THE AI Thread

[CYDJ]

Disclaimer: This isn’t intended to be combative in tone. If it comes off that way, I’m sorry.

Weirdly I think we probably agree on the potential scale of impact but disagree on much of the rest of it.

I’ll start with most trivial, pedantic disagreement. I think the notion that AI is incomprehensible and opaque is greatly overblown.

Not to be overly facile, but virtually everything in existence is incomprehensible to humans at some level of abstraction.

With AI I think this belief stems from our inability to describe the behavior of the system as we would using a transfer function in a linear system.

Sure, the output is both emergent and non-deterministic, but the process of how it’s created and a conceptual description of what it is doing are embarrassingly simple at the fundamental level (ANNs). It is also basically unchanged over the last 40 years (and the most basic element was invented ~70 years ago!).

While I agree governance and control are the foremost concerns, I don’t think this has anything to do with AI itself

The structural problem is with how we’ve organized our civilization. The concentration of power that already exists is what is causing the abdication by governments of their duty to provide oversight. AI didn’t create that, we all allowed that to happen.

My hunch is that access to data centers needed to keep up in the arms race to build the biggest models will continue to be monetized, not restricted. Those massive models will either become economically viable or they’ll shrink to fit the power budgets to make them profitable enough. I’d bet on a hybrid model of training in the cloud and running inference at the edge becoming the most popular paradigm.

The damage I’m most worried about is due less to control, but to widespread access.

From what I gather the “agentic” AI you’re all talking about currently is integration of LLMs with business tooling wrapped in somewhat bespoke programmatic “agents” to provide domain-specific executive function. Does that sound right?

Putting that firepower in the hands of roving gangs of CFO warlords may well cause economic genocide on a massive scale at precisely a time when social safety nets and cohesion are weakened.

Or maybe it’ll trigger a social remodeling that rebalances power in a more equitable manner?

I think I agree with just about everything here. Except the bolded.

What I hear when I talk to my "smart friends" and listen to those that have been in this area of study for a significant chunk of the true application of ANNs in modern AI (Geoffrey Hinton, Yann Lecun, Et al.) is that we definitely know HOW to make a statistical model and the infrastructure necessary to feed and power it. At the same time they are saying they definitely DON'T KNOW HOW the model chooses the connections necessary to produce the work that is, in an ever increasing fashion, impressing and baffling their creators.

They can not explain exactly how it does everything it does. That is not a fact that makes me think that we have no worries at all.

Thank you for the complete thoughts though. They were very enlightening.

 

[CYDJ]

Besides the em dashes? The tone and structure felt like something generic I’ve read 100 times via ai.

GREAT, what did you agree with or disagree with?

I'm not sure the point is if you can identify AI editing structure. That is not the point.

I guess if you wrote per The Elements of Style I could see that and call it out too. But, what good does that do?

If you feel what I said was slop, please elaborate. That, at least, would be joining the conversation.

 

[CYDJ]

I'm sorry that I spammed this evening. Spent some time away from the computer today and had a few thoughts I wanted to express to certain posters (which were all respectful, which is nice to see here.)

I understand we have differences of opinion. And I am glad we all still have those rights.

Have a great evening.

 

[simply1]

GREAT, what did you agree with or disagree with?

I'm not sure the point is if you can identify AI editing structure. That is not the point.

I guess if you wrote per The Elements of Style I could see that and call it out too. But, what good does that do?

If you feel what I said was slop, please elaborate. That, at least, would be joining the conversation.

I mean I could write 3 sentences and tell AI to come to with more and call it my own. Give it your post. Give a couple sentences. Paste 5 paragraphs and say I’m looking for conversation.
When it’s that easy to tell AI is writing it, it’s hard to engage with the person who may or may not have anything to do with the thoughts coming out. You can say it all you want. What I read is AI and thus not interesting.

 

[xr4ticlone]

From what I have seen so far, AI has brought way more harm than good. I used to trust a couple of reputable neutral news sources to see what is going on in the world, but I seriously do not trust anything anymore. All content (text, pictures, videos) is hard to decipher what is real. Couple this with social media and we have a real disaster brewing. Entities creating fake/bogus content to stir emotions, pit people against one another, and get clicks so they get $'s in their pocket. Quite sad really.

That’s the point IMHO.

It’s not new. It’s a common theme in communist nations.

When everything is a lie, there is no truth.

If you’ve ever meet anyone from the former Soviet Union, they lie about even the most mundane things. You see it in China as well.

Having seen the end results first hand…we should all be very concerned.

 

[xr4ticlone]

This is different on my opinion. The corporate goal is to eliminate the 'labor tax' of people. (Their words not mine) This is nearly all industry too. Not just one or two.

Employees are the largest expense for almost every business…top 3 for almost all.

Not sure who buys things when 70% of the workforce is watching AI Jerry Springer on their gov cellphones all day.

 

[xr4ticlone]

It could also be argued that it will provide a tool to make radiologists, cardiologists, scribe, psychologists, psychiatrists more accurate and faster and maybe replace them with something more accurate down the road. Wouldn't that be good for you if you are the patient???

I get it, this is hitting close to you for some reason. Yes it’s going to change things, but maybe some of the changes will be for the better too.

Imagine a psychologist that you could talk to whenever that didn’t just talk about their own bull ****…or try and drag treatment out for job security.

*I’ve only been to one for ADHD meds…which was a $150 for 3 minutes of their time every 3 months to get refills. But have talked to many who went or still go about their time.

 

[xr4ticlone]

So depending how quickly and dramatically the entire structure of society changes both remodel and handyman type work could be safe.

If there is a market big enough to sustain work for everyone displaced from new construction then I imagine it is a market big enough to warrant investment.

Maybe that investment results in imaging tools that can account for tiny variances in plumb and square, etc to get extremely precise, complete models (maybe even from a phone). Those models are used to generate kits with many precut (or preprinted) pieces, step-by-step instructions, and even guides/patterns based designs created through chat bots.

Maybe the investment goes the other way to use draconian code and zoning laws under the guise of safety, environmental, etc improvements to make new builds with manufactured/composite printing/etc construction more cost effective than maintaining old stock for the corporate owners that gobble up most of the real estate to rent back to the serfs?

And also, just building fairly affordable, capable carpentry robots isn’t all that far off.

Under the incentives of our current economic system anything worth paying a living wage to enough people to do is worth

All of this assumes anyone will have money to pay for these services.

 

[12191987]

I think I agree with just about everything here. Except the bolded.

What I hear when I talk to my "smart friends" and listen to those that have been in this area of study for a significant chunk of the true application of ANNs in modern AI (Geoffrey Hinton, Yann Lecun, Et al.) is that we definitely know HOW to make a statistical model and the infrastructure necessary to feed and power it. At the same time they are saying they definitely DON'T KNOW HOW the model chooses the connections necessary to produce the work that is, in an ever increasing fashion, impressing and baffling their creators.

They can not explain exactly how it does everything it does. That is not a fact that makes me think that we have no worries at all.

Thank you for the complete thoughts though. They were very enlightening.

No, we agree here. That’s what I was trying to get across (poorly).

Unlike say, a traditional linear system, we can’t describe accurately and precisely what it is doing.

Nobody can explain how features are spread through the latent space.

However, it is possible to explain gradient descent and back propagation so that it is intuitively understandable to a kid who gets how a dot product between two vectors gives a correlation.

It isn’t too unwieldy to describe how you might do that with a fully-connected, two-layer ANN to classify simple images of cats using labeled data.

We can say that through a massive amount of trial-and-error all those weights and biases have encoded a set of features that describe a pic of a cat. We just don’t exactly know what/where/how about the feature encoding. We could probably speculate about certain shapes, textures, etc that might be encoded, but we can’t know.

I think that is a believable enough story for the human mind.

Now, as the complexity of the model grows the emergent features from that incomprehensible what/when/how of the encoding themselves become inconceivable.

So yeah, it is essentially magic at a point.

To be clear, I am not built to just accept this scale of ignorance about anything, particularly a human invention. But here we are.

Honestly, consuming as much as I can about neuroscience and theories for a materialist origin of consciousness (as well as random **** from the Santa Fe Institute) has helped with just accepting it.

Random AI celebrity flex: Andrew Ng and I were classmates. We probably had 1-2 classes together most semesters over 4 years. I don’t believe we ever spoke a word to each other. He is a super smart dude.

 

[12191987]

All of this assumes anyone will have money to pay for these services.

100% agree. Our social/economic structure does not seem suited for the moment at hand.

**** could get really, really bad.

 

[NATEizKING]

NASA’s Perseverance rover completes the first AI-planned drive on Mars

NASA’s Perseverance rover has just made history by driving across Mars using routes planned by artificial intelligence instead of human operators. A vision-capable AI analyzed the same images and terrain data normally used by rover planners, identified hazards like rocks and sand ripples, and...

www.sciencedaily.com

 

[CYDJ]

I mean I could write 3 sentences and tell AI to come to with more and call it my own. Give it your post. Give a couple sentences. Paste 5 paragraphs and say I’m looking for conversation.
When it’s that easy to tell AI is writing it, it’s hard to engage with the person who may or may not have anything to do with the thoughts coming out. You can say it all you want. What I read is AI and thus not interesting.

Gotcha.

I don't think that will bode well for anyone in the future. But, to each their own.

 

[GBlade]

Just curious, how are you determining “quicker”? There is an open source study that showed developers thought they were saving time and really weren’t.

Measured by the number of tickets I finish. Not scientific. I definitely agreed with that study when it came out. I was re-writing most of the AI output so savings were a wash or a net sink, but the ability for agents to call into bash tools has really changed things since they can self validate now.

Out of curiosity, which tools have you used?

My work provides GitHub Copilot Enterprise and Cursor Enterprise subscriptions as model providers. Cursor is good and a bit less friction then using VSCode (very similar though). The new CLI tools are what I was referring to though. Claude Code is the most famous, but there is Copilot CLI and Codex CLI too. Work has not approved those though, so I am using OpenCode which has integrations with many providers. What's nice about a CLI app is it's a lot easier to just open the Terminal and go then importing a project into an IDE. Allows some crazy workflows that I'm just scratching the surface of if Twitter is any indication.

My main workflow is to open the project in Plan mode and prompt it with a detailed feature specification. But I tell it to only write the tests for the specification and not the code. I review the tests to make sure they capture my requirements and only then do I have it proceed with writing the code. There is a rule in the context to tell it to run the tests to validate. It will loop until the tests pass and then it's a matter of making sure it didn't go off the rails and cleaning up anything that was ambiguous or my preferences.

 

[cycloneretro]

Nvidia shares are down after a report that its OpenAI investment stalled. Here's what's happening

Nvidia's Jensen Huang criticized OpenAI's business strategy as the chip giant plans to invest up to $100 billion into the AI startup, per reports.

www.cnbc.com

I don’t question the value AI brings to those that can use it as a tool along with their domain knowledge. What I question is the business model sustainability. OpeAI Isn’t making any money. None of these companies are. Businesses aren’t seeing returns that they were promised.

 

[iowastatefan1929]

Nvidia shares are down after a report that its OpenAI investment stalled. Here's what's happening

Nvidia's Jensen Huang criticized OpenAI's business strategy as the chip giant plans to invest up to $100 billion into the AI startup, per reports.

www.cnbc.com

I don’t question the value AI brings to those that can use it as a tool along with their domain knowledge. What I question is the business model sustainability. OpeAI Isn’t making any money. None of these companies are. Businesses aren’t seeing returns that they were promised.

I already got lured over to Google Gemini, it was a bit cheaper and I think its just as good if not better. They will be giving it away for free soon.

 

[cyclone83]

https://www.google.com/url?sa=t&sou...QIFRAB&sqi=2&usg=AOvVaw3ZPhqZrGmyMQO1djzwzhXU Are the tech bros who had front row seats at Trumps inauguration the priests of the temples of Syrinx? The late great Neal Peart wrote this in 1975. However he was off a few years, it's not quite 2112 yet.

 

[mramseyISU]

If this is real then it's kind of concerning I think.

 

[FriendlySpartan]

If this is real then it's kind of concerning I think.

This is insanely misleading (it’s from FB so of course it is) Moltbook was created by a human scientist to be a social media site for AI chatbots and agents to see how they interact only with each other.

 

[mramseyISU]

This is insanely misleading (it’s from FB so of course it is) Moltbook was created by a human scientist to be a social media site for AI chatbots and agents to see how they interact only with each other.

Yeah I'm usually skeptical of those sensational headlines, this the first time I'd heard of Moltbook so I wasn't really sure what it is. The only place where stuff is more misleading than Facebook is Linked In.

 

[HoraceGrant]

The accounting firm I work for is going to sell to private equity in the next year, so they can make large capital investments into AI (accounting firms don't traditionally have a ton of working capital). We currently use an AI program for tax research and it is pretty dang good (although it does make mistakes). We also currently use CoPilot for kind of every day things which I kind of feel like sucks and has almost no practical applications yet besides some help in excel. The push from the top is getting pretty strong "you must embrace AI and you must promote AI". Weird times - I am mixed on it overall.