpretations and why they are less likely: What if the peach was not placed along the width?
Along the length (if defined): The problem only mentions desk width. Length is not explicitly defined. Even if there were a length, placing fruits along the length when the width is already being used as a reference dimension seems less likely in a simple word problem like this. It would make the problem more complex without need.
Diagonal or arbitrary direction: Placing the peach at a diagonal or in an arbitrary direction would also unnecessarily complicate a simple problem about distances on a desk. Word problems usually aim for the most straightforward interpretation unless explicitly stated otherwise.
Perpendicular to the width (along the depth, if any): Similar to length, depth isn’t defined. And moving in a direction perpendicular to the established “width” dimension seems less intuitive when discussing positions “along the width” for the pear.
Focus on the most reasonable interpretation: In the absence of explicit direction for the peach and given the prior context of placement “along the width” for the pear, the most reasonable and simplest assumption is that the peach is also placed along the width. This keeps the problem in one dimension and makes it solvable in a straightforward manner as intended for this type of problem.
