Jump to content

Recommended Posts

Posted

Hung out with some old friends over Thanksgiving that I haven't seen in a while, and it's amazing how almost all of them simp for AI. I think they were surprised by my pushback, but I'd guess it's only because so few people around them have a strong and vocal anti-AI stance.

One is a surgeon and says he uses AI at work to help him. Great! Using AI as an extra tool for healthcare sounds good to me. But in general, they're just completely amenable with ChatGPT-ing everything and haven't spent a second thinking about how it is horrible for arts/music/film. All of them also have kids in high school or younger, and can't see the forest for the trees with how it will negatively affect them.

Likening it to Mad Cow disease  is on point. I'm not hopeful, but how AI is currently being used/marketed is a lame party trick and people need to wake up quickly. 

 

  • Hook 'Em 3
  • Fuck Around and Find Out 1
Posted
5 minutes ago, Hank Kingsley said:

Click the link in post above mine. I'm not your ChatGPT. 

 

 

no offense intended.  didn't realize the link went to an actual article.

I didn't get the "feeding on itself" context of the comment and was thinking more about the images of cows unable to walk and falling down that played on a loop anytime Mad Cow disease was discussed in MSM.

  • Hook 'Em 1
Posted
35 minutes ago, Incredulity said:

no offense intended.  didn't realize the link went to an actual article.

I didn't get the "feeding on itself" context of the comment and was thinking more about the images of cows unable to walk and falling down that played on a loop anytime Mad Cow disease was discussed in MSM.

[context: I am literally a certified AI/ML specialist and builder and have taken enough high-order math at UT to at least accurately conceptualize what's happening in the black box]

The end state of "unable to walk and falling down" is effectively the end-state of a "Hapsburg" model. When you train a model on synthetic data produced by a similar model, you get the same concentration of error effect that is observed in genetics whenever inbreeding happens in a population. 

In the current and active development state of AI, these LLM companies like Anthropic and Gemini and OpenAI are scraping the internet for ANY data they can, and they don't really care about the provenence of its origin. As a result, the "for-profit" LLM's are getting trained on a heavily poisoned and synthetic dataset that contains a bunch of data that will embed defects into the model and train it to produce worse errors. Not to mention the general problem with model poisoning that Anthropic published a paper on a month or two ago

  • Hook 'Em 6
Posted
4 minutes ago, Captainant said:

[context: I am literally a certified AI/ML specialist and builder and have taken enough high-order math at UT to at least accurately conceptualize what's happening in the black box]

The end state of "unable to walk and falling down" is effectively the end-state of a "Hapsburg" model. When you train a model on synthetic data produced by a similar model, you get the same concentration of error effect that is observed in genetics whenever inbreeding happens in a population. 

In the current and active development state of AI, these LLM companies like Anthropic and Gemini and OpenAI are scraping the internet for ANY data they can, and they don't really care about the provenence of its origin. As a result, the "for-profit" LLM's are getting trained on a heavily poisoned and synthetic dataset that contains a bunch of data that will embed defects into the model and train it to produce worse errors. Not to mention the general problem with model poisoning that Anthropic published a paper on a month or two ago

tl;dr

AI haemophilia

Posted
5 minutes ago, Captainant said:

I am literally a certified AI/ML specialist

Have you considered a certificate in disaster science? I hear Harvard Extension School has a program.

  • Like 1
  • Haha 1
Posted
Just now, Bozo_Casanova said:

Have you considered a certificate in disaster science? I hear Harvard Extension School has a program.

I was doing this BEFORE it was genAI nonsense, I swear! Most of my time is spent convincing customers they don't need an LLM and traditional ML modeling is more than sufficient to meet their business problem - AND they can still tell their business they're """using AI"""

It's frustrating because there are genuinely useful genAI use cases out there, but there's a much higher volume of snake oil being bought and sold. Not to mention the army of idiot middle managers that don't understand the jobs they manage, so they think they can "just replace everyone with AI!"

  • Hook 'Em 5
  • Like 2
Posted (edited)
2 minutes ago, Captainant said:

Not to mention the army of idiot middle managers that don't understand the jobs they manage, so they think they can "just replace everyone with AI!"

This is exactly what happened with my former job when we brought in a new manager who had no idea what we actually did.  Not so much replace the employees with AI, but have all the current employees rely on AI to start doing their jobs for them instead of using our brains.  When I pushed back against that, it was all over for me.

Edited by Biff Tannen
  • Rage+1 4
Posted
1 hour ago, Hank Kingsley said:

Hung out with some old friends over Thanksgiving that I haven't seen in a while, and it's amazing how almost all of them simp for AI. I think they were surprised by my pushback, but I'd guess it's only because so few people around them have a strong and vocal anti-AI stance.

One is a surgeon and says he uses AI at work to help him. Great! Using AI as an extra tool for healthcare sounds good to me. But in general, they're just completely amenable with ChatGPT-ing everything and haven't spent a second thinking about how it is horrible for arts/music/film. All of them also have kids in high school or younger, and can't see the forest for the trees with how it will negatively affect them.

Likening it to Mad Cow disease  is on point. I'm not hopeful, but how AI is currently being used/marketed is a lame party trick and people need to wake up quickly. 

 

I think this just highlights the larger issue with how most people rarely see the bigger picture with anything. They care about right now and how it directly impacts them at this very moment. If it's convenient for me, I want it.

  • Hook 'Em 3
  • Like 1
Posted
14 minutes ago, Okie State said:

I think this just highlights the larger issue with how most people rarely see the bigger picture with anything. They care about right now and how it directly impacts them at this very moment. If it's convenient for me, I want it.

thats-the-sad-truth-thats-the-painful-tr

  • Hook 'Em 1
Posted

I feel like an outlier. Claude.ai helps me write oracle stored procedures, job schedules, triggers in seconds. I do have to make corrections because it doesn’t know our db schema that well. But it will generate the syntax for artifacts I’ve never created before. It’s made me >10x more productive. I come up with the “what do I need”, explain it to claude, and claude does 90% of the work.

It’s also helpful to interpret 1000 lines of audit logs and summarize it.

Shit. I’m on a system integration issue call right now with a parcel vendor and enterprise system vendor where we have too many end point connections to the parcel vendor system and it’s causing errors in the enterprise system’s manifesting. Claude looked at a screenshot of the connections and diagnosed the issue.

  • Hook 'Em 1
Posted
2 hours ago, Captainant said:

I was doing this BEFORE it was genAI nonsense, I swear! Most of my time is spent convincing customers they don't need an LLM and traditional ML modeling is more than sufficient to meet their business problem - AND they can still tell their business they're """using AI"""

It's frustrating because there are genuinely useful genAI use cases out there, but there's a much higher volume of snake oil being bought and sold. Not to mention the army of idiot middle managers that don't understand the jobs they manage, so they think they can "just replace everyone with AI!"

Have you read "AI Snake Oil" by Kapoor and Narayanan?  

16 minutes ago, GenXer said:

I feel like an outlier. Claude.ai helps me write oracle stored procedures, job schedules, triggers in seconds. I do have to make corrections because it doesn’t know our db schema that well. But it will generate the syntax for artifacts I’ve never created before. It’s made me >10x more productive. I come up with the “what do I need”, explain it to claude, and claude does 90% of the work.

It’s also helpful to interpret 1000 lines of audit logs and summarize it.

Shit. I’m on a system integration issue call right now with a parcel vendor and enterprise system vendor where we have too many end point connections to the parcel vendor system and it’s causing errors in the enterprise system’s manifesting. Claude looked at a screenshot of the connections and diagnosed the issue.

I think programming is one area where AI is very useful.  You have a ton of data to learn from and an easy way to track result vs input.  

  • Like 1
Posted
27 minutes ago, GenXer said:

I feel like an outlier. Claude.ai helps me write oracle stored procedures, job schedules, triggers in seconds. I do have to make corrections because it doesn’t know our db schema that well. But it will generate the syntax for artifacts I’ve never created before. It’s made me >10x more productive. I come up with the “what do I need”, explain it to claude, and claude does 90% of the work.

It’s also helpful to interpret 1000 lines of audit logs and summarize it.

Shit. I’m on a system integration issue call right now with a parcel vendor and enterprise system vendor where we have too many end point connections to the parcel vendor system and it’s causing errors in the enterprise system’s manifesting. Claude looked at a screenshot of the connections and diagnosed the issue.

I don't think anyone's saying that AI, as currently constituted, doesn't provide a beneficial service to most industries, in that it increases efficiency and allows a worker to accomplish exactly what you've described above.

The issue is the over reliance that society is placing on AI, industries that seem to be forcing AI onto their customers without a real need, the environmental and energy issues, and the future unknowns regarding capability and ability to control and regulate AI once it really takes off.

That's not even getting into the human jobs that will be eliminated. Like, I don't know, jobs that already allow AI to do 90% of the work...

  • Hook 'Em 3
  • Like 1
  • Fuck Around and Find Out 1
Posted
9 minutes ago, Tuco said:

Have you read "AI Snake Oil" by Kapoor and Narayanan?  

Nope I haven't 

29 minutes ago, GenXer said:

I feel like an outlier. Claude.ai helps me write oracle stored procedures, job schedules, triggers in seconds. I do have to make corrections because it doesn’t know our db schema that well. But it will generate the syntax for artifacts I’ve never created before. It’s made me >10x more productive. I come up with the “what do I need”, explain it to claude, and claude does 90% of the work.

It’s also helpful to interpret 1000 lines of audit logs and summarize it.

Shit. I’m on a system integration issue call right now with a parcel vendor and enterprise system vendor where we have too many end point connections to the parcel vendor system and it’s causing errors in the enterprise system’s manifesting. Claude looked at a screenshot of the connections and diagnosed the issue.

This is an actual use case for LLMs. All the cloud providers have pretty handy built in agents now that do a similar function. Can't think of the right cloud trail query? No problem, just describe it and they'll generate the query and drop it into the editor box for you.

I use Cline with a locally hosted LLM to accelerate local development and it's been very handy for covering unit test gaps,and building new projects from source that have complex dependencies. It's great for reading through dense logs and crap that I don't want to look through but need a morsel of information out of. 

Posted

I'm a noted bear on *this* particular flavor of AI and this AI bubble (while noting that AI is a transformational technology which will change everything we have known about labor, work, social, etc. in the future).

That said I was listening to one of the world's prominent quants talking about pod shops and hedge funds and we all know they are super advanced on all things technology to make a buck, and he made the point in a way that I've been thinking about it: today AI is not ready to take over. It's only synthesizing words and videos and text and images that man has already created for it to process (upon processing). But one day, in a few years even, maybe? WIth the advent of the sensory stuff (Meta glasses) and other stuff that will come (touch, taste, tactiles, etc.). The complexity of human sensing and consuming will get closer to what a machine can also consume and sense.

It will be an AI version of the meme "The Yellowstone Park Ranger has noted there is a large overlap between the smartest bear and the dumbest human, upon trying to create a bear-proof trash can".

The overlap between the dumbest person and the smartest machine COULD be a critical mass. And then maybe there is a tipping point.

And then what to do about the useless eaters?

21 minutes ago, Foosters said:

I don't think anyone's saying that AI, as currently constituted, doesn't provide a beneficial service to most industries, in that it increases efficiency and allows a worker to accomplish exactly what you've described above.

The issue is the over reliance that society is placing on AI, industries that seem to be forcing AI onto their customers without a real need, the environmental and energy issues, and the future unknowns regarding capability and ability to control and regulate AI once it really takes off.

That's not even getting into the human jobs that will be eliminated. Like, I don't know, jobs that already allow AI to do 90% of the work...

Saw this and it makes one think:

https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fed6b6fe8-cab7-4a26-921e-4d7d5b49fb76_1536x2752.png

Posted
4 minutes ago, Vegas64 said:

I'm a noted bear on *this* particular flavor of AI and this AI bubble (while noting that AI is a transformational technology which will change everything we have known about labor, work, social, etc. in the future).

That said I was listening to one of the world's prominent quants talking about pod shops and hedge funds and we all know they are super advanced on all things technology to make a buck, and he made the point in a way that I've been thinking about it: today AI is not ready to take over. It's only synthesizing words and videos and text and images that man has already created for it to process (upon processing). But one day, in a few years even, maybe? WIth the advent of the sensory stuff (Meta glasses) and other stuff that will come (touch, taste, tactiles, etc.). The complexity of human sensing and consuming will get closer to what a machine can also consume and sense.

It will be an AI version of the meme "The Yellowstone Park Ranger has noted there is a large overlap between the smartest bear and the dumbest human, upon trying to create a bear-proof trash can".

The overlap between the dumbest person and the smartest machine COULD be a critical mass. And then maybe there is a tipping point.

And then what to do about the useless eaters?

Saw this and it makes one think:

https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fed6b6fe8-cab7-4a26-921e-4d7d5b49fb76_1536x2752.png

Will ai be able to read that cause I can’t?

Posted
3 hours ago, Captainant said:

[context: I am literally a certified AI/ML specialist and builder and have taken enough high-order math at UT to at least accurately conceptualize what's happening in the black box]

The end state of "unable to walk and falling down" is effectively the end-state of a "Hapsburg" model. When you train a model on synthetic data produced by a similar model, you get the same concentration of error effect that is observed in genetics whenever inbreeding happens in a population. 

In the current and active development state of AI, these LLM companies like Anthropic and Gemini and OpenAI are scraping the internet for ANY data they can, and they don't really care about the provenence of its origin. As a result, the "for-profit" LLM's are getting trained on a heavily poisoned and synthetic dataset that contains a bunch of data that will embed defects into the model and train it to produce worse errors. Not to mention the general problem with model poisoning that Anthropic published a paper on a month or two ago

 

Charles GPT 2

 

1*L_XRr-GcwpnpzsT9Rj1OzA.jpeg&f=1&ipt=21

“did not contain a single drop of blood; his heart was the size of a peppercorn; his lungs corroded; his intestines rotten and gangrenous; he had a single testicle, black as coal, and his head was full of water.

 

  • Haha 1
  • Fuck Around and Find Out 1
Posted
33 minutes ago, Dnaguy said:

 

Charles GPT 2

 

1*L_XRr-GcwpnpzsT9Rj1OzA.jpeg&f=1&ipt=21

“did not contain a single drop of blood; his heart was the size of a peppercorn; his lungs corroded; his intestines rotten and gangrenous; he had a single testicle, black as coal, and his head was full of water.

 

Charles II as "short, lame, epileptic, senile, and completely bald before 35, he was always on the verge of death, but repeatedly baffled Christendom by continuing to live."

  • Haha 1
Posted
1 hour ago, Vegas64 said:

image.png.cbdb04628876d8a510bf32ba75157a57.png

I disagree pretty strongly with this chart, mainly because it's making the same fallacious leaps in logic that idiot middle managers do about how much """AI""" can do. Assuming that every judge, patient, and consumer will just happily lap up AI slop is a little outrageous. This is still extremely bullish on AI in general, despite your preferred posturing as "bearish"

Posted (edited)
3 minutes ago, Captainant said:

I disagree pretty strongly with this chart, mainly because it's making the same fallacious leaps in logic that idiot middle managers do about how much """AI""" can do. Assuming that every judge, patient, and consumer will just happily lap up AI slop is a little outrageous. This is still extremely bullish on AI in general, despite your preferred posturing as "bearish"

To be clear I didn't generate the chart and I don't necessarily agree with it, I was responding to the poster who mentioned AI with certain industries as conversation point and to add to that particular discussion. This post was a disconnect from MY personal opinion which was the bear point of view. This site software combined the two into one post as I posted back to back too quickly.

Edited by Vegas64


×
×
  • Create New...