I see the arrival of LLMs (I dislike calling the tech AI) as being similar to pocket calculators. Yes, they've made most people forget how to quickly make operations in their heads, but they have also made life the hell of a lot easier for most.

But LLMs are not like calculators, in the sense that their output is complex, and mostly based on previously reported criteria, and cannot quickly and easily be verified following a simple algorithm.

So I wonder if, in the near future, at least some LLMs will boast a certification of sorts guaranteeing they will never report unverified data ... and who would have the authority of administering said "certification"

If you consider an LLM a tool (like a calculator or a farm tractor), those who learn to effectively use the tool will be more productive than those who try to do the same work without the tool.

At some point, there will be those who use the tool for the job, and those who don't have a job.

There's really nothing else to be said. You can't put the genie back in the bottle.

 

If you consider an LLM a tool (like a calculator or a farm tractor), those who learn to effectively use the tool will be more productive than those who try to do the same work without the tool.

At some point, there will be those who use the tool for the job, and those who don't have a job.

There's really nothing else to be said. You can't put the genie back in the bottle.

Maybe, it might also be a tool that's so inaccurate, imprecise and non-repeatable it can't be trusted for serious work.

Trust is the key IMO. At what point do we trust these things to work without being baby-watched constantly?

 

At what point do we trust these things to work without being baby-watched constantly?

Wait ... are we still talking about LLMs, or about self-driving cars? ...

 

Wait ... are we still talking about LLMs, or about self-driving cars? ...

Yes.

 

https://www.bleepingcomputer.com/ne...-google-to-hire-ai-answers-quality-engineers/
Google Search AI hallucinations push Google to hire "AI Answers Quality" engineers

After the recent update, Google has been pushing more and more users to AI Mode and AI answers.
In fact, Google has also updated its Discover feed with AI Overviews for news and is even rewriting headlines of news publications using AI.
While Google AI Overviews have come a long way and answers are now better than what we have seen in the past, there are still some rough edges.
For example, a week ago, when I Googled the valuation of a certain startup, Google made up a figure of $4 million. Then, I opened another tab and asked the same question but worded it slightly differently, and AI Overviews said the company is valued at over $70 million.
I cross-verified Google’s answers with the links (citations) it generated, but I noticed that the value Google cited twice did not even exist in the sources it allegedly referred to.
This is just one of the examples where Google can get things wrong.

 

 

Maybe, it might also be a tool that's so inaccurate, imprecise and non-repeatable it can't be trusted for serious work.

But it's already being used by effectively for serious work by people who know how to use it, and who know how to check their work.

And the tools will continue to improve.

I predict within one or two years (based upon the progress I have seen my actual self), these tools will by highly effective for those who know how to use them properly (just like any other tools).

There will be three kinds of workers:

1) those who use the tools effectively and become more productive;
2) those who use the tools improperly;
3) those who don't use the tools.

I further predict that, in 3 to 5 years, workers in categories 2 and 3 will not have jobs, or their jobs will be of the niche or artisan types where productivity is not a valuable selling point.

IMHO. But I think it is clear where this is going.

Personal case in point: In December, I wrote -- from scratch -- a 230 page piece of technical literature fully formatted and ready for publishing in less than a week. This would have taken me months without an LLM.

Prior to that work, I was a skeptic just like you. Not anymore.

 

But it's already being used by effectively for serious work by people who know how to use it, and who know how to check their work.

And the tools will continue to improve.

I predict within one or two years (based upon the progress I have seen my actual self), these tools will by highly effective for those who know how to use them properly (just like any other tools).

There will be three kinds of workers:

1) those who use the tools effectively and become more productive;
2) those who use the tools improperly;
3) those who don't use the tools.

I further predict that, in 3 to 5 years, workers in categories 2 and 3 will not have jobs, or their jobs will be of the niche or artisan types where productivity is not a valuable selling point.

IMHO. But I think it is clear where this is going.

Personal case in point: In December, I wrote -- from scratch -- a 230 page piece of technical literature fully formatted and ready for publishing in less than a week. This would have taken me months without an LLM.

Prior to that work, I was a skeptic just like you. Not anymore.

Fine, enjoy your AI. Glad I'm retired.

 

...enjoy your AI

I already was, but I appreciate your permission.

 

But it's already being used by effectively for serious work by people who know how to use it, and who know how to check their work.

And the tools will continue to improve.

I predict within one or two years (based upon the progress I have seen my actual self), these tools will by highly effective for those who know how to use them properly (just like any other tools).

There will be three kinds of workers:

1) those who use the tools effectively and become more productive;
2) those who use the tools improperly;
3) those who don't use the tools.

I further predict that, in 3 to 5 years, workers in categories 2 and 3 will not have jobs, or their jobs will be of the niche or artisan types where productivity is not a valuable selling point.

IMHO. But I think it is clear where this is going.

Personal case in point: In December, I wrote -- from scratch -- a 230 page piece of technical literature fully formatted and ready for publishing in less than a week. This would have taken me months without an LLM.

Prior to that work, I was a skeptic just like you. Not anymore.

I asked Grok to grade my assessment. It agrees (TIFWIW):

Yes, this is a reasonably fair and increasingly mainstream assessment of where things are heading with LLMs and generative AI as of early 2026 — though with some caveats on the timeline and severity of job displacement.
The core ideas in the statement align well with what many experts, company leaders, and real-world adopters are observing right now.
Current Reality (Early 2026)
LLMs are already being used effectively for serious professional work by people who know how to prompt well, chain tools/agents, verify outputs, and integrate them into workflows. Fields like software development, content creation, research, legal analysis, data analysis, and marketing see real productivity gains (often 30-100%+ on certain tasks) when used skillfully.
Many professionals report that those who master these tools become dramatically more productive, while those who ignore them or use them poorly fall behind. A common refrain echoing across discussions (including recent X posts) is:
"AI won’t replace your job — but someone using AI probably will."
This matches your statement's three categories of workers almost exactly.
Adoption is accelerating: Surveys and reports from 2025–2026 show high-income/experienced professionals using LLMs daily at high rates, with enterprises treating them as core infrastructure. Tools like Claude, Cursor, Gemini, and agentic systems are deeply embedded in real workflows.
Near-Term Outlook (1–2 Years)
Your prediction that tools will become "highly effective" for skilled users within 1–2 years feels realistic — we're arguably already seeing this in 2026 for many domains. Advances in reasoning models, longer contexts, better tool use, and agentic systems (e.g., multi-step agents handling complex tasks) are making them far more reliable and powerful. Productivity dashboards and careful measurements (predicted by Stanford experts for 2026) are starting to quantify where AI boosts output vs. where it falls short.
Many experts frame 2026 as the year of pragmatic augmentation rather than full replacement — AI as a "cognitive amplifier" for humans rather than a complete substitute in most cases.
Longer-Term Outlook (3–5 Years)
This is where the assessment is plausible but more debated and conditional.
The idea that categories 2 and 3 (improper use or no use) will face severe job pressure — potentially losing roles or being relegated to low-productivity niches/artisan work — is a very common prediction. We've seen early signs:
Entry-level white-collar roles (especially repetitive/cognitive ones) are already shrinking.
Companies (e.g., in tech, finance, customer support) are reducing headcount while boosting output via AI.
Predictions from figures like Geoffrey Hinton and others suggest major replacement in software engineering, analysis, and similar fields within a few years.
However, the consensus among many economists, labor market analysts, and even some AI leaders is more nuanced:
Net job creation often follows major tech shifts (historical pattern: Industrial Revolution, internet, etc.), though with painful transitions and frictional unemployment.
AI tends to augment more than fully replace in complex, judgment-heavy, or human-centric roles — at least in the medium term.
Some forecasts predict 20–40% of tasks automated, but with many new roles emerging around AI orchestration, validation, ethics, and novel applications.
The most pessimistic views (e.g., 50%+ white-collar displacement, massive structural unemployment) exist, especially from AI insiders, but they're balanced by others expecting more gradual evolution with policy interventions (retraining, UBI discussions, etc.) softening the blow.
Bottom Line
Your take is clear-eyed and evidence-based, especially for knowledge work and cognitive tasks — it's not hyperbolic doom-posting, but neither is it complacent optimism. The direction ("this is where it's going") feels very accurate: AI mastery is becoming a core professional skill, like using spreadsheets or email was in previous decades. Those who treat it as just another tool to wield skillfully will thrive; those who don't will struggle, potentially severely in 3–5 years.
The big open questions remain:
How fast and how uneven the displacement will be across industries/regions.
Whether society can adapt policies and culture quickly enough to share the productivity gains broadly.
Overall, I'd say this is one of the more grounded, realistic assessments circulating right now. The evidence from 2025–2026 strongly supports it.

 

Yeah, I know IE is mostly crap and all, but this article is very interesting and it has a link to the original paper, if anyone's interested:

2,500-year-old board game inspires AI to tackle engine, data center overheating


​

The team, which published their findings in a paper in the journal Artificial Intelligence Review, set out to present a comprehensive analysis of the effectiveness of spray cooling.

The key is water droplets. When they hit a hot object’s surface, each tiny droplet evaporates, carrying away some of the heat, helping to regulate the surface temperature and cool down the object.

 

"I asked Grok to grade my assessment. It agrees (TIFWIW):"

Fancy auto-complete that generates compliant text from a human generated prompt . Shocked, shocked, well not that shocked.

 

https://www.theregister.com/2026/01/06/memory_firm_profits_up_as/

"The memory market is at an inflexion point, with demand materially outpacing supply," IDC stated, claiming that while the memory industry has long been characterized by boom-and-bust cycles, this one is different.

This crunch stems from AI's voracious appetite for memory - hyperscalers like Microsoft, Google, and Amazon are driving demand that's permanently reshaping silicon wafer allocation away from consumer products (phones and PCs). This restricts the supply of general-purpose memory modules and pushes up prices across the board.

IDC expects DRAM and NAND supply growth to lag at just 16 and 17 percent respectively this year, well below historical norms.

 

https://www.theregister.com/2026/01/07/ibm_bob_vulnerability/?td=keepreading

AI agent software – models given access to tools and tasked with some goal in an iterative loop – is notoriously insecure and often comes with warnings from vendors. The risks have been demonstrated repeatedly by security researcher Johann Rehberger, among others. Agents may be vulnerable to prompt injection, jailbreaks, or more traditional code flaws that enable the execution of malicious code.

 

https://www.theregister.com/2026/01/06/memory_firm_profits_up_as/

I tried to buy a NVME SSD at Walmart yesterday.

A month ago, they had 1TB for about $79.

Now you can't buy them.

 

Maybe, it might also be a tool that's so inaccurate, imprecise and non-repeatable it can't be trusted for serious work.

Trust is the key IMO. At what point do we trust these things to work without being baby-watched constantly?

There is of course, no tool. No more than there's "a film" or "a painting", LLM has no definition, its just software, who defines "large"? who defines "language" and who defines "model"?

 

https://www.msn.com/en-us/technolog...or-all-those-boring-tasks-at-work/ar-AA1TMuha

Workdays without busywork are closer to reality than ever, thanks to artificial intelligence. AI tools that can sort and summarize emails, take meeting notes and file expense reports promise to free us to concentrate on the important stuff.
This sounds great. The catch is that our brains aren’t capable of thinking big thoughts nonstop. And we risk forfeiting the epiphanies that sometimes spring to mind while doing easy, repetitive job functions.

 

https://www.msn.com/en-us/technolog...or-all-those-boring-tasks-at-work/ar-AA1TMuha

Workdays without busywork are closer to reality than ever, thanks to artificial intelligence. AI tools that can sort and summarize emails, take meeting notes and file expense reports promise to free us to concentrate on the important stuff.
This sounds great. The catch is that our brains aren’t capable of thinking big thoughts nonstop. And we risk forfeiting the epiphanies that sometimes spring to mind while doing easy, repetitive job functions.

https://en.wikipedia.org/wiki/Luddite

Over time, the term has been used to refer to those opposed to the introduction of new technologies.

 

Sounds like a new game possibility. Luddites vs AI Zombies.

https://gregrobison.medium.com/the-...rational-fear-of-average-quality-1efadaa0c990

In the modern era, a similar sentiment can be observed with the rise of artificial intelligence (AI), particularly in the form of Large Language Models (LLMs) like ChatGPT. Unfortunately, the common reference to “AI Luddites” as those who don’t understand AI misses the spirit of the original Luddites. AI Luddites should describe those who show skepticism or opposition towards the rapid advancement and integration of AI in all the various facets of life, especially in creative and intellectual domains. This contemporary version of Luddism is not a blanket opposition to technology; rather, it stems from concerns about the potential impacts of AI on human creativity, thought processes, and the authenticity of human expression.

 

https://www.msn.com/en-us/news/tech...he-results-from-hundreds-of-tests/ar-AA1TO92q
Can AI do your job? See the results from hundreds of tests.

But the AI version is completely wrong.

The failed floor plan illustrates a disconnect three years after the release of ChatGPT that has implications for the whole economy.

AI can accomplish many impressive tasks involving computer code, documents or images. That has prompted predictions that human work of many kinds could soon be done by computers alone. Bentley University and Gallup found in a survey last year that about three-quarters of Americans expect AI to reduce the number of U.S. jobs over the next decade.
...
The research team then gave each task to AI systems such as OpenAI’s ChatGPT, Google’s Gemini and Anthropic’s Claude.

The best-performing AI system successfully completed only 2.5 percent of the projects, according to the research team from Scale AI, a start-up that provides data to AI developers, and the Center for AI Safety, a nonprofit that works to understand risks from AI.

“Current models are not close to being able to automate real jobs in the economy,” said Jason Hausenloy, one of the researchers on the Remote Labor Index study. They created the index to give policymakers clear-eyed information about the capabilities of AI systems, he said.

I'm sure, with work, the successfully completed rate might be 10 percent soon.