labor-economics on S Anand

When to choose AI over humans

Tue, 28 Oct 2025 04:37:14 +0000

I charted the OpenAI GDPVal paper with industry compensation as the size and AI augmentation as color. Big green areas are we’re paying people where AI does better.

Click here to see the interactive visualization.

Clicking to see some actual tasks compared.

I use this to check whom to ask advice: AI or professional.

AI beats Personal Financial Advisors ~64% of the time. So I invested half my money using ChatGPT’s recommendation. (UTI Nifty 50, if you’re curious.)

Accountants beat AI 76% of the time. So my tax returns are still filed by Ventura Pranas.

AI beats Government - Administrative Services Managers 62% of the time. I haven’t figured out how to bypass them yet. Nor customer service representatives.

Overall, AI beats most managers and clerks, not industrial engineers and pharmacists.

Here’s my current thought where I wouldn’t hire AI if a high portion of their work

Has legal liability (e.g. pharmacist/judge vs shop attendant/lawyer)
Is subjective (e.g. perfumer/auction appraiser vs lab chemist/insurance appraiser)
Needs rapid contextual decisions (e.g. detective/fireman/ER vs parking enforcer)

But this apart, if they charged half, would their demand double? If so, even with AI, they’ll have more demand.

Technology efficiency affects jobs differently

Sat, 14 Jun 2025 04:55:26 +0000

Jobs fall with technological efficiency.

Farmers in the US fell from 40% (1900) to ~2.7% (1980) and ~74% drop from 1948 to 2019 despite ~175% output growth; wheat harvest efficiency rose ~75* (300>3-4 man-hours).
Mechanics & repairers grew from ~140 k (1910) to ~4.64 M (2000); machinery reliability lagged so technician demand surged over decades.
Construction workers doubled from 1.66 M (1910) to 3.84 M (2000) even as labor share fell (4.3>3.0%); 5-10* productivity gains met booming development.
Switchboard operators plunged from ~1.34 M (1950) to ~40 k (1984) and ~4 k today as rotary-dial and digital switching automated call handling.
Travel agents dropped >50% from ~100 k (2000) to ~45 k (2022) while travel demand rose; online booking doubled trips per agent.
Elevator operators went from building-staff staple to near zero by the 1940s once automatic doors and button controls arrived.
Lamplighters vanished from thousands to near zero post-1907 electrification; Edison’s incandescent lamps eliminated manual lighting.

Jobs also grow with technology efficiency.

Software/IT workers surged from ~450 k (1970) to 4.6 M (2014); RAM price-performance jumped >100 000* (1 MB at $5 k>1 GB at <$0.03). 1980 IBM PC launch triggered a scramble for COBOL and BASIC coders-employers flew recruiters with cash-filled briefcases to MIT; “Y2K bounty hunters” later earned $1 k/hr inspecting two-digit dates.
Registered nurses climbed from ~12 k (1900) to ~3 M (2024); medical tech (antibiotics, MRI, EHRs) boosted care per nurse >10*. WWI field hospitals proved that trained nurses cut mortality by half; politicians returned home demanding hospital schools-enrollment tripled in one decade.
Wind-turbine technicians rose from ~4.7 k (2012) to 11.4 k (2023) and head for 18.2 k (2033); turbine capacity (660 kW>4+ MW) and remote SCADA expanded roles. First U.S. “windtech” apprentice class (Minnesota West, 2004) trained atop a decommissioned 90-ft tower welded to the parking lot; grads had 100 % placement.
Solar-PV installers jumped from ~4.7 k (2012) to 24.5 k (2023) as panel costs collapsed 80% and snap-in racking doubled installs per crew-day. An Arizona roofer who added PV installs in 2011 hit $1 B revenue by 2022-outselling five coal mines combined.
Social-media managers grew from ~2 k (2010) to ~61 k (2024); auto-schedulers let one person handle 50+ brand channels. Oreo’s “dunk in the dark” tweet (Super Bowl 2013) was crafted by a 15-person war-room-today a solo creator can replicate that reach on TikTok.

What drives head-counts growth despite efficiency jumps? Here’s ChatGPT’s guess:

Elastic new demand - Cheaper output unlocks previously-priced-out customers (software for every desk, electricity from prairie winds).
Complementary task creation - Tech automates routine sub-tasks, freeing humans for non-routine extensions: nurses moved from bed-making to ICU monitoring; developers shifted from punch-cards to UX design. ([nber.org][12])
Regulatory or safety mandates - Every extra MRI or turbine needs certified operators; compliance can outrun automation.
Network effects & ecosystems - Social platforms, app stores and open-source stacks spawn whole job families (community managers, DevRel).
Local installation & maintenance bias - PV panels assembled in Asia still need boots-on-roof locally.

So in the context of LLMs, here’s a guess on roles that could grow.

Edge-of-frontier complements. Prompt engineers, AI ethics leads, autonomous-fleet technicians emerge where tech leaves gaps.
Orchestrators & translators. Roles that fuse domain expertise with LLM tooling (e.g., “AI curriculum architect” in education).
Ecosystem enablers. Marketplace ops, trust & safety reviewers, fine-tuning data curators.
Regtech & audit. Assurance positions verifying model compliance and bias-often imposed by statute (EU AI Act clones).
Experience designers. As core tasks automate, differentiation shifts to narrative, community and emotion; expect growth in AI game writers, “synthetic brand” managers.

PS: The image was vibe-coded from BLS stats.

Sat, 24 May 2025 07:58:38 +0000

How much does an LLM charge per hour for its services?

If we multiple the Cost Per Output Token with Tokens Per Second, we can get the cost for what an LLM produces in Dollars Per Hour. (We’re ignoring the input cost, but it’s not the main driver of time.)

Over time, different models have been released at different billing rates.

New powerful models like O3 cost ~$7/hr – Poland’s minimum wage rate.
Gemini 2.5 Pro costs ~$12/hr – France’s minimum wage rate.
The latest Claude 4 Sonnet costs ~$2/hr – India’s minimum wage rate.

Once a language model’s run-time cost drops below the local minimum wage, the “offshoring” advantage disappears. AI becomes the cheapest employee in every country at once.

Paradoxically, workers in countries with strong labor protections, unions, and higher wages (like Germany and France) may be safer from AI displacement.

But countries whose economies depend on wage arbitrage face economic disruption.

Visualization: https://public.flourish.studio/visualisation/23384855/
Analysis: https://chatgpt.com/share/68317a06-0cac-800c-ad6f-13646ceb489f
Source code: https://github.com/sanand0/llmpricing/blob/9b66480458b712de52ea14502a64bf4c3b6a98ed/billing.py

Why your boss is overpaid

Sat, 27 May 2006 12:00:00 +0000

Why your boss is overpaid.

The ugly truth is that your boss is probably overpaid–and it’s for your benefit, not his. We work our socks off in underpaying jobs in the hope that one day we’ll win the rat race and become overpaid fat cats ourselves. Economists call this “tournament theory.”

The second, and more counterintuitive, prediction of tournament theory is that the more luck is involved in work, the larger the pay gaps should be between the winners and the losers. If Jack’s promotion is 90% luck and 10% effort, Jack may be inclined to goof off–unless, of course, the rewards for promotion are absolutely astronomical. And they sometimes are.

Finally, tournament theory also helps to explain why insiders, not outsiders, get cushy jobs. You thought it was all about the old-boy network, but in fact, the logical reason for promoting insiders is clear: These jobs are designed to keep your workforce motivated.

Comments

joey 6 Jun 2006 1:44 am: interesting analysis. no wonder the top execs in usa are paid enormously regardless of whether the company does well or not
Rghu 6 Jun 2006 2:55 am: A professor in a US school once told to me â€˜one of the reasons you have a job is most managers are unsure of anything. All they do is follow the ’latest trend’, be it Java or .NET or something fancy demanding more funds every few months and adding more peopleâ€™. Well I worked for a boss who didnâ€™t know the difference between a â€˜bitâ€™ and a â€˜byteâ€™ and supposedly led technical teams for many many years!

Beautiful people are paid more

Fri, 07 Apr 2006 12:00:00 +0000

Beautiful people are paid more because of a positive feedback mechanism. Employers expect them to perform better. They expect themselves to perform better. And these come across.

Comments

Madhu 10 Apr 2006 7:12 am: he he.. my research deals with whether board of directors have any impact on CG et al. I should probably consider the looks of the board as an important parameter
S Anand 12 Apr 2006 7:10 pm: It’s supposed to be a stronger effect with women. You could factor that in :-)
Madhu 14 Apr 2006 7:28 am: and probably suggestion reservation for women in the board;)

Labour statistics

Sat, 24 Feb 2001 12:00:00 +0000

The US Bureau of Labor Statistics has published a list of the fastest growing occupations from 1998-2008. The Indian Labour Bureau publishes some useful statistics too.