Model Makers
United States · BLS Employment Projections 2024–34 (figures for SOC 51-4061 occupational group)
AI exposure in published research
Figures below are reproduced from external datasets without modification. Where a dataset does not cover this occupation, the value is shown as — rather than as zero.
Data provider: OpenAI · "GPTs are GPTs"
Time basis: 2023 baselineex-ante estimate
Scored against GPT-4-generation capability.
- Human rater basis
- 9.7%
- GPT-4 rater basis
- 16.1%
β = direct exposure (E1) + 0.5 × exposure when tools are available (E2), per the source repository's definition.
16 rated tasks · 3 tasks with β ≥ 0.5 (18.8%)
- Version:
- gh-main-0471612
- License:
- MIT License · Copyright (c) 2024 OpenAI
Data provider: Anthropic Economic Index
Time basis: Published 2026-03-05composite index
- Observed exposure
- 0.0%
Theoretical exposure index weighted by measured Claude usage, per the original report's definition.
13 tasks · usage observed in 0 · mean 0.0%
Data: Anthropic Economic Index — labor_market_impacts, CC-BY, https://huggingface.co/datasets/Anthropic/EconomicIndex
AI Changing Work uses Anthropic Economic Index data under CC-BY; Anthropic does not endorse or sponsor this site or its analyses.
- Version:
- hf-lmi-2026-03
- License:
- CC BY 4.0
- Observation period:
- — (not applicable to this release)
- Model:
- — (not stated by the source)
- Definition source:
- https://www.anthropic.com/research/labor-market-impacts
2023 prediction vs observation-based index published 2026-03-05
One card (GPTs are GPTs) is a 2023 estimate of what AI could theoretically do; the other (Anthropic Economic Index) is built from observed usage and was published on 2026-03-05 — that is its publication date, not the period it observed. They measure different things, so the two figures cannot be added, averaged, or ranked against each other. The older figure is kept here as a baseline for comparison rather than removed.
Caution: the Anthropic figures — observed exposure and task penetration — take the Eloundou β as one of their inputs. The two sides resembling each other is therefore not evidence that the earlier prediction came true; reading it that way is circular reasoning.
The Anthropic figures are measured on Claude users, who are not the whole economy and not the whole workforce.
The mapping of O*NET tasks and occupation codes was performed by AI Changing Work. The source figures themselves were not modified.
These indices are not forecasts. Which point in time each one belongs to is stated on the badge on its card.
Task Breakdown
- Create prototypes and models
- Use CAD/CAM systems
- Read blueprints and specifications
About This Occupation
If you work as a Model Makers, AI is augmenting your role. Risk 26/100, exposure 36%.
ISCO-08 classification
Toolmakers and Related Workers
Definition
ILO original text (English)
Toolmakers and related workers make and repair custom-made and specialised tools, sports guns, locks, dies, patterns, machinery components and other metal articles using hand and machine tools to work metal to fine tolerances.
Definition & vocabulary source
Source: International Labour Organization (ILO) — ISCO-08 Structure
License: ILO CC BY 4.0
Official occupational information
- ONET49-9094.00
Locksmiths and Safe Repairers
Repair and open locks, make keys, change locks and safe combinations, and install and repair safes.
View original - ONET51-4061.00
Model Makers, Metal and Plastic
Set up and operate machines, such as lathes, milling and engraving machines, and jig borers to make working models of metal or plastic objects. Includes template makers.
View original
Frequently Asked Questions
The Anthropic Economic Index puts observed exposure at 0.0%. The OpenAI "GPTs are GPTs" rubric puts occupation-level β at 9.7% under human raters. Both figures are reproduced from published research without modification.
They come from two published datasets: the Anthropic Economic Index (labor_market_impacts, CC BY 4.0) and the OpenAI "GPTs are GPTs" exposure rubric (MIT License, Copyright (c) 2024 OpenAI). AI Changing Work maps them onto O*NET occupation and task codes and does not calculate exposure scores of its own. AI Changing Work uses Anthropic Economic Index data under CC-BY; Anthropic does not endorse or sponsor this site or its analyses.
No. They are a diagnosis of exposure as measured at the time each source dataset was published. AI Changing Work publishes no prediction of future automation or job displacement for this occupation.