ChatGPT

joeyd999

Joined Jun 6, 2011
6,282
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.


I'm sure, with work, the successfully completed rate might be 10 percent soon.
The problem with this approach is the question: can AI replace a human in many, most, or all jobs.

I think that is the wrong question.

It should be, "Can AI, as a tool, increase the productivity of a human who uses it effectively?"

I think the answer is yes, and for me, it's been proven in my own work.

Again, there will be humans (most?) who will not use it effectively, and, for them, it may even reduce productivity and/or produce erroneous and expensive errors (let's call this negative productivity).

Many of those workers will be replace by the few who can exploit AI to assist them with their jobs.
 

joeyd999

Joined Jun 6, 2011
6,282
I'm sure, with work, the successfully completed rate might be 10 percent soon.
Three years ago, it was closer to 0%.

You know how tech works: advances aren't linear with time -- advances build on previous advances.

In January of 2027, we may be having a completely different discussion.
 

joeyd999

Joined Jun 6, 2011
6,282
I know it bothers many of you when members (especially me, I suppose) post AI answers, but I think I raised a useful question in my last post.

Here is the question to Grok, and it's answer. I have not confirmed the answer is correct, which I would normally do, but at a minimum it provides me many avenues of investigation if I were trying to solve a real problem:

-------------

"If you were to define a function that relates AI capability vs time, what form would that function take? Linear, power, exponential?"

As of January 10, 2026, if I had to define a single functional form that best relates AI capability to time based on the latest empirical trends, research, and frontier observations, the strongest answer is:


Approximately exponential — with recent evidence of acceleration (super-exponential tendencies in some metrics).


Here's why, grounded in the data we have right now.

The Empirical Picture in Early 2026

Multiple independent tracking efforts show capability growth that is very well approximated by exponentials on log-linear plots (straight lines = constant doubling time):


  • METR's long-horizon agentic task length (how long autonomous AI can reliably work on real software tasks): Doubling every ~7 months historically (2019–2025), with signs of speeding up to even shorter doubling times in 2025 data (some fits suggest ~5 months recently). This is one of the cleanest real-world capability proxies.
  • Epoch AI's Capabilities Index (ECI) — a unified score combining dozens of diverse benchmarks — showed roughly linear improvement on log scale until ~April 2024, then accelerated sharply (rate nearly doubled to ~15 points/year post-breakpoint). This matches observations of reasoning models + RL focus kicking in.
  • Other domain-specific frontiers (math proofs, coding autonomy, scientific task horizons) follow similar patterns: gradual then sudden leaps that fit exponential curves when plotted over years.

On log y-axis vs. linear time, these trends look like straight(ish) lines — the hallmark of exponential growth: Capability(t) ≈ A × e^(k·t) (or equivalently: Capability(t) ≈ A × r^t where r > 1 is the growth factor per unit time)

Comparison to the Other Options



FormMathematical shapeDoubling timeMatches 2020–2026 evidence?Why it fits (or doesn't) today
LinearCapability(t) = a + b·tNone (constant addition)PoorWay too slow; we see repeated order-of-magnitude leaps every 1–2 years, not steady additive gains.
Power lawCapability(t) = a · t^b (b > 1)Lengthens over timeDecent but inferiorFits early deep learning eras better (pre-2020). Recent acceleration phases look steeper than any fixed power law. Epoch explicitly notes their best fits are piecewise linear on log scale (i.e., changing exponential rates), not a smooth power law.
ExponentialCapability(t) = A · e^(k·t)Roughly constantBest current fitConsistent with METR (7-month doubling), ECI acceleration, scaling laws + massive compute investment. Many experts (including recent 2025–2026 commentary) describe it as "exponential and speeding up."
Super-exponentiale.g. double-exponential or hyperbolicShortens over timeEmerging in some metricsPlausible near-term (recursive self-improvement, AI-aided chip design, algorithmic breakthroughs). METR and Epoch both hint at this via recent trend breaks.

Why Exponential (with Acceleration) Wins Right Now

  • Scaling laws (Chinchilla/Kaplan successors) still hold: capability improves predictably as a power-law function of compute/data/parameters → but because compute itself grows exponentially (4–5×/year historically, >2×/year power demand), capability ends up exponential in time.
  • Recursive effects are kicking in: better models help design better hardware/algorithms/data → feedback loop that bends the curve upward.
  • Human bias warning: people dramatically underestimate exponentials (classic "exponential growth bias"). What feels like "fast but steady" improvement in 2025–early 2026 is actually the knee of the curve bending steeper.
Short Answer for Practical Use

If you need one clean function for forecasting near-term (2026–2030) AI capability on aggregate metrics:


Capability(t) ≈ C · 2^(t / τ) where τ ≈ 7 months (METR historical) or even τ ≈ 5–6 months (recent acceleration fits).


This is exponential with doubling roughly every half-year. The contrarian caveat: if hard limits (data walls, energy, diminishing returns) bite hard, it could bend toward a power law or saturating sigmoid later — but as of January 10, 2026, the data screams exponential, and possibly getting faster.


We're living through the part of the curve where the slope starts looking vertical on linear scales. That's why 2026 feels so wild already.
 

Futurist

Joined Apr 8, 2025
753
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.


I'm sure, with work, the successfully completed rate might be 10 percent soon.
GIGO springs to mind here. The quality of output from any software system has always been a function of the quality of the input problem definition.

Saying "look at how poorly this AI problem was solved" is meaningless if the statement of the problem was itself poorly stated.
 

joeyd999

Joined Jun 6, 2011
6,282

joeyd999

Joined Jun 6, 2011
6,282
How long do you think before they have ads for free and lower tiers?
Who knows. But let me ask you this:

If it has value (which it does for me as I'm willing to pay for it), why should it be free (or ad free for zero/low cost)?

I'd rather be the customer than the product!

BTW, the next tier for Grok, "Super Grok", is $300 per month. I don't need the capability (yet), but if I can find a way to make it productive in my office (make my employees more effective), it's far cheaper than more payroll!
 

nsaspook

Joined Aug 27, 2009
16,325
Who knows. But let me ask you this:

If it has value (which it does for me as I'm willing to pay for it), why should it be free (or ad free for zero/low cost)?

I'd rather be the customer than the product!

BTW, the next tier for Grok, "Super Grok", is $300 per month. I don't need the capability (yet), but if I can find a way to make it productive in my office (make my employees more effective), it's far cheaper than more payroll!
I'm almost shocked it took this long but it's true in most domains, first you groom the addict (addiction can be a good thing with moderation) and then you groom the wallet.
 

WBahn

Joined Mar 31, 2012
32,844
How long do you think before they have ads for free and lower tiers?
I'm using Perplexity and while it doesn't have ads, yet, it is always pushing me to subscribe to their paid tier -- about every other query I have to click one or two places to avoid subscribing and see the response.
 

cmartinez

Joined Jan 17, 2007
8,762
It's only a matter of (a very short) time, before politicians start saying "I never said that! ... that's fake AI !"

And it won't matter if there were witnesses or physical records of any sort. People will simply go flat-earth on anything they like or dislike. Welcome to the Post-Truth Era of human history ... frightening indeed.


Look closer and you'll see the indications of an AI artist. Rose has no social media presence, has never played a gig, has no videos, and has released an improbable number of songs in a short space of time.

Between 28 September and 5 December, she uploaded at least 45 tracks to streaming services. Even Prince, an artist known for restless creative mania, would have struggled to match that figure.
 

nsaspook

Joined Aug 27, 2009
16,325
It's only a matter of (a very short) time, before politicians start saying "I never said that! ... that's fake AI !"

And it won't matter if there were witnesses or physical records of any sort. People will simply go flat-earth on anything they like or dislike. Welcome to the Post-Truth Era of human history ... frightening indeed.


What? There are people that blanket believe things?
https://toolbox.google.com/factcheck/explorer/search/list:recent;hl=en

We are still in the Pre-Truth-Era of human history. This, now easy method of making and eventually (making a consistent, past fake life trail, is still tricky) exposing fakers might get actually get us to the Truth-Era of human history.
 

nsaspook

Joined Aug 27, 2009
16,325
https://www.theguardian.com/technology/2026/jan/15/elon-musk-xai-datacenter-memphis
1768819591697.jpeg
A US regulator ruled on Thursday that Elon Musk’s artificial intelligence company had acted illegally by using dozens of methane gas turbines to power huge datacenters in Tennessee.

xAI has been fighting for a year and a half over truck-sized gas turbines the company had parked near its Colossus 1 and 2 facilities, arguing to local authorities that the electricity-generating turbines were exempt from requirements for air quality permits.
 

nsaspook

Joined Aug 27, 2009
16,325
https://futurism.com/artificial-intelligence/ai-failing-boost-productivity

AI Completely Failing to Boost Productivity, Says Top Analyst
"Where we are today, we're not seeing it."
Mountains of research as well as cases of workplace deployment of AI have suggested that the tech is far from being ready for primetime. One notable MIT study found that 95 percent of companies that integrated AI saw zero meaningful growth in revenue. For coding tasks, one of AI’s most widely hyped applications, another study showed that programmers who used AI coding tools actually became slower at their jobs.

AIs designed to automate entire tasks aren’t looking too hot either: when researchers at the Center for AI Safety tested AI agents’ ability to complete remote work tasks, not a single model was able to complete more than three percent of their assignments. The very introduction of AI into the workplace appears to be disastrous for employee relations, suggests another study, which found that AI resulted in employees passing off low quality “workslop” with the expectation that someone else down the line would polish the AI’s shoddy output.

“A lot of generative AI stuff isn’t really working,” Gownder told The Register. “And I’m not just talking about your consumer experience, which has its own gaps, but the MIT study that suggested that 95 percent of all generative AI projects are not yielding a tangible [profit and loss] benefit. So no actual [return on investment.]”
“Outsourcing is a very popular one,” he told The Register. “They’re firing people because of AI, and then three weeks later they hire a team in India because the labor is so much cheaper.”
 

nsaspook

Joined Aug 27, 2009
16,325
https://torrentfreak.com/nvidia-con...o-secure-access-to-millions-of-pirated-books/
‘NVIDIA Contacted Anna’s Archive’
Last Friday, the authors filed an amended complaint that significantly expands the scope of the lawsuit. In addition to adding more books, authors, and AI models, it also includes broader “shadow library” claims and allegations.

The authors, including Abdi Nazemian, now cite various internal Nvidia emails and documents, suggesting that the company willingly downloaded millions of copyrighted books.

The new complaint alleges that “competitive pressures drove NVIDIA to piracy”, which allegedly included collaborating with the controversial Anna’s Archive library.
allegedly
 
Top