發生了什麼事
Business Insider reports that workers are retroactively editing old LinkedIn job titles and descriptions to add terms such as “AI,” “GPT,” “LLM,” and “artificial intelligence.” The report attributes the findings to a new, non-peer-reviewed National Bureau of Economic Research working paper based on monthly snapshots of 29.4 million U.S. profiles collected by Revelio Labs from August 2020 through January 2026.
Business Insider reports that the underlying NBER working paper examined monthly snapshots of 29.4 million U.S. LinkedIn profiles. The profiles were collected by labor-market data firm Revelio Labs between August 2020 and January 2026. The paper’s authors included Stanford economist Nicholas Bloom and researchers at Stanford and Revelio Labs. Because the paper is non-peer-reviewed, its methods and conclusions have not yet passed formal peer review according to the report. The large sample gives the analysis broad coverage of public professional-profile changes, but the supplied source does not provide the paper’s full methodology, statistical tables, or a separate replication.
Business Insider says that nearly one-fifth of the accounts in the study had retroactively changed the title or description of a job they had already left. Since ChatGPT’s release in late 2022, the paper reported that retroactive additions of terms including “AI,” “GPT,” “LLM,” and “artificial intelligence” to previous roles rose more than sixfold. The report describes this as workers going back to revise their professional histories and notes that the researchers called the practice “time travel.” The evidence concerns changes to profile language; it does not establish that the workers’ underlying duties changed or that they began using AI only when the terms were added.
The pattern varied by industry. Business Insider reports that 31.6% of workers in technology and information had retroactively edited an old role, compared with a study-wide rate of 19.7%. The reported rates were 25% in arts, entertainment, and recreation and 24.1% in professional, scientific, and technical services. The median edit occurred more than four years after the job ended. The report says these changes often happened when workers were preparing to move jobs, which the researchers interpreted as a possible signal of what applicants believe employers want. The source does not say how many profiles were edited more than once or how the researchers identified job-search timing.
The same analysis found changes in other kinds of language. Business Insider reports that workers became less likely to add terms such as “remote,” “work from home,” and “WFH” to past roles. By the end of 2025, additions and deletions of those terms were occurring at roughly similar rates, a shift the report says coincided with return-to-office mandates. The paper also found a sharp decline in additions of “diversity,” “equity,” “inclusion,” and “DEI” at the beginning of 2025, coinciding with executive orders targeting DEI programs. The supplied material reports these timing relationships but does not establish that the policy changes directly caused individual profile edits.
為什麼這很重要
The findings suggest that professional profiles are changing records of past work as workers respond to shifting employer preferences. They also show why LinkedIn data should not automatically be treated as a stable historical record of when people acquired or used AI skills.
The most important implication is that a professional profile can be a moving interpretation of the past rather than a fixed historical record. Business Insider reports that the paper estimated a 2026 snapshot would overstate the prevalence of AI-related skills in 2022 by about 30%. In practical terms, a researcher studying older profiles could mistake later wording changes for evidence that AI skills were already commonly documented at the earlier date. That matters for labor-market analysis, hiring research, and any attempt to measure when AI capabilities spread through occupations.
The findings also offer evidence about how workers respond to changing labor-market signals. The paper’s authors wrote that workers modify résumés to reflect what they believe employers want to hear. The reported increase in AI language may therefore measure both genuine experience and strategic presentation. A worker might be adding detail that was previously omitted, translating existing tasks into terminology that is more attractive to employers, or making a stronger claim about the relevance of past work. The source does not provide enough information to distinguish reliably among those possibilities.
The AI connection is consequential because the reported edits concern how people represent their careers as employers place more value on AI-related capabilities. The study does not show that AI has increased productivity, eliminated jobs, or improved hiring outcomes. It instead documents a change in the language of professional identity. That narrower finding is still useful: if job seekers increasingly describe old work through an AI lens, employers and researchers may need to evaluate the underlying tasks, evidence of competence, and dates of experience rather than relying on keyword counts alone.
The comparison with remote-work and DEI terminology shows that profile language can respond to broader institutional and political shifts, not only to changes in a person’s work. Business Insider links the remote-language pattern to return-to-office mandates and the DEI-language pattern to executive orders, while reporting those relationships as timing coincidences from the study. This suggests that retrospective profiles may reflect perceived reputational or hiring incentives. It does not prove that workers abandoned remote work or DEI practices, nor does it establish that every deletion represented a change in belief or experience.
互動機制：它實際上是如何運作的
以互動方式探索這項發展背後的基礎技術。
A route planner searches possible journeys using explicit rules. What does this illustrate about AI?
接下來看什麼
The supplied report does not independently confirm the paper’s findings through a separate or study. Important unknowns include how many edits reflect previously omitted experience rather than new claims, whether employers reward the changes, and whether the pattern continues as demand for AI-related skills evolves.
A key next step is independent scrutiny of the NBER working paper. The supplied report identifies the and broad time period but does not provide an independent replication, a detailed account of sampling limitations, or evidence that the profile changes were verified against employment records. Peer review or a separate analysis using archived profiles, résumé data, job applications, or employer records could test whether the reported pattern is robust and whether it generalizes beyond LinkedIn users in the United States.
Researchers and employers should watch how AI-related language is defined and interpreted. The terms “AI,” “GPT,” “LLM,” and “artificial intelligence” can describe many different activities, from using a general-purpose tool to building or evaluating models. The source does not state whether the study separated technical model development from ordinary workplace use, nor whether it measured proficiency. Without that distinction, rising keyword prevalence should not be treated as a direct measure of the number of skilled AI practitioners.
Hiring practices are another area to monitor. The report does not say whether retroactive AI wording improves interview rates, compensation, promotions, or job placement. It also does not establish whether employers can identify inflated or vague claims. More evidence would be needed to determine whether keyword changes merely follow employer demand or actively influence hiring decisions. Practical evaluation may increasingly require candidates to explain specific tasks, tools, outputs, and results rather than relying on broad AI labels.
Finally, future snapshots should be interpreted carefully. The paper’s reported estimate that a 2026 view could overstate 2022 AI-skill prevalence by about 30% indicates that historical datasets may need versioning so later edits do not overwrite earlier records. The supplied source leaves open whether the pace of retrospective edits will continue, slow, or reverse if employer preferences change. It also leaves unknown whether the same patterns appear in other countries, occupations, professional networks, or workers who do not maintain public online profiles.