Working with AI: Measuring the Applicability of Generative AI to Occupations

Analyzing 200k real Bing Copilot chats, researchers found AI is used most for information work, ranking occupations by AI applicability.

2025/07/10 by Kiran Tomlinson, Sonia Jaffe, Tomlinson, Kiran +8 · 110 voices · 10 citations
Social Sciences · Computer Science · Medicine · #Ethics and Social Impacts of AI #AI in Service Interactions #Artificial Intelligence in Healthcare and Education

paper · pdf · doi:10.48550/arxiv.2507.07935

Abstract

With generative AI emerging as a general-purpose technology, understanding its economic effects is among society's most pressing questions. Existing studies of AI impact have largely relied on predictions of AI capabilities or focused narrowly on individual firms. Drawing instead on real-world AI usage, we analyze a dataset of 200k anonymized conversations with Microsoft Bing Copilot to measure AI applicability to occupations. We use an LLM-based pipeline to classify the O*NET work activities assisted or performed by AI in each conversation. We find that the most common and successful AI-assisted work activities involve information work--the creation, processing, and communication of information. At the occupation level, we find widespread AI applicability cutting across sectors, as most occupations have information work components. Our methodology also allows us to predict which occupations are more likely to delegate tasks to AI and which are more likely to use AI to assist existing workflows.

Summary

The authors analyzed 200,000 anonymized Microsoft Bing Copilot conversations, using an LLM pipeline to classify what the user wanted and what the AI actually did in O*NET work-activity terms. Most usage and success clusters around information work such as writing, explaining, and answering questions, and this translates into an occupation-level 'AI applicability score' that ranks media, sales, and clerical jobs highest and manual, physical jobs lowest.

machine-generated · claude-sonnet-5

Outline

• Introduction — Frames the goal: use real Copilot usage, not expert forecasts or benchmarks, to see which work activities and occupations AI actually touches.
• Results: Work activities — Classifies each conversation's user goal and AI action into O*NET work activities; finds AI usage concentrated in learning, communicating, teaching, and writing.
• Results: Metrics of success — Scores each activity on completion rate and scope (how much of the activity Copilot can handle); AI does best on communication/writing, worst on image generation and data analysis.
• Results: AI applicability to occupations — Aggregates activity-level scores, weighted by O*NET task importance, into a per-occupation AI applicability score; ranks all SOC occupations and correlates the ranking against prior AI-exposure predictions and wages.
• Results: User goals vs. AI actions — Uses PCA to show some occupations lean toward delegating tasks to AI (high AI-action score) while others lean toward using AI to assist ongoing work (high user-goal score).
• Discussion — Argues AI mainly affects information work, warns against assuming automation-scored jobs will simply be replaced, and lists limitations and future research directions.
• Materials and methods — Details the Copilot dataset, O*NET/BLS/CPS data sources, the two-stage GPT-4o classification pipeline, the applicability-score formula, and human validation of the pipeline.

machine-generated · claude-sonnet-5

Claims

• The most common and most successful AI-assisted work activities in real Copilot usage are information work (creating, processing, and communicating information), not physical or manual tasks. [experiment]
• In 40% of conversations the user's stated goal and the AI's actual action map to entirely non-overlapping O*NET work activities, and on average the AI performs about twice as many distinct work activities as the user's single stated goal. [experiment]
• An occupation-level AI applicability score built from this usage data correlates strongly (r=0.73, employment-weighted) with Eloundou et al.'s independent expert-based predictions of which occupations would see AI-driven time savings. [experiment]
• AI applicability score is only weakly correlated with wage (employment-weighted r=0.13) and spans a broad range across education levels, in contrast to prior work reporting strong positive wage correlations for AI exposure. [experiment]
• Occupations split into two patterns: some (e.g., business and financial operations) score higher on AI-action applicability, suggesting workers there may delegate tasks to AI, while others (e.g., computer and mathematical occupations) score higher on user-goal applicability, suggesting workers there use AI to assist their own ongoing work. [experiment]
• High AI-action applicability should not be read as a prediction of job or wage loss, nor high user-goal applicability as a prediction of wage gains, because historical technology adoption (e.g., ATMs and bank tellers) shows employment effects of automating a task are hard to predict and can run counter to naive expectations. [argument]

machine-generated · claude-sonnet-5

Key figure

Figure 1 — Four charts showing which work activities came up most in Bing Copilot chats: the 15 most common things users wanted help with, the 15 most common things the AI actually did, how these compare to how often those activities happen across the whole U.S. workforce, and which activities the AI handled most and least successfully.

machine-generated · claude-sonnet-5

Glossary

O*NET

A U.S. Department of Labor database that breaks occupations down into their component tasks and work activities.

Intermediate Work Activity (IWA)

A mid-level, cross-occupation description of a job task from O*NET, e.g. 'Edit written materials or documents.'

Generalized Work Activity (GWA)

A broader O*NET grouping of related IWAs (37 total) used to compare Copilot usage against overall workforce activity.

User goal

What the human in a Copilot conversation was trying to accomplish, as classified by the paper's LLM pipeline.

AI action

What the AI itself did in the conversation, classified separately from the user's stated goal.

Scope (of impact)

A six-point rating (none to complete) of how much of a work activity's total work Copilot demonstrated it could handle.

Completion rate

The share of conversations in which an LLM classifier judged that Copilot successfully completed the user's request.

AI applicability score

The paper's composite per-occupation metric combining activity coverage, completion rate, and scope, weighted by how important each activity is to that occupation.

Standard Occupational Classification (SOC)

The U.S. government's standard system for coding and grouping detailed occupations, used here to report results by job and job group.

Activity share

The fraction of all conversations attributed to a given work activity, splitting credit evenly when a conversation matches multiple activities.

machine-generated · claude-sonnet-5

Audience

Labor economists, workforce and AI policy researchers, and product or policy teams at AI companies who need evidence-based (not forecast-based) estimates of which jobs and tasks generative AI already affects.

prerequisites: Basic familiarity with occupational task taxonomies like O*NET or SOC, Comfort reading correlation coefficients and PCA-based scatterplots, General understanding of how an LLM classification pipeline works

machine-generated · claude-sonnet-5

Open questions

• How will occupations restructure their mix of tasks and responsibilities as AI capabilities keep improving?
  The paper measures where AI overlaps with current work activities today but states this is an open question about how jobs will actually change in response.
• What new occupations will emerge to perform new kinds of work activities that AI makes possible?
  The authors note most of today's jobs arose from past technologies within the last century and expect the same could happen with generative AI, but the paper cannot say what those jobs will be.
• Does an AI successfully completing a work activity in a conversation translate into real productivity gains for the person doing that job?
  The paper explicitly states its completion and scope metrics measure AI's utility on an activity, not its actual productivity impact on workers.
• How would results differ for other AI platforms, including workplace-specific or occupation-specific AI tools not captured in consumer Bing Copilot data?
  The authors flag their data as only one slice of the AI market, potentially missing usage patterns from task- or occupation-specific LLM products.

machine-generated · claude-sonnet-5

Supplementary links

• https://github.com/microsoft/working-with-ai [code]

machine-generated · claude-sonnet-5

Citations

• Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce
• Towards Accurate Differential Diagnosis with Large Language Models
• How will Language Modelers like ChatGPT Affect Occupations and Industries?

Cited by

• The atomic structure of work: a micro-action instrument reveals two-pole AI occupational exposure and its decade-scale polar inversion
• AI Fiction in the Wild
• LLMs Corrupt Your Documents When You Delegate
• Generative AI Practices, Literacy, and Divides: An Empirical Analysis in the Italian Context
• Bowling with ChatGPT: On the Evolving User Interactions with Conversational AI Systems
• Readers Prefer Outputs of AI Trained on Copyrighted Books over Expert Human Writers
• A theory-based AI automation exposure index: Applying Moravec's Paradox to the US labor market
• Flipping the Dialogue: Training and Evaluating User Language Models
• DuetUI: A Bidirectional Context Loop for Human-Agent Co-Generation of Task-Oriented Interfaces
• Workplace impact of artificial intelligence [wikipedia]

Discussions

• This incredibly worrying study suggests Historians are the 2nd most likely occupation to be supplanted by AI. That’s despite the fact that current AI history output is wildly unreliable, unsourced, ha [bsky, 302 points, 34 comments]
• Also, just because the makers of a technology promise that it is the second coming of Christ, we do not have to believe them. We don’t have to use it. These people/companies seek to gut our field, and [bsky, 223 points, 0 comments]
• Talk to your kids! Here’s the link to the Microsoft report listing jobs likely to be taken over by AI. arxiv.org/pdf/2507.07935 [bsky, 122 points, 10 comments]
• i assume the finance newsletter dot com assumed, correctly, that most of bluesky wouldn't bother to try reading it, but it's not long and not hard to understand [bsky, 84 points, 2 comments]
• Yup. The study: arxiv.org/abs/2507.07935 What Andrew claims are the conclusions of this study are not at all the conclusions drawn from this study. [bsky, 61 points, 6 comments]
• I've seen variations on "top 40 jobs replaced by AI" etc. FWIW, that's not what the paper said. It's not a great paper, but it's appropriately limited to whether AI uses (specifically Copilot) general [bsky, 54 points, 4 comments]
• Here's the original paper, looks like it's not quite as was presented: arxiv.org/abs/2507.07935 [bsky, 27 points, 2 comments]
• Microsoft published research showing top 40 jobs "with highest AI applicability score" (translation: we're boned) arxiv.org/pdf/2507.07935 [bsky, 25 points, 13 comments]
• I read the paper that is the basis for this headline/take and it truly does not say this, it is an attempt to create an "AI applicability" score for various tasks, which ofc is highly limited in terms [bsky, 24 points, 3 comments]
• This chart is from a preprint of a paper some Microsoft people did. I find it hard to believe that "Writers and Authors" and Historians are more likely to be replaced by AI than Telemarketers. What a [bsky, 18 points, 6 comments]
• New Microsoft study reveals the 40 most-likely professions to be impacted by AI growth and 40 least-likely. White-collar work is going to be radically transformed. Please share this information with t [bsky, 15 points, 9 comments]
• The actual paper is here: arxiv.org/pdf/2507.07935 One major flaw, IMO, is the evaluation of whether AI can do the job. User asks Copilot a question, Copilot responds, user gives thumbs up/thumbs down [bsky, 15 points, 3 comments]
• Sooooo this paper is annoyingly written, and they didn’t put their findings in the abstract (cowards :-P ), but there’s some interesting stuff in it. arxiv.org/pdf/2507.07935 My main takeaway is that [bsky, 13 points, 1 comments]
• You can find the research paper below [bsky, 13 points, 0 comments]
• Read it and weep / swear / etc. Tables of the bottom and top 40 occupations with "highest AI applicability score" from a pre-print Microsoft paper. arxiv.org/pdf/2507.07935 [bsky, 12 points, 4 comments]
• microsoft paper essentially says that AI is best used for management and sales tasks not creative or specialist ones arxiv.org/pdf/2507.07935 [bsky, 11 points, 0 comments]
• An interesting but, I think, partially misguided paper looking at occupations’ vulnerability to AI. It has ‘archivist’ in its top 40 most vulnerable. But this is because it has determined that archivi [bsky, 8 points, 1 comments]
• here's the actual thing arxiv.org/abs/2507.07935 [bsky, 8 points, 3 comments]
• Working with AI: Measuring the Occupational Implications of Generative AI - Microsoft Research arxiv.org/pdf/2507.07935 (PDF) - 마이크로소프트 코파일럿 실제 대화 20만 건을 분석한 결과, 영업·컴퓨터·사무직의 AI 적용 가능성이 가장 높았으며, 반면 육체노 [bsky, 7 points, 1 comments]
• found it; study's from june last year but was updated in december. the job categories are generated by an llm by cross-referencing bing ai conversations with a job database, it seems. but the actual " [bsky, 7 points, 3 comments]
• Only someone who's never read a history book or met a historian could possibly believe AI can replace 91% of historians' labor. This isn't research, it's an anti-labor manifesto crossed with a technoc [bsky, 6 points, 1 comments]
• Working with AI: Measuring the Occupational Implications of Generative AI [hn, 6 points, 1 comments]
• Yes, it did. "we use an LLM-based pipeline to map the user goals and AI actions in each conversation to the set of work activities being performed." arxiv.org/pdf/2507.07935 Based on the idea that see [bsky, 6 points, 2 comments]
• Does anyone find it ironic that the analysis of chats reported in Microsoft's paper "Working with AI: Measuring the Occupational Implications of Generative AI" (arxiv.org/pdf/2507.07935) was performed [bsky, 5 points, 0 comments]
• Research paper is here: arxiv.org/pdf/2507.07935. I do not believe this is a credible method to answer the question asked [bsky, 4 points, 1 comments]
• I need folks to know that this list is basically a list of jobs created to match the most common things that people used Bing Copilot to help them do. Paper here: arxiv.org/pdf/2507.07935 [bsky, 4 points, 2 comments]
• Here's the paper arxiv.org/abs/2507.07935 [bsky, 4 points, 2 comments]
• listeye bak. yapay zeka geliştikçe yok olma riski taşıyan 2. meslek tarihçilik mi? ne iş yaptığımızı bilmediklerinden, pek naifler. hem tarihçiyi yok ederse yapay zekanın tarihini kim yapacak? beni yo [bsky, 4 points, 0 comments]
• I understand the anxiety around this, but the paper is claiming AI's usefulness as an assistant tool for those professions, not its competency to "replace" people. You can see it in the screenshot's f [bsky, 3 points, 1 comments]
• Which #jobs will or will not be replaced by #AI. Top 40 arxiv.org/abs/2507.07935 [bsky, 3 points, 0 comments]
• Measuring the Occupational Implications of Generative AI [hn, 3 points, 1 comments]
• #Microsoft released a detailed report on specific tasks AND jobs most at risk to AI — a useful guide to what might lie ahead for your job or specific tasks. See our list of the 10 most-threatened (int [bsky, 3 points, 0 comments]
• People should read the actual paper, rather than the reporting summarizing it. arxiv.org/abs/2507.07935 [bsky, 2 points, 1 comments]
• The pre-print (i.e., not peer reviewed) is here: arxiv.org/abs/2507.07935 [bsky, 2 points, 0 comments]
• Here is the paper if anyone wants to actually read it arxiv.org/abs/2507.07935 [bsky, 2 points, 0 comments]
• If you've read the Microsoft paper that implied your job will be taken away, know that the paper has a lot of logical pitfalls. A logical equivalent of what this paper says: Because more people google [bsky, 2 points, 0 comments]
• Uh oh #polisky Microsoft thinks we're more AI-replaceable than basically everyone arxiv.org/abs/2507.07935 [bsky, 2 points, 2 comments]
• Working with AI: Measuring the Occupational Implications of Generative AI arxiv.org/abs/2507.07935 [bsky, 2 points, 0 comments]
• Working with AI: Measuring the Occupational Implications of Generative AI [hn, 2 points, 0 comments]
• Also nächster Job Einbalsamierer. Den kann die KI nicht so leicht klauen und gestorben wird immer. arxiv.org/abs/2507.07935 [bsky, 2 points, 0 comments]
• You will hear a LOT about this paper in the coming weeks: Here's actual #Microsoft Research, Top 40 jobs most susceptible to #AI automation: arxiv.org/pdf/2507.07935 Two, things to know: 1. It used 'C [bsky, 2 points, 0 comments]
• Has anyone examined the Microsoft paper that lists historians as second from the top in careers under pressure from AI? arxiv.org/pdf/2507.07935 They use a database of job requirements/tasks O*Net and [bsky, 2 points, 1 comments]
• Hey buddy you dropped this on the way to posting this table arxiv.org/abs/2507.07935 [bsky, 2 points, 0 comments]
• New study analyzes 200k Copilot chats to map how #generativeAI impacts jobs. Most affected: info-heavy roles like sales, writing, & admin. Least affected: manual labor arxiv.org/pdf/2507.07935 [bsky, 1 points, 1 comments]
• Nesen Microsoft publicēja rakstu par profesijām, kuru darbus varētu darīt LLM: arxiv.org/abs/2507.07935 . Taču ChatGPT-5 nebūs tas īstais, kas vēl aizvietos - tas pat nemāk apkopot datus no šī raksta. [bsky, 1 points, 0 comments]
• According to a Microsoft-sponsored study (make of that what you will), the occupations find AI will most significantly impact are translators and interpreters - then historians ... 😳 arxiv.org/pdf/25 [bsky, 1 points, 0 comments]
• Investigadores de Microsoft analizaron cómo el uso de la IA generativa (Microsoft Bing Copilot) impacta en las diferentes actividades laborales y ocupacionales de las personas. 🔗 Artículo completo en [bsky, 1 points, 0 comments]
• Andrew stop lying. That’s not what the study is about: “Drawing instead on real-world AI usage, we analyze a dataset of 200k anonymized conversations with Microsoft Bing Copilot to measure AI applicab [bsky, 1 points, 1 comments]
• AI adoption concerns notwithstanding, data-driven analysis of adoption trends and use cases is always useful. (Friend of a friend co-authored this paper from Microsoft Research.) https://arxiv.org/pdf [bsky, 1 points, 0 comments]
• suena muy interesante. seguro te será muy útil. te comparto este paper escrito por investigadores de Microsoft que analiza cómo la inteligencia artificial está impactando realmente el mundo laboral. p [bsky, 1 points, 1 comments]
• No surprises in this paper, really. The magic I can do with some of these tools and tech is astonishing, but it's nothing compared to what I might if someone realized how my specialties and this paper [bsky, 1 points, 0 comments]
• KI wird den Arbeitswelt so stark verändern wie keine Technologie zuvor. Eine neue Studie von Microsoft - arxiv.org/pdf/2507.07935 - zeigt in welchen Berufe generative KI "anwendbar" ist und in welchen [bsky, 1 points, 0 comments]
• So embalming sounds promising, given all the humans AI is gonna end up erasing. arxiv.org/pdf/2507.07935 [bsky, 1 points, 0 comments]
• A prediction of #work most easily and least easily replaced by #AI. (Salt required, but interesting). Source: arxiv.org/abs/2507.07935 [bsky, 1 points, 0 comments]
• Según este estudio (PDF, página 12) de Microsoft, estos son los 40 trabajos que la IA tiene más probabilidades de impactar y los 40 que son seguros que se eliminarán. Aparecen traductores, escritores, [bsky, 1 points, 1 comments]
• Parece que, según el reciente estudio de Microsoft (arxiv.org/pdf/2507.07935), la profesión de historiador o historiadora es la segunda que más va a ser sustituida por la Inteligencia Artificial en el [bsky, 1 points, 1 comments]
• If I can summarize: The study finds that people in white collar jobs who hit the “thumbs up” on LLM-generated query responses are more likely to be in professions replaced by AI. Not sure where to sta [bsky, 1 points, 0 comments]
• And the tasks they have listed for these "replaceable" occupations are things like "examine materials for accuracy,"research historical or social issues,""compile records and documentation," and "prov [bsky, 1 points, 0 comments]
• Top 40 des métiers qui sont, seront probablement les plus impactés par l'IA selon un rapport de Microsoft. À l'inverse, les métiers les moins impactés sont ceux qui nécessitent des actions manuelles. [bsky, 1 points, 0 comments]
• arxiv.org/pdf/2507.07935 [bsky, 1 points, 0 comments]
• Entramos en una nueva etapa donde la automatización del conocimiento destruirá empleos como traductores, escritores, docentes y matemáticos. Vía Paper de Microsoft. arxiv.org/pdf/2507.07935 [bsky, 1 points, 0 comments]
• Une étude de Microsoft sur l'impact des IAs génératives sur le monde du travail liste les historiens et archivistes dans les quarante professions avec le plus haut "AI applicability score" arxiv.org/p [bsky, 1 points, 1 comments]
• Je vous vois partager ça et le gars il est sympa mais il a transformé "highest AI applicability score" par "40 jobs most at risks" et c'est un menteur. Lisez le résumé de l'étude arxiv.org/abs/2507.07 [bsky, 1 points, 1 comments]
• J'ai lu ca. Working with AI: Measuring the Occupational Implications of Generative AI. Franchement c'est moyen voire mauvais. arxiv.org/pdf/2507.07935 [bsky, 1 points, 1 comments]
• Les journalistes sont parmi les 20 métiers (16e rang exactement) les plus menacés ou transformés par l'IA, selon Microsoft arxiv.org/abs/2507.07935 🤖🧐 [bsky, 1 points, 1 comments]
• I ricercatori di Microsoft hanno sviluppato una bella indagine sul rapporto tra IA e lavoro. La Big Tech ritiene che i lavori usuranti e denigrati dalla società non saranno automatizzati, una buona no [bsky, 1 points, 0 comments]
• Working with AI: Measuring the Occupational Implications of Generative AI [hn, 1 points, 0 comments]
• Working with AI: Measuring the Occupational Implications of Generative AI [hn, 1 points, 0 comments]
• This is a Microsoft Research paper. Please do your research lol arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• AI is coming for the desk jobs you've been told to get. A recent Microsoft study proves it. This isn't a future threat for Gen Z—it's happening now. Your degree isn't the shield it used to be. Wake up [bsky, 0 points, 1 comments]
• So embalming sounds promising, given all the humans AI is gonna end up erasing. https://arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• Microsoft research into ‘AI application’ in employment sectors and jobs. It will probably be interpreted as ‘replaceability’ but it doesn’t go that far. #UncertaintyStrategies #AIStrategies arxiv.org/ [bsky, 0 points, 1 comments]
• ... 2/2 but i found the paper accessible and short and clear on what it was and was not trying to show. Here's the pdf of the paper. I'd you do a search for "bank teller" you'll find a paragraph on wh [bsky, 0 points, 0 comments]
• Here's the link: arxiv.org/abs/2507.07935 [bsky, 0 points, 0 comments]
• Microsoft's #AI job study notably omits positions like CEO and programmer. Turning over the preservation of history to a notoriously inaccurate system? Authors/Writers are once again seen as mere word [bsky, 0 points, 0 comments]
• arxiv.org/abs/2507.07935 [bsky, 0 points, 1 comments]
• arxiv.org/pdf/2507.07935 This is a comprehensive research paper analyzing how people use Microsoft Bing Copilot for work activities and its implications for different occupations. #ai #unemployment He [bsky, 0 points, 1 comments]
• I’m as wary of AI as the next guy, but the study referenced here isn’t saying these jobs are at risk of replacement but rather will get the greatest benefit from AI assistance. arxiv.org/abs/2507.0793 [bsky, 0 points, 0 comments]
• Microsoft study reveals the 40 jobs most at risk by AI and the 40 most secure Read the full report 👉 arxiv.org/pdf/2507.07935 #AI #Microsoft [bsky, 0 points, 0 comments]
• Occupational Implications [bsky, 0 points, 0 comments]
• This is not at all what the study is about or concludes. I'm a pretty anti-AI guy but this study was finding which professions would find the most use out of an AI assistant. arxiv.org/abs/2507.07935 [bsky, 0 points, 0 comments]
• This days a lot about Andrew, as this is an Arxiv artivle that's still in work. The latest version does not include this table. Also switchboard operators??? arxiv.org/abs/2507.079... [bsky, 0 points, 0 comments]
• To be fair, that's not Fortune's list; it's from a Microsoft preprint paper (arxiv.org/pdf/2507.07935). Still utter bollocks though. Though, if the argument is about 'impact' - fair to debate whether [bsky, 0 points, 0 comments]
• The full study: "Measuring the Occupational Implications of Generative AI" arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• Source: arxiv.org/abs/2507.07935 [bsky, 0 points, 0 comments]
• Los 40 empleos que mejor aprovechan el uso de IA en la actualidad según un estudio de Microsoft Y no, no perderán ni están perdiendo sus puestos. Fuente: arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• Here top 40 professions that will be replaced by AI first and the last, published by Microsoft arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• [PDF] Working with AI: Measuring the occupational implications of Generative AI arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• Is your job at stake because of AI? That depends of what your job is. See: arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• OP is misleading about the study. The study didn't look at which jobs could be replaced with AI, but which tasks in those jobs are most likely to either use an AI chatbot. Also the values in OP's tabl [bsky, 0 points, 1 comments]
• arxiv.org/abs/2507.07935 [bsky, 0 points, 1 comments]
• Dishwashers, you are safe. arxiv.org/abs/2507.07935 [bsky, 0 points, 0 comments]
• arxiv.org/abs/2507.07935 [bsky, 0 points, 0 comments]
• Evidently that list comes from this research paper, and that’s not an accurate description of what the authors are saying arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• New paper from Microsoft re-runs the analysis of which jobs will be automated first. Translators come top, which seems right, but passenger assistants? I can't imagine GPT-4 pointing out the exit door [bsky, 0 points, 0 comments]
• Microsoft reports on what people most commonly use copilot for and what impact that has on jobs: arxiv.org/abs/2507.07935 [bsky, 0 points, 1 comments]
• #AI is going to put us all out of work if we #write or have other white collar jobs. But if you’re a dredge operator you are safe. This new study should stop you in your tracks. arxiv.org/pdf/2507.079 [bsky, 0 points, 0 comments]
• New Microsoft AI paper asks a big question: What do users want Copilot to do, and what does it actually do? Turns out, there’s a measurable gap between user intent, Copilot behavior, and real-world wo [bsky, 0 points, 0 comments]
• arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• Interestingly, the safest future occupation is "Dredge Operator," one who "removes sand, gravel, or other materials in order to excavate and maintain navigable channels in waterways." Given that DATA [bsky, 0 points, 0 comments]
• No, they don't. The study clearly states: "It is tempting to conclude that occupations that have high AI action applicability score will be automated and thus experience job or wage loss [...] This wo [bsky, 0 points, 0 comments]
• You can find the research paper here: arxiv.org/pdf/2507.07935 [bsky, 0 points, 0 comments]
• Embalsamador Reparador de tejados, de neumáticos, de barcos Operarios de dragas, de maquinaria de asfaltado, de ETAP/EDAR En definitiva, las mayores ganancias en productividad (y despidos) se van a da [bsky, 0 points, 1 comments]
• Source: arxiv.org/abs/2507.07935 [bsky, 0 points, 0 comments]
• The paper is linked in the caption of the graph. They used ChatGPT arxiv.org/pdf/2507.07935 [bsky, 0 points, 1 comments]
• See the actual Microsoft Research: Measuring the Applicability of Generative AI to Occupations arxiv.org/pdf/2507.079... [bsky, 0 points, 0 comments]
• Para quien quiera leerse el "estudio" que a mi parecer suena torticero https://arxiv.org/pdf/2507.07935 Porfi periodistas, no le pidáis fuentes a los LLM, y pedid asistencia a expertos independientes [bsky, 0 points, 0 comments]
• Microsoft maakte een rapport over de beroepen die het meest en het minst zullen worden aangetast door AI. Voor je beroepskeuze: bouwvakker is meer toekomstvast dan programmeur. Rapport: arxiv.org/pdf/ [bsky, 0 points, 1 comments]
• Microsoft araştırmacıları yapay zekânın en çok (ve en az) uygulanabildiği 40 işi açıklayan bir rapor yayınladılar. Listenin başında tercümanlar ve çevirmenler var. arxiv.org/abs/2507.079... [bsky, 0 points, 0 comments]
• You can find the article here (pdf link at bottom). arxiv.org/abs/2507.07935 [bsky, 0 points, 0 comments]

Related

• Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversations
• Towards the Terminator Economy: Assessing Job Exposure to AI through LLMs
• AI and jobs. A review of theory, estimates, and evidence
• What Work is AI Actually Doing? Uncovering the Drivers of Generative AI Adoption
• The Potential Impact of AI Innovations on U.S. Occupations