Artificial Intelligence In Our Lives

[turbineseaplane]

We don’t need the world to move faster. We don’t need society to move faster. We don’t need to “progress”

No kidding. When people advocate for that, I don’t even know what they think we are racing towards?

There is no race to win. There is no end point.

 

[BSDnostalgia]

According to data, checking AI’s work is slower than just doing it yourself for most things, like code for example

Interesting. Do you have one or more sources for that? In these forums, we often see people stating they get significantly improved productivity using Claude tools, but, who has actually done full project comparisons?

 

[MicroLab]

According to data, checking AI’s work is slower than just doing it yourself for most things, like code for example

Not if you have absolutely no idea what you’re doing (like me)

 

[turbineseaplane]

Not if you have absolutely no idea what you’re doing (like me)

Doesn’t that also mean you’re gonna have no idea if what it’s doing is correct or has other issues?

 

[turbineseaplane]

Seeing more and more stuff like this.
Not good.

 

[Howard2k]

Not if you have absolutely no idea what you’re doing (like me)

It does depend on the task too. I had a repetitive task I needed to complete over and over, and it was a manual and burdensome task.

I had AI write a macro to do the work for me, then I had AI write another macro do double check the work of the first macro.


I may be wrong, but I think the overwhelming portion of the negativity on AI comes from people who don't know how to use AI. Not all, but a significant amount.

Just like the datacentre noise.

Driven by ignorance and conspiracy.

 

[Flowstates]

I use AI for most of my professional function and have a pretty good grip on the science behind it (Academically).

The last 8 months have been a complete shift in capability, with an extremely big caveat that there needs to be formal verification to the result in order to make the results actually useable (whether that formal verification is part of the training process, or during "Test time compute" which is when the model reasons and bounces on it's own thread).

This is the main reason that we still don't see the exponential that we see with Programming elsewhere is that some fields are innately harder to create a formal verification system for.

The models have gained orders of magnitudes of capability (that is if you agree as capability in autonomy to affect change in a given bounded context measured by autonomous time with no outside input and output quality.) Yet the still fail in ridiculous manners, eg. Hallucinations (that are driven by its inability to understand which set of information to retrieve the optimal gradient from, eg. external tool use vs internal model knowledge).

The real productivity for myself has skyrocketed, as did the confidence and ability to tackle bigger problems.

The flip side of this faustian bargain has been a growing distrust for any contemporary unit of information published in a non-trusted environnement. An inability to forsee and apprehend the future more than 6 months out. Intense dread at the scale of incoming societal disruption and growing pessimism into our ability to process it as an empathic society.

With the deepest cut being the downright demotion to pointlessness of certain pursuits; writing, visual arts and music have become unsustainable as anything else than dilettante musing.

And I for once fear for the spirit of community when we all become intermediated by stochastic artifacts, our embodied sensitivities displaced by gradients ran by corporations for profit.

 

[turbineseaplane]

The real productivity for myself has skyrocketed

I guess I would ask if this is ultimately, actually, a net good in any way beyond "metric go up"?

Lately (in recent years) I'm feeling like everyone is trying to run faster and faster in some race, with no finish line or even intermediate goal.

It just seems to be "lets boil the planet, our lives, our sanity, our work/life balance, to do MOAR".

Why?
"Who knows ... just MOAR!"

 

[turbineseaplane]

So much of this tech enables such awfulness.
It's really sad to watch

How Deepfakes Tore a High School Apart

After five teen girls were targeted by AI-generated child sexual abuse material, Radnor Township High School in Pennsylvania has become a case study in how schools and police around the country grapple with how to response to deepfake crimes involving children.

www.404media.co

 

[Flowstates]

I guess I would ask if this is ultimately, actually, a net good in any way beyond "metric go up"?

Excellent question.

If you mean productivity as unit output by unit time, yes definitely. I just finished the work that was piloted by a team of contractors for 6 months alone in 3 (that being said, the scenario was in the goldilocks of what was optimal for models to manipulate).

Nevertheless, it took some experimentation to get there and verification is still a real cost given that current systems still collapse under complexity.

Now to address the second part of your statement, assuming a new equilibrium where everyone has access to such tools and you are assuming similar ability to yours from your competitors. No inherent moat (lasting business advantage in a market) can be created from the use of the technology itself.

You either create a lasting business advantage from being enabled in a tertiary function from the systems or cutting overhead. Most leadership is not capable enough to successfully pivot to achieve the first, hence why we are seeing widespread overhead cut.

Unless we start seeing material advances in materials sciences from the technology (which has transformative effects in healthcare, construction and engineering that would unlock more than just automation. As in the dichotomy between automation and transformative change (which leads to a net benefit).

We just added ressource overhead for power centralization and surveillance with no obvious quantitative benefit to humans (That is in the mass of productivity added to society under the assumption that the work that was then done by a thread of human intelligence is now being done by a computational thread with a radically different consumption and externality profile).

The answers to those issues are political, IP attribution, externalities mitigation (Pollution, Displacement, Social Change, Climate Change).

Yet we have just shifted into an era where the political for the masses has become pantomime, hence the pessimism.

 

[Analog Kid]

If anyone is in this thread complaining about AI is also in another thread complaining that Apple doesn't add features or fix bugs fast enough, it's worth stepping back and reconciling your belief systems.

 

[Luftkopf]

Excellent question.

If you mean productivity as unit output by unit time, yes definitely. I just finished the work that was piloted by a team of contractors for 6 months alone in 3 (that being said, the scenario was in the goldilocks of what was optimal for models to manipulate).

Nevertheless, it took some experimentation to get there and verification is still a real cost given that current systems still collapse under complexity.

Now to address the second part of your statement, assuming a new equilibrium where everyone has access to such tools and you are assuming similar ability to yours from your competitors. No inherent moat (lasting business advantage in a market) can be created from the use of the technology itself.

You either create a lasting business advantage from being enabled in a tertiary function from the systems or cutting overhead. Most leadership is not capable enough to successfully pivot to achieve the first, hence why we are seeing widespread overhead cut.

Unless we start seeing material advances in materials sciences from the technology (which has transformative effects in healthcare, construction and engineering that would unlock more than just automation. As in the dichotomy between automation and transformative change (which leads to a net benefit).

We just added ressource overhead for power centralization and surveillance with no obvious quantitative benefit to humans (That is in the mass of productivity added to society under the assumption that the work that was then done by a thread of human intelligence is now being done by a computational thread with a radically different consumption and externality profile).

The answers to those issues are political, IP attribution, externalities mitigation (Pollution, Displacement, Social Change, Climate Change).

Yet we have just shifted into an era where the political for the masses has become pantomime, hence the pessimism.

I'm not asking to be dismissive, I'm just curious. Which LLM do you use to refine your posts?

 

[Flowstates]

I'm not asking to be dismissive, I'm just curious. Which LLM do you use to refine your posts?

That's good ol me.

Although I will concede that the "Excellent Question" open feels icky, it would seem that I have started to absorb the sycophancy.

 

[turbineseaplane]

I mean, it is a little concerning to think about where this leads us..

Lawyers know AI can hallucinate. Judges have warned them. Courts have sanctioned them for it. They keep citing fake AI cases anyway.

The trend of attorneys getting caught citing AI-hallucinated cases points to a broader problem: instead of checking AI’s work, people keep trusting it

www.scientificamerican.com

 

[Luftkopf]

That's good ol me.

Although I will concede that the "Excellent Question" open feels icky, it would seem that I have started to absorb the sycophancy.

I guess it was bound to happen that people start writing in the style of LLM output. Now I don’t trust my own internal AI radar!

Thanks for answering my question. 🙂

 

[BSDnostalgia]

Excellent question.

If you mean productivity as unit output by unit time, yes definitely. I just finished the work that was piloted by a team of contractors for 6 months alone in 3 (that being said, the scenario was in the goldilocks of what was optimal for models to manipulate).

Nevertheless, it took some experimentation to get there and verification is still a real cost given that current systems still collapse under complexity.

Pardon my ignorance, but, I've never used AI for coding, and, I have no idea how this works in a real project.

In a "traditional" large software development process, normally, actually writing the code is anywhere from 50% down to as low as 10% of the total effort. Project work, defining and dividing the work, defining and refining interfaces, defining the verification tests functionally, (coding), code reviews, writing test code and setting up the verification tests, documenting, etc., are the other 50%-90%.

All we hear about is AI writing code, but, how much of the other stuff can AI do? It sounds like AI can do the "fun" 10% and leave people with the tough stuff-- tough, like, long meetings stuff. But, I guess, "vibe" coding obviates the need for meetings? Are there any actual "scientific" tests comparing the processes?

 

[turbineseaplane]

Maybe in the interests of "good business", among other benefits, we should use human employees?

 

[Mainsail]

Pardon my ignorance, but, I've never used AI for coding, and, I have no idea how this works in a real project.

In a "traditional" large software development process, normally, actually writing the code is anywhere from 50% down to as low as 10% of the total effort. Project work, defining and dividing the work, defining and refining interfaces, defining the verification tests functionally, (coding), code reviews, writing test code and setting up the verification tests, documenting, etc., are the other 50%-90%.

All we hear about is AI writing code, but, how much of the other stuff can AI do? It sounds like AI can do the "fun" 10% and leave people with the tough stuff-- tough, like, long meetings stuff. But, I guess, "vibe" coding obviates the need for meetings? Are there any actual "scientific" tests comparing the processes?

This is true of many white collar planning and analysis jobs. The up front work defining the goals and purpose and major assumptions are the most important and difficult part of the project. Most of my limited experience with AI has required that I feed that info into the model first before it generates much of anything useful. Granted, as a retired guy, my use is pretty limited, so maybe others have had other experiences .

 

[nnoble]

AI undoubtedly has the ability to improve our lives, even more so if it were better regulated. I am proud of the stand taken by mayor of London Sir Sadiq Khan against Palantir. We need AI but we do not want something that has the potential to be weaponised by hostile powers and dubiously led companies like Palantir.

 

[Flowstates]

Pardon my ignorance, but, I've never used AI for coding, and, I have no idea how this works in a real project.

In a "traditional" large software development process, normally, actually writing the code is anywhere from 50% down to as low as 10% of the total effort. Project work, defining and dividing the work, defining and refining interfaces, defining the verification tests functionally, (coding), code reviews, writing test code and setting up the verification tests, documenting, etc., are the other 50%-90%.

All we hear about is AI writing code, but, how much of the other stuff can AI do? It sounds like AI can do the "fun" 10% and leave people with the tough stuff-- tough, like, long meetings stuff. But, I guess, "vibe" coding obviates the need for meetings? Are there any actual "scientific" tests comparing the processes?

Echoing what has been said above, planning and requirements definition is still the lion's share of time spent.

The radical shift has been that AI has been handling the verification test definition, test code, verification test and documentation by itself.

We also rarely delegate tasks anymore and have a much higher ownership surface, what get's divvied up is high level ownership surface over a product / domain. Which I assume is the way successful orgs are going to restructure but bodes badly for low level (in the sense of reactive, not proactive) workers.

 

[Luftkopf]

Echoing what has been said above, planning and requirements definition is still the lion's share of time spent.

The radical shift has been that AI has been handling the verification test definition, test code, verification test and documentation by itself.

We also rarely delegate tasks anymore and have a much higher ownership surface, what get's divvied up is high level ownership surface over a product / domain. Which I assume is the way successful orgs are going to restructure but bodes badly for low level (in the sense of reactive, not proactive) workers.

It sounds like there won't be room for junior devs to come through any more. I think this is adding to the debt being built up. Things are going to be rough in ten years.

 

[Analog Kid]

It sounds like there won't be room for junior devs to come through any more.

I’ve seen studies that suggest there is less demand for junior devs, and “you need experience to get experience” isn’t something that started with the advent of AI. but I think it’s worth recognizing specifically what point @Flowstates is making:

low level (in the sense of reactive, not proactive)

The distinction being made is high level (big picture, lean forward, creative, willing to take risk and ownership) vs low level (transactional, small work batch, measure by lines of code written) personalities. This doesn’t directly map to junior versus senior. I’d argue there are some junior devs who have been held back by management that forces high functioning junior employees into low level roles. And senior employees have been lost early because they’re viewed as expensive per line of code.

What @Flowstates is correctly identifying, in my opinion, is that the nature of the gig is changing. This is going to mean junior employees are going to need different skills. I think the demand gap right now is because we aren’t training new grads with the right skills for this new era yet.

 

[Luftkopf]

I’ve seen studies that suggest there is less demand for junior devs, and “you need experience to get experience” isn’t something that started with the advent of AI. but I think it’s worth recognizing specifically what point @Flowstates is making:



The distinction being made is high level (big picture, lean forward, creative, willing to take risk and ownership) vs low level (transactional, small work batch, measure by lines of code written) personalities. This doesn’t directly map to junior versus senior. I’d argue there are some junior devs who have been held back by management that forces high functioning junior employees into low level roles. And senior employees have been lost early because they’re viewed as expensive per line of code.

What @Flowstates is correctly identifying, in my opinion, is that the nature of the gig is changing. This is going to mean junior employees are going to need different skills. I think the demand gap right now is because we aren’t training new grads with the right skills for this new era yet.

Either way, it's going to be disruptive, and we're all in for a lot of pain before things even out.

 

[throAU]

Sometimes, when I read all this, I get the feeling that people are listing the pros and cons
and passing judgment without even looking at what AI can actually do for people today !

Many people, even in industrialized nations, will hardly be able to use AI effectively because
the average person is unlikely to ask the right, detailed questions to get the (thoughtful)
answers that will help them move forward in life. So it won’t raise personal educational standards.

If people are thinking of AI as a basic chatbot and thinking "hurrr, it got the answer to this trivia question wrong! AI is dumb!" they're so far behind the reality of what is happening they may as well be living in the 1980s.

 

[throAU]

That's a good point, I've installed Linux (an Arch derivative ) and I've run into some issues, here and there. googling was fruitless, unless I wanted to watch a 20 minute video which may or may not help me. I got the exact steps from claude.

I've been experimenting with openclaw in a VM.

Its really interesting having the machine able to configure and diagnose itself by just talking to it.

Also interesting to get it to give me a daily briefing based on criteria I ask it to keep an eye on, on the internet.

I told it (Bob) to generate an email template in the style of the OpenClaw website as it saw fit.