Yes, the lying is not from the fancy autocomplete computer logic and programming, the intelligence and intent come from the human data sources it uses to autocomplete the next word. The machine mirrors the data input.

How do you account for the same model giving right and then wrong answers to the same prompt over and over? Sometimes I'll get a correct formula then I run the same code again and it's now wrong. Other times it will assume it's right if I tell it is and wrong if I tell it it's wrong but it can't really tell.

The output is currently highly dependent on the user's technical prowess not just the internal dataset and this seems to be affected by the user's cache as well. If I was to personify the machine, I'd say it's delirious not an habitual fibber.

 

How do you account for the same model giving right and then wrong answers to the same prompt over and over? Sometimes I'll get a correct formula then I run the same code again and it's now wrong. Other times it will assume it's right if I tell it is and wrong if I tell it it's wrong but it can't really tell.

The output is currently highly dependent on the user's technical prowess not just the internal dataset and this seems to be affected by the user's cache as well. If I was to personify the machine, I'd say it's delirious not an habitual fibber.

Because they have stochastic randomness in the system as a mixer.
https://en.wikipedia.org/wiki/Stochastic_process

https://medium.com/geekculture/why-chatgpt-lies-4d4e0c6e864e
Why ChatGPT lies
The problem with Generative AI that causes it to spew misinformation.

A Quick Refresher on ChatGPT
To understand the reason it lies, let’s first do a very quick refresher on how it works. Specifically, I will be focusing on the latent space embedding, the pretraining with large data corpora, and the generative word-by-word output format since they are the biggest culprits to ChatGPT/other Deep Learning modules and their tendency to lie. These 3 components work as follows-

1. The datasets are encoded into a latent space.
2. The latent space is used to train the data.
3. ChatGPT then uses this encoded space to process your query. The query is fed into the latent space. The AI traverses the latent space to find the best outputs.[/quote

ChatGPT is Fancy Autocomplete
Here is how ChatGPT produces its output- it uses the embedding and the datasets, to predict the most likely next word. This process is stochastic, which is why the same query can create different outputs. But at its core, ChatGPT functions very similarly to the autocomplete on your phone.

The fact it depends (for better responses) on the users technical prowess for crafting the query only means it can narrow the range of autocomplete responses.

 

And if the data was input by biased jerks ....

That's a given seeing how the entire internet of social media is part of the input data.

 

Because they have stochastic randomness in the system as a mixer.
https://en.wikipedia.org/wiki/Stochastic_process

https://medium.com/geekculture/why-chatgpt-lies-4d4e0c6e864e
Why ChatGPT lies
The problem with Generative AI that causes it to spew misinformation.


The fact it depends (for better responses) on the users technical prowess for crafting the query only means it can narrow the range of autocomplete responses.

My understanding is the stochastic randomness is a bug not a feature. It is a necessary bug in the sense that humans don't always use the same set of words in the same way. We constantly mix our way of speaking just as I'm doing now and I don't really choose these exact words. They are my way of speaking but they are also random in their occurrence which bubble up from my sub-conscience. I find it striking AI is plagued by this sort of ineffable quality just as the human mind (entering one wrong figure into the calculator ruins the entire operation). From my experience, our current AI systems are not able to identify when they enter this persona. Part of my job is to specify when this happens in addition to form of misinformation itself. The former is not so easy to do because part of the prompt invariably includes telling the bot to take on some sort of persona. We get around this by inputting an elaborate system prompt prior which govern any subsequent prompts. In other words, the user MUST supply a LOT of data to get anything realistic and worthwhile back.

 

Part of my job is to specify when this happens in addition to form of misinformation itself. The former is not so easy to do because part of the prompt invariably includes telling the bot to take on some sort of persona. We get around this by inputting an elaborate system prompt prior which govern any subsequent prompts. In other words, the user MUST supply a LOT of data to get anything realistic and worthwhile back.

That's what we once called, putting a straitjacket on a mad man. Yes, he's still crazy but he can't bite you while spitting at the pink elephant.

 

​

 

 

https://arstechnica.com/information...3-release-excels-at-ai-generated-body-horror/
New Stable Diffusion 3 release excels at AI-generated body horror

AI image fans are so far blaming the Stable Diffusion 3's anatomy failures on Stability's insistence on filtering out adult content (often called "NSFW" content) from the SD3 training data that teaches the model how to generate images. "Believe it or not, heavily censoring a model also gets rid of human anatomy, so... that's what happened," wrote one Reddit user in the thread.

Basically, any time a user prompt homes in on a concept that isn't represented well in the AI model's training dataset, the image-synthesis model will confabulate its best interpretation of what the user is asking for. And sometimes that can be completely terrifying.

Looking at lots of nudes and studying human anatomy might actually be important to an artist.

 

https://www.msn.com/en-us/money/technology/ar-BB1oDl5z
AI is exhausting the power grid. Tech firms are seeking a miracle solution.

A spike in tech-related energy needs in Georgia moved regulators in April to green-light an expansion of fossil fuel use, including purchasing power from Mississippi that will delay closure of a half-century-old coal plant there. In the suburbs of Milwaukee, Microsoft’s announcement in March that it is building a $3.3 billion data center campus followed the local utility pushing back by one year the retirement of coal units, and unveiling plans for a vast expansion of gas power that regional energy executives say is necessary to stabilize the grid amid soaring data center demand and other growth.

In Omaha, where Google and Meta recently set up sprawling data center operations, a coal plant that was supposed to go offline in 2022 will now be operational through at least 2026. The local utility has scrapped plans to install large batteries to store solar power.
...
But there is deep skepticism in the scientific community that Helion or other fusion start-ups will be sending juice to the power grid within a decade, much less the kind of too-cheap-to-meter, safe electricity the tech companies are chasing.

“Predictions of commercial fusion by 2030 or 2035 are hype at this point,” said John Holdren, a Harvard physicist who was White House science adviser during the Obama era. “We haven’t even yet seen a true energy break-even where the fusion reaction is generating more energy than had to be supplied to facilitate it.”


Promises that commercial fusion is around the corner, he said, “feeds the public’s belief in technological miracles that will save us from the difficult task of dealing with climate change … with the options that are closer to practical reality.”

AI and Fusion, one hell of a hype train.

 

https://www.msn.com/en-us/money/technology/ar-BB1oDl5z
AI is exhausting the power grid. Tech firms are seeking a miracle solution.


AI and Fusion, one hell of a hype train.

Remember, the human body generates more bio-electricity than a 120-volt battery and over 25000 BTUs of body heat.

AI might see a solution there.

Fate, it seems, is not without a sense of irony...

 

AI might see a solution there.

I just hope that AI will have the decency of letting us decide between the blue and the red pill ...

 

I just hope that AI will have the decency of letting us decide between the blue and the red pill ...

 

https://www.businessinsider.com/ai-return-investment-disappointing-goldman-sachs-report-2024-6
Goldman Sachs says the return on investment for AI might be disappointing

"AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn't designed to do," Jim Covello, the head of Global Equity Research at Goldman Sachs, said in the report.

"The starting point for costs is also so high that even if costs decline, they would have to do so dramatically to make automating tasks with AI affordable," he added. "In our experience, even basic summarization tasks often yield illegible and nonsensical results."

He's not wrong. Google scaled back its AI use in search after its bot began making some odd suggestions, including telling a Business Insider correspondent to put glue on their pizza to keep the cheese in place.

The tech industry is also "too complacent in its assumption that AI costs will decline substantially over time," especially when that assumption seems to rely on competition dethroning Nvidia, which dominates the market with its AI chips, Covello said.

 

https://www.datacenterdynamics.com/...n-for-2024-ai-startup-set-to-lose-5bn-report/
OpenAI training and inference costs could reach $7bn for 2024, AI startup set to lose $5bn - report

The company now employs about 1,500 people, which could cost $1.5 billion as it continues to grow, The Information estimates - OpenAI had originally projected workforce costs of $500 million for 2023 while doubling headcount to around 800 by the end of that year.

The company is bringing in about $2bn annually from ChatGPT, and could be set to bring in nearly $1bn from charging access to LLMs.

OpenAI recently generated $283 million in total revenue per month, which could mean full-year sales of between $3.5bn and $4.5bn.

That would leave a $5 billion shortfall, with the company likely in need of fresh funds within the next 12 months.

 

https://www.ft.com/content/ae507468-7f5b-440b-8512-aea81c6bf4a5
The problem of ‘model collapse’: how a lack of human data limits AI progress

The use of computer-generated data to train artificial intelligence models risks causing them to produce nonsensical results, according to new research that highlights looming challenges to the emerging technology. Leading AI companies, including OpenAI and Microsoft, have tested the use of “synthetic” data — information created by AI systems to then also train large language models (LLMs) — as they reach the limits of human-made material that can improve the cutting-edge technology. Research published in Nature on Wednesday suggests the use of such data could lead to the rapid degradation of AI models. One trial using synthetic input text about medieval architecture descended into a discussion of jackrabbits after fewer than 10 generations of output.

These systems are fancy autocomplete programs totally dependant on human intelligence and actual data. They have zero 'artificial' intelligence and it been shown time and time again that 'artificial/synthetic' data sources produce total garbage.

 

https://www.atlasobscura.com/articles/is-this-the-first-time-anyone-printed-garbage-in-garbage-out
Is This the First Time Anyone Printed, ‘Garbage In, Garbage Out’?

 

https://www.theregister.com/2024/08/14/microsoft_services_agreement_update_warns/

Microsoft is notifying folks that its AI services should not be taken too seriously, echoing prior service-specific disclaimers.
In an update to the IT giant's Service Agreement, which takes effect on September 30, 2024, Redmond has declared that its Assistive AI isn't suitable for matters of consequence.
"AI services are not designed, intended, or to be used as substitutes for professional advice," Microsoft's revised legalese explains.

 

https://www.theregister.com/2024/08/22/gartner_agi_hype_cycle/
AGI is on clients' radar but far from reality, says Gartner
Controversial concept may not even be a useful goal in computing

In its Hype Cycle for Emerging Technologies, 2024, Gartner says it distills "key insights" from more than 2,000 technologies and, using its framework, produces a succinct set of "must-know" emerging technologies that have the potential to deliver benefits over the next two to ten years.

The consultancy notes that GenAI – the subject of volumes of industry hype and billions in investment – is about to enter the dreaded "trough of disillusionment." Arun Chandrasekaran, Gartner distinguished VP analyst, told The Register:

"The expectations and hype around GenAI are enormously high. So it's not that the technology, per se, is bad, but it's unable to keep up with the high expectations that I think enterprises have because of the enormous hype that's been created in the market in the last 12 to 18 months."
...
"Even the timeline for reaching it or even what AGI means is uncertain. I believe that machines are good at certain things and human beings are good at certain things, and I don't know whether trying to create a machine that thinks and acts like a human being may be the most desirable or the most optimal goal."

"at least 10 years away"
The most common tech support question now is, "how can I turn this off?"

 

https://www.theregister.com/2024/08/22/gartner_agi_hype_cycle/
AGI is on clients' radar but far from reality, says Gartner
Controversial concept may not even be a useful goal in computing


"at least 10 years away"
The most common tech support question now is, "how can I turn this off?"

"as least 10 years away" -- Translation -- just after we have commercially viable nuclear fusion.

My guess is that most of the players riding the AI hype wave fully understand it's a crap pipe dream, but that's okay, because it is something that is allowing them to rake in lots of money right now and they don't care whether it will (or can) ever amount to anything truly useful. They will simply move onto to the next hype wave, confident that one will always be around to take advantage of. On the other hand, sometimes, as an unintended consequence, long-shot technologies get the change to become something useful because of the presence of a hype wave and the people that ride it.

 

"as least 10 years away" -- Translation -- just after we have commercially viable nuclear fusion.

My guess is that most of the players riding the AI hype wave fully understand it's a crap pipe dream, but that's okay, because it is something that is allowing them to rake in lots of money right now and they don't care whether it will (or can) ever amount to anything truly useful. They will simply move onto to the next hype wave, confident that one will always be around to take advantage of. On the other hand, sometimes, as an unintended consequence, long-shot technologies get the change to become something useful because of the presence of a hype wave and the people that ride it.

EVs and autonomous vehicles come to mind.