Three years of Chat GPT – generating headlines but few job losses (so far)

The last day of next month will mark the third anniversary of OpenAI launching ChatGPT.

OpenAI described their new product at the time, thus.

We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer follow up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.

GPT 5.0 was released just over two months ago and, it’s fair to say, things didn’t go quite to plan for OpenAI.

Matthew Smith, contributing editor for IEEE Spectrum, wrote last month in his article, GPT-5’s Rocky Launch Underscores Broader AI Disappointments: Is AI headed toward the trough of disillusionment?

GPT-5 was supposed to be the model that proved artificial general intelligence (AGI) is within reach. 

Cognitive scientist and AGI skeptic Gary Marcus called GPT-5 “overhyped and underwhelming“ in a Substack post, and the deluge of negative feedback eventually prompted (Open Ai founder, Sam) Altman to admit OpenAI “totally screwed up” the launch.

As part of the new model’s release, OpenAI removed the earlier GPT-4o model from ChatGPT on the apparent assumption that users would find GPT-5 an upgrade in every situation. Instead, many ChatGPT users complained that the new model seemed worse than its predecessor. The criticism caused OpenAI to change course and restore access to GPT-4o just 24 hours after its removal.

It was an embarrassing turn of events for OpenAI. In 2024, Altman predicted that GPT-5 would make GPT-4 feel “mildly embarrassing” by comparison. Instead, user feedback to GPT-5 was so negative that OpenAI decided to restore its predecessor.

Recent releases from Grok and Anthropic also received a tepid response.

The underwhelming recent model updates of the major GenAI companies has led to increased frequency of headlines speculating on the health of the “AI bubble”.

This speculation has been fuelled by data showing how dominant investment in GenAI tech has been in the United States in the past two years.

Earlier this week Investing.com published an article “America is now one big bet on AI”, in which head of Rockefeller Capital Management’s international business, Ruchir Sharma is quoted, “AI companies have accounted for 80 per cent of the gains in US stocks so far in 2025.” In fact, more than a fifth of the entire S&P 500 market cap is now just three companies — Nvidia, Microsoft, and Apple — two of which are basically big bets on AI.

The chart below is a vivid illustration of how quickly investment in tech assets has dominated recent growth in the United States’ GDP.

Investors must have been slightly more than nervous when Bloomberg reported, “Leaders at AI computing company CoreWeave Inc. sold shares worth more than $1 billion after a lockup on the stock lifted in mid-August, putting them among the top 10 individual insider sellers of the third quarter. Some of New Jersey-based CoreWeave’s investors followed suit, with its biggest institutional owner Magnetar Financial LLC selling nearly $1.9 billion worth of shares over the same time period.”

If company insiders are selling in lockstep then other investors would have every right to wonder what those insiders know that they don’t.

To date, the labour market impact of GenAI is not quite what the AI-bolsters would have expected when large scale data sets have been examined.

The pwc 2025 Global AI Jobs Barometer reported

• Industries most able to use AI have 3x higher growth in revenue generated by each employee
• Since 2022, productivity growth in industries best positioned to adopt AI has nearly quadrupled (while falling slightly in industries least exposed to AI)
• Jobs are growing in virtually every type of AI-exposed occupation, including highly automatable ones.

Al job loss alarmism was further soothed two weeks ago when Yale University published Evaluating the Impact of AI on the Labor Market: Current State of Affairs.

To answer these questions, the researchers compared how quickly the occupational mix has changed across a range of measures since ChatGPT’s launch and examined the evidence of economy-wide employment effects, compared this to past disruptions from computers and the internet.

The researchers’ two main conclusions from comparing the pace of labour market change in the 33-month period since the launch of ChatGPT, to the employment disruption from past periods of early technological change, were:

1. While the occupational mix is changing more quickly than it has in the past, it is not a large difference and predates the widespread introduction of AI in the workforce.
2. Currently, measures of exposure, automation, and augmentation show no sign of being related to changes in employment or unemployment.

The researchers noted that while their findings may contradict the most alarming headlines, it is not surprising given past precedents. Historically, widespread technological disruption in workplaces tends to occur over decades, rather than months or years.

In August, U.S. independent think tank, Economic Innovation Group, reported that US workers whose jobs involve tasks that AI can do are actually much less likely than other workers to be unemployed (see chart, above). They’re also much less likely to be leaving the labour force.

“The evidence that we have so far about the level of diffusion and the actual application of AI in the labour market suggests that these more recent movements [in US jobs data] are not going to be related to AI,” said Nathan Goldschlag, EIG’s director of research.

“When zooming out across the whole US economy, it’s still too early to see widespread job loss from AI,” said Bharat Chandar, a labour economist at the Stanford Digital Economy Lab, speaking to Bloomberg Weekend.

However, he cautions that Census datasets can involve relatively small samples for certain subsets of workers — like recent college graduates. “The main data sources used by researchers are not detailed enough yet to give us an up-to-date view of what’s happening with 22-25 year olds in the most AI-exposed jobs, for example,” he said.

It’s been proven that GenAI can undertake specific work-related tasks with a high degree of skill, which is not the same as GenAI replacing jobs.

Wharton School professor Ethan Mollick, who specialises in innovation and AI, summarised the most recent research at OpenAI last month.

OpenAI released a new test of AI ability, but this one differs from the usual benchmarks built around math or trivia. For this test, OpenAI gathered experts with an average of 14 years of experience in industries ranging from finance to law to retail and had them design realistic tasks that would take human experts an average of four to seven hours to complete..

OpenAI then had both AI and other experts do the tasks themselves. A third group of experts graded the results, not knowing which answers came from the AI and which from the human, a process which took about an hour per question.

Human experts won, but barely, and the margins varied dramatically by industry.

If the current patterns (of advancement in AI models) hold, the next generation of AI models should beat human experts on average in this test

Mollick poses, and answers, the obvious question, “Does this mean AI is ready to replace human jobs?”

Although AI has the capability to undertake specific tasks of a job better than a human, it does not necessarily replace the entire job, it shifts where the job holder spends their time and adds value.

As Mollick summarises, “…as long as AI is jagged in its abilities and cannot substitute for all the complex work of human interaction, it cannot easily replace jobs as a whole.”

This is even less surprising when research shows that organisational efforts to drive GenAi’s promised transformative effect on worker productivity have, so far, bombed.

The GenAI Divide: State of AI in Business 2025, a new report published by MIT’s NANDA initiative, based on 150 interviews with leaders, a survey of 350 employees, and an analysis of 300 public AI deployments, reveals that despite the rush to integrate powerful new models, about 5% of AI pilot programs achieve rapid revenue acceleration; the vast majority stall, delivering little to no measurable impact to profitability.

The core issue, according to the researchers was not the quality of the AI models, but the “learning gap” for both tools and organisations. While executives often blame regulation or model performance, MIT’s research points to flawed enterprise integration. Generic tools like ChatGPT excel for individuals because of their flexibility, but are far less effective in enterprise use as they fail to learn from or adapt to workflows.

Despite a widespread fear by company boards and CEOs they may be late starters in the pursuit of “AI-led business transformation” the evidence, from the U.S. at least, shows that even among large companies AI adoption rates, although rising, remain very low (See chart, below).

Five months ago, a doomsday quote from Dario Amodei the CEO of Anthropic and former VP of Research, OpenAI, generated headlines around the world.

“AI could wipe out half of all entry-level white-collar jobs — and spike unemployment to 10-20% in the next one to five years.

AI companies and government need to stop “sugar-coating” what’s coming: the possible mass elimination of jobs across technology, finance, law, consulting and other white-collar professions, especially entry-level gigs.”

At this stage in Amodei’s predicted five-year timespan the evidence suggests a jobs apocalypse remains in the distance, rather than just around the corner.

Note: I have a 50 minute keynote and a two hour seminar about this topic that I update weekly, adaptable for both inhouse and public audiences, either industry-specific or broad-based. If you want your team, company, members or clients to be up-to-date about the how GenAI is impacting jobs, careers, and the labour market, then contact me via [email protected] for more information.

Related blogs

Recruiters get ready – the unsettling impact of GenAI on careers is just beginning

GenAI in hiring reaches flashpoint as legal precedent looms

AI clones Australian radio host for 6 months (and nobody notices)

Could Hollywood writers have pointed the way to a sensible AI future?

AI powers its way into recruiters’ jobs with GPT-4 (and its cousins)