- Joined
- Nov 16, 2005
- Messages
- 21,017
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.
How AI will fail
- Thread starter rinaldo00
- Start date
The thread topic title is "How AI will fail" and it severely lacking in advanced functionality (call it intelligence or whatever else) is one of the main reasons it will fail to deliver on promises, at least all the currently known forms of AI.
StevenJamesBorchard
Limp Gawd
- Joined
- Feb 22, 2025
- Messages
- 271
With regards to OP title, AI and automated people interaction ( whatever) will fail because actual people usually can tell authentic real empathy and compassion from NOT.
and specifically from what a lot of "AI" and automated scripting takes over is customer care and information positions.
Take it from someone who has well over 70,000 phone calls taken across 3-4 different companies.
Actual humans want another real human in front of them or on the phone to argue with and get pissed at, if nothing else to vent and be angry even if no progress or positive outcome is from it.
AI's/auto mated scripts just aren't the same, while the real person just turns phone volume down takes a deep breath and often really hangs in there!
Sorry if this doesn't really seem relevant but I too also engage in " TL: DR " procedures.
and specifically from what a lot of "AI" and automated scripting takes over is customer care and information positions.
Take it from someone who has well over 70,000 phone calls taken across 3-4 different companies.
Actual humans want another real human in front of them or on the phone to argue with and get pissed at, if nothing else to vent and be angry even if no progress or positive outcome is from it.
AI's/auto mated scripts just aren't the same, while the real person just turns phone volume down takes a deep breath and often really hangs in there!
Sorry if this doesn't really seem relevant but I too also engage in " TL: DR " procedures.
mullet
2[H]4U
- Joined
- Aug 19, 2004
- Messages
- 3,291
I don't know which is going to run out first the money or the power, that is the true race to the failing point.
Except just as we all know and understand (even if we pretend not to) ML=AI now, they're interchangeable (and have been used this way for a long, long time) - and what is 'traditionally' though of/referred to as 'AI' = AGI
And to that point, the thread's/video title isn't apt in the same way, as it wasn't talking about 'AI itself' failing, but referring to 'AI' as in 'cloud/non-local/datacenter' 'bIg MeAn CoRpOrAtE AI' failing - in leu of locally ran AI.
The number of people able and willing to run a meaningfully capable local AI is a statistical error. That will be the last thing that kills the corpo AI.
I agree - as I touched upon earlier, as well as mentioned other reasons why - just because I run a local Plex server at home does not mean I also think Netflix will die. Same thing essentially.
Edit: And just to clarify - I think among 'normies' it will be a statistical anomaly - I could see non-statistical anomaly but still low single-to-double-digit on-prem for corp use for sake of argument - but that still won't negate the big guys/cloud/datacenter.
Last edited:
- Joined
- Jul 13, 2006
- Messages
- 2,192
I don't think I am super amazing technically advanced and I am running Qwen3.6 locally with vllm and an open source mcp and vectordb works pretty well for me.
You being here on this site just talking about it puts you outside of the group of normies into edge case/power user/enthusiast - no matter how many more 'power users' are actually more advanced than you etc
- Joined
- Jun 7, 2004
- Messages
- 3,653
AI is going to fail exactly how Google web search "failed".
- Joined
- Nov 16, 2005
- Messages
- 21,017
It's just gonna be better search? With a ton of ads.
- Joined
- Jul 13, 2006
- Messages
- 2,192
That's what "local ai" kinda means at this moment in time. If I can get it running well in my spare time, it won't be long before there are fully built tested and hardened local ai inference boxes for 2-3k (maybe more at first).
(currently 3d printing a fan shroud for a 32gb v100 i got off ebay so i have other gpus to play with).
But again, the normies (and especially en masse) are not going to buy those. Another example is Synology BeeStation (or tons of similar products released over the years/decades) that is a 'preset/set it and forget it NAS/Server' - meant for normies - yet not popular, still a niche in a niche, didn't kill cloud/cloud still alive and well.
- Joined
- Jul 13, 2006
- Messages
- 2,192
My mother and I had a relatively long argument about me spending my small savings to buy a used commodore 64 at a yard sale because she didn't think computers were very important. Now we (people) can't imagine living without them.
Difference is, I don't think we'll have that long of an adoption period for local ai.
But you are not the benchmark for you are not a normie.
Become more and more useless over time?
littlegreenmenfromhome
Limp Gawd
- Joined
- Aug 27, 2025
- Messages
- 372
How do you get those calls from voice to text?
Or do you just do the wrong thing and hope every detail is logged correctly by humans and miss the patterns in large volume of calls.
The proper way would be to:
1. First step use A.I. models to get the recorded call to text. (Voice‑to‑Text / Interference‑Robust / Noise‑Resistant AI)
2. Then another trained models analyze, catagorize or trace resolution. (Text Classification Model / Intent Detection Model / Root‑Cause Analysis Model / LLM Reasoning Layer)
3. Have yet another set of models optimize based on the real world data and make sure call-center traning, customer material and guides ect. are adapted, so the volume of calls, time-to-resolution are redcued and right-resolution-first-time becomes king. (Pattern‑Mining AI / Process Optimization AI / LLM‑Based Training Generator / Continuous Improvement Model )
You are blinded by bias and habbits and cannot see the forest for trees.
(Guess what the 8 GPU server I am implementing is doing?
)A lot of people crying here are the ones missing the big picture and will be the ones "left behind".
Last edited:
Soulstorm brew
Weaksauce
- Joined
- Mar 29, 2019
- Messages
- 125
Or our water.
- Joined
- Jul 13, 2006
- Messages
- 2,192
The "left behind" mantra I don't think is accurate either.How do you get those calls from voice to text?
Or do you just do the wrong thing and hope every detail is logged correctly by humans and miss the patterns in large volume of calls.
The proper way would be to:
1. First step use A.I. models to get the recorded call to text. (Voice‑to‑Text / Interference‑Robust / Noise‑Resistant AI)
2. Then another trained models analyze, catagorize or trace resolution. (Text Classification Model / Intent Detection Model / Root‑Cause Analysis Model / LLM Reasoning Layer)
3. Have yet another set of models optimize based on the real world data and make sure call-center traning, customer material and guides ect. are adapted, so the volume of calls, time-to-resolution are redcued and right-resolution-first-time becomes king. (Pattern‑Mining AI / Process Optimization AI / LLM‑Based Training Generator / Continuous Improvement Model )
You are blinded by bias and habbits and cannot see the forest for trees.
(Guess what the 8 GPU server I am implementing is doing?
A lot of people crying here are the ones missing the big picture and will be the ones "left behind".
To most people these are a black box, and so they're filling in the missing details often incorrectly.
But AI will become as ubiquitous as laptops. It will be a commodity. People will form better understandings and will eventually embrace it and like any tool this will expand the capacity of people and be a net enabler. People can and do argue this point but that counter claim has no historical precedent.
- Joined
- May 18, 1997
- Messages
- 58,009
- Joined
- Jun 7, 2004
- Messages
- 3,653
Well... I was thinking sort of the opposite (sorta). As someone that hires and interviews people, tech candidates are completely lost without Google nowadays, and IMHO, in a very short period of time, it will be AI that is required by the candidate to "present actual knowledge".
- Joined
- Jun 7, 2004
- Messages
- 3,653
Oddly, AI probably cuts out the middle manager even moreso than the worker bee. But again, in a short time, the worker bee will lose that final bit of creative problem solving skill by leaning on AI for all answers. And no, we "won't check" AI's answers anymore. Just like we assume that Google results are "gospel".Cut out the middle-man, just hire the AI
- Joined
- Jul 13, 2006
- Messages
- 2,192
If you look at what Nvidia is pushing at GTC you will notice a shift. In 2024, it was mostly pushing AI, with hints of the next phase. Phase 2 is how it comes together, and if you're working in robotics or any 'cyber physical system' you can't miss it.
https://www.nvidia.com/gtc/keynote/?regcode=no-ncid&ncid=no-ncid
Jensen sees things with more clarity than the average person, and imho, people should really pay attention to what he is selling and think about scale. The highlighted companies, I have some insights into and they are all going to need big iron to run phase 2. (I might have missed some, and I don't know all of these companies or their exact business)
Last edited:
StevenJamesBorchard
Limp Gawd
- Joined
- Feb 22, 2025
- Messages
- 271
How do you get those calls from voice to text?
Or do you just do the wrong thing and hope every detail is logged correctly by humans and miss the patterns in large volume of calls.
The proper way would be to:
1. First step use A.I. models to get the recorded call to text. (Voice‑to‑Text / Interference‑Robust / Noise‑Resistant AI)
2. Then another trained models analyze, catagorize or trace resolution. (Text Classification Model / Intent Detection Model / Root‑Cause Analysis Model / LLM Reasoning Layer)
3. Have yet another set of models optimize based on the real world data and make sure call-center traning, customer material and guides ect. are adapted, so the volume of calls, time-to-resolution are redcued and right-resolution-first-time becomes king. (Pattern‑Mining AI / Process Optimization AI / LLM‑Based Training Generator / Continuous Improvement Model )
You are blinded by bias and habbits and cannot see the forest for trees.
(Guess what the 8 GPU server I am implementing is doing?
A lot of people crying here are the ones missing the big picture and will be the ones "left behind".
automation and use of AI is flawed unless alternative and proactive positions and proactive care is first implemented for the displacement and change of human beings and workers.
But its just different world views and philosophies. That and our freedom of speech and opinions are often difficult to understand when we don't know or understand each other's time scales and value we place on specific things in our world views and philosophies.
Most people have a good understanding of how the world is, but can't envision what it can ultimately be. That's the difference between the guys on the high tech side pushing an envelope and everyone else. Humanity went from the Wright brothers to landing men on the moon in 66 years, literally within the lifespan of a single human. I can't stress enough how insane that is, and if that doesn't make people believe that almost anything is possible, I don't know what can.
I'm actually surprised at the number of people on tech forums that are bearish and unimaginative about the trajectory and potential with AI. I understand everyone's bitter about the negative impact on our tech hobby, but I will contend that people who are doubting the future and potential of AI are simply not paying attention to what's happening in that space. Anthropic went from a $1 billion ARR in 2025 to an estimated $30 billion ARR now, and that's real revenue. Stuff is happening whether we like it or not.
- Joined
- Nov 16, 2005
- Messages
- 21,017
- Joined
- Jul 13, 2006
- Messages
- 2,192
Im a huge bull on AI as a concept and a tool. Im a little bearish on the short term economics of it.
I am pointing this out because AI as a tool vs AI as it is being built out today are not the same thing and is also a limited imagination of what AI can be and probably will be long terms as a decentralized commodity like anything else.
The Linux kernel is on its way to be completely vibe coded... The whole thing is a complete disaster in the making.
- Joined
- Jul 13, 2006
- Messages
- 2,192
https://www.techspot.com/news/112166-linux-may-drop-legacy-network-drivers-amid-surge.html
https://www.techradar.com/pro/linux...coding-just-dont-use-it-on-anything-important
I don't see any actual maintainers accepting ai code.
I see reports bug reports and wasting dev time on ai code, but that's a different problem.
Is there evidence AI code is actually being accepted by maintainers?
The technology is in its infancy right now. It requires a lot of up-front investment, sure, but it's a necessary part of the process. Most of the companies pouring the cash in can support it from cash flows from operations.
Recall back in the dot-com era that a whole ton of fiber had to be laid. Initially, this looked silly. Who can possibly use all of this bandwidth? Now we have people streaming literally everything over the internet. We needed that robust network to facilitate what a lot of people never envisioned would ever be a thing. Smart tech people saw the infrastructure being laid and knew what that would enable them to ultimately do once the capacity existed. It's going to be the same with AI. A lot of datacenters are going up right now, and these companies are literally burning out these GPUs. As I said, Anthropic's REVENUE has exploded. It's currently the fastest revenue growth rate of any company in history. This is revenue from operations that they're bringing in. AI is coming whether we like it or not.
- Joined
- Nov 11, 2012
- Messages
- 1,475
What we have now isn't AI. At best, it's a LLM scaled per datacenter. The mistake that's being made is to make LLM-AI supposedly like an all-knowing entity. Basically, a single entity with exceptional micro and macro abilities, like a hive mind (think of a Queen Bee or a Queen Ant). The problem is, humans know nothing of what that's like, even using the military chain-of-command model, it's still nothing like a hive mind.
A better model of AI, is to make a population of different AI entities that have exceptional micro abilities, but can communicate with other AI's to complete macro-level tasks. Basically, how humans in a group operate. LLM-AI will fail because the humans creating it are idiots. The ones that could actually create AI, won't, because they'd just be creating artificial humans that lack empathy.
A better model of AI, is to make a population of different AI entities that have exceptional micro abilities, but can communicate with other AI's to complete macro-level tasks. Basically, how humans in a group operate. LLM-AI will fail because the humans creating it are idiots. The ones that could actually create AI, won't, because they'd just be creating artificial humans that lack empathy.
- Joined
- Jul 13, 2006
- Messages
- 2,192
You're not telling me new information here and it doesn't change any of the analysis. I simply refuse to believe the data center build out is going to sustain global gdp growth numbers it needs in the short term. Long term is an entirely different discussion (15 years from now).
Classical over investment in the short term and under investment in the long term.
Nothing being discussed anywhere or any information factually contradicts this view.
Nonsense - there's a mad scientist for everything. (I think we lack the capability/technology/know-how to create AGI, not the will or the person willing to)
Very true. There are two ways to build out AI ... one is a slow and methodical process that focuses on specific tasks and optimizes the models for those tasks. It tests repeatedly until the reliability is good enough that they'd be willing to bet their life on it. Then they release it to the public as a single tool that gets used in a specific industry. In parallel they could be working on other tools (i.e. MRI, X-Ray, etc for the medical field, coding assistant (bug hunting), and so forth). The second approach is throwing everything and the kitchen sink at AI and just tossing it out and using the public as testers. They won't take the methodical and safe approach because it doesn't make for sexy headlines and investors these days just seem to chase sexy headlines. In the background, yes, some companies are building specific tools to do specific jobs but the big data centre build out is chasing the dream of artificial general intelligence. If it doesn't pan out then the specific tools will be buying data centre space on the cheap because the people running the things will be desperate. And, in the meantime, those data centre builds will ruin a whole lot of lives.
It isn't that AI can't be valuable its that the people running the show are a bunch of narcissistic, greedy buggers who just want to ride off into the sunset on investor money and be heroes.
littlegreenmenfromhome
Limp Gawd
- Joined
- Aug 27, 2025
- Messages
- 372