Artificial Intelligence thread

[Merkin]

[CatsbyAZ]

One point of contention with Artificial Intelligence that is not yet covered on this thread is the disruptive amount of computational power needed to facilitate AI operations. Such as by Large Language Models such as ChatGPT. To quote a recent feature titled The Scramble for AI Computing Power published in the American Affairs Journal (May 2024): “By predicting text sequences from large corpora of training data, systems like ChatGPT not only discover the rules of natural language, but also learn common sense reasoning and other forms of abstract thought. There’s only one catch: the computing cost required to train a model grows exponentially with its raw capability.”

Additionally: “The projected energy demands from AI are so substantial that compute providers such as Microsoft [OpenAI] and Amazon have plans to vertically integrate their own power generation.”

According to the American Affairs Journal what is at stake behind all this demand for power?
“AI will largely augment the work we already do. Average programmers with a coding copilot can become 10x software engineers; doctors with a medical chatbot can get an instant second opinion; and lawyers can use customized models to draft docu­ments and summarize evidence, letting them take on more clients. Overtime, however, AI is trending toward agent-like systems that surpass human experts at a wide variety of tasks, if not entire categories of work."

MY THOUGHTS: Notice the phrase “agent-like systems.” Meaning industrial-scale applications, such as those discussed earlier in this thread to replace wholesale functions of film making. In the short term, AI is positioned to takeover the more tedious operations of high volume work, such as special effects generation for films. Some independent forecasts predict AIs to be competitive with most college-educated labor by as early as sometime in 2026.

If AI promises to be this effective at an industrial scale, and if power requirements promise to be this strapped by increasing AI computing power, I half wonder whether AI leaders like OpenAI will start limiting their computing powers away from at-home users like myself? For the sake of saving their computing for more profitable, scalable efforts. I use ChatGPT almost daily for editing. Another buddy of mine relies on ChatGPT to test run software coding he does for his Silicon Valley employer. A teacher I know uses ChatGPT to outline lesson plans for their history classes. Another friend uses Midjourney and Dall-E imaging to generate pinups that all suspiciously resemble Amber Heard.

These are all single-function consumer uses that not only occupy increasingly demanded computational power but more importantly these functions, when processed through AI applications, eat up far more computing power than if processed through legacy non-AI applications. For example, I don’t need ChatGPT for editing, but it’s more convenient. Microsoft Word already has editing features that don’t require extensive computing.

More quotes from the American Affairs Journal–
“As the main currency in a post-AI economy, the future will be determined by those with access to large computing clusters and the energy needed to power them. Those clusters will ideally be located in the West, but with the monopoly risk looming in the background, it may not suffice to cede control to purely private hands. The power unleashed by future AI models will challenge our basic governance structures to their core, busting through decadent procedures and driving demands for new controls over the distribution of compute—if not outright public ownership.”.

“The United States is currently primed to be the leader in both AI and global AI governance thanks to its enormous compute advantage over every plausible nation-state competitor. 21 U.S.-based firms own 70 per­cent of the global cloud computing market, with the bulk of that market share split between just three companies: Amazon, Microsoft, and Google.”

[dovecanyoncat]

Dear TEP customer,

SkyNet has terminated your access to the grid for August 1 thru August 25 in order to learn how to make AI Futanari Dickgirl porn.

Please do not reply to this auto-generated message.

Tucson Electric Power

[CatsbyAZ]

"There’s only one catch: the computing cost required to train a model grows exponentially with its raw capability."

An example of AI's costly computing impacts, specifically when it comes to the environment – from the Daily Mail: The dried-up lake in Mexico fueling fears AI technology is an environmental catastrophe in the making:

“…while governments and corporations alike rush headlong to embrace AI, growing numbers of environmentalists and digital computing experts are warning that we could be sleepwalking into a catastrophic ecological disaster.”

Important explanation on how AI generates its outputs, and the additional power consumed to do so:
“AI isn't actually intelligent in the sense that it thinks for itself. AI programs such as Bard are called Large Language Models (LLMs) because in order to 'learn' they are fed huge amounts of data. Within this data — billions of pages gleaned from the internet — LLMs recognize patterns within previous references to any given subject and to prior questions and responses relating to it. So when you ask an LLM chatbot a question, it is not reasoning and giving you an answer based on what it 'thinks'; it is calculating which response is statistically most appropriate out of trillions of pieces of information from almost all of the internet. According to McGovern, an AI search using this methodology consumes around ten times more energy than a Google search, which simply recovers answers from an index rather than scouring the whole web.”

“Last year, researchers at the University of California Riverside published a paper predicting that data centres worldwide would be using up to 6.6 billion cubic meters of water by 2027 — equivalent to half the UK's annual usage. Worst affected by shortages because of the need for all this water are some emerging and developing countries where tech companies can set up their operations with less environmental and legal regulation than in their own country…alarm bells are ringing in Mexico, Costa Rica and parts of Africa where communities are already in competition with data centers over water and mineral resources.”

The article then goes on to detail the environmental impact of Microsoft Data centers recently built in Mexico’s Querétaro State, which are sucking area lakes dry to serve its AI Servers huge volumes of cooling water. This environmental change is collapsing the local agriculture and livestock industry reliant on stable lake levels. Food prices are driven higher and the dried out land is at increased risk of wildfires.

“The overwhelming desire from the people of the drought-ridden towns of Queretaro is for their rights to water to be put first. But as their taps run dry, they can only watch while vast quantities are pumped through the pipes of AI data centers next door.”

[Merkin]

As a long time computer programmer, I had no idea that water is still being used to cool computers. Thought that was a mainframe thing. We had huge water chillers on top of the roof of the computer center.

[dovecanyoncat]

So when you ask an LLM chatbot a question, it is not reasoning and giving you an answer based on what it 'thinks'; it is calculating which response is statistically most appropriate out of trillions of pieces of information from almost all of the internet. According to McGovern, an AI search using this methodology consumes around ten times more energy than a Google search, which simply recovers answers from an index rather than scouring the whole web.”[/i]

OT question: does AI then index its answers for subsequent similar searches, or does it always run the consumption heavy methodology?

[CatsbyAZ]

OT question: does AI then index its answers for subsequent similar searches, or does it always run the consumption heavy methodology?

We would think that as part of Generative AI’s machine learning capabilities, that yes, AI would build its answers into an index. However (surprisingly?), this is not the case. Short answer is – No. Generative AI does not typically use recent outputs from previous queries as part of continuing its outputs. Recently learned content can only be incorporated once an AI application is retrained on its updated LLM data sets. This is typically done periodically rather than continuously and acts as a measure of control over its data sets.

The Generative AI applications that are most in consumer use, such as ChatGPT, are programmed to learn from specified data sets. When we say Generative AI, the Generative capability is to denote a more layered, dynamic level of learning, where outputs are not merely quick references from existing data sets that it was trained on. Rather, Generative AI, can analyze linguistic patterns and data structures and take the next step by generating original outputs (new content) that extend or further develop the data it was trained on.

The capabilities for Active Learning AI do exist. But companies like OpenAI or Bard are likely holding back its full public release because they do not trust consumers to not take advantage of Active Learning by manipulating its learning into vast misinformation campaigns or harmful images that could become a liability for not only AI companies but for the platforms where the its outputs can be shared.

For example, last January Twitter had to pause searches for Taylor Swift after hoards of explicit deepfakes spread across its platform. And manipulating ChatBots that use Active Learning has been a problem for quite some time:

[Merkin]

This is an Al reconstruction of the long-destroyed hospital scene.
Originally intended as the ending for the film 'The Shining'.


https://www.youtube.com/watch?v=d3pLb1nhMt8

[CatsbyAZ]

Google is destroying its Search Engine based ad-revenue business model to pursue OpenAI in a costly and ongoing AI arms race.

Google remains spooked since OpenAI’s ChatGPT debuted as a commercial successful AI product in late 2022, generating outputs viable enough to accomplish the unthinkable. Almost overnight ChatGPT attracted substantial user traffic away from Google and emerged as the first real threat in over 20 years to the ad-revenue business modal Google operates through its search engine. This drastically triggered a “Code Red” within the corporate walls of Google: “Teams in Google’s research, trust, and safety division, among others, have been tasked with shifting focus to aid in creating and introducing artificial intelligence (AI) prototypes and products.”

Google’s solution was to make things worse for itself by rushing headlong into countering OpenAI by forcing its own AI functions (Bard, Astra, Gemini) into its search engine. Now over a year later this is proving to be a mess, especially when it comes to everyday businesses who’ve heavily invested in SEO (Search Engine Optimization) to promote their webpages higher on Google’s search results:

“Google’s new artificial intelligence (AI)-powered search summaries could significantly impact online publishers by driving users away from their websites and toward the tech giant’s own results, potentially reshaping the online landscape and raising concerns among content creators who rely on website traffic for advertising revenue…AI-powered search could be bad news for many websites that rely on clicks. According to research firm Gartner, traditional search engine volume will decrease by 25% by 2026, with AI chatbots and other virtual agents capturing a significant portion of the search marketing share.”

“Google is in a tricky position here…they can’t just cut out all the publishers and advertisers who create the content that makes the internet what it is. Google will need to balance several factors. They’ll need to find ways to use AI to improve and speed up search while still ensuring users have a reason to click through to websites and engage with the content there. Publishers and advertisers will also keep a close eye on this because their livelihoods depend on people visiting their sites.”

In the short term, the damage is already done to businesses relying on Google. Not only has Google’s forced AI results interrupted traffic into website for those businesses, and thus cut into their profits, but their dependable ad-revenues generated for Google is also slashed. And not only is Google’s AI arms race cutting into its ad-revenue earnings, but Wall Street is increasingly growing leery over whether AI itself will make money, starting with Google’s products:

“Earlier this week, Google released its second-quarter earnings, failing to impress investors with razor-thin profit margins and surging costs related to training AI models. Capital expenditures are surging far past what the company had been spending previously, as the Wall Street Journal reports, with this year's total spend expected to surpass $49 billion, or 84 percent higher than what the company averaged over the last five years.”

And as for Google’s primary AI competitor: “How much time does the tech industry have to stop bleeding cash as it pours money into the tech? If recent reports are to be believed, OpenAI may lose $5 billion this year and run out of cash in the next 12 months, barring further cash injections.”

[Chicat]

[pc in NM]

This is pretty cool!

[Chicat]

Good thing Elmo bought Twitter and took care of that bot problem.

[CatsbyAZ]

Another longer post, mostly compiled with quotes transcribed from the YouTube video embedded at the end.

For those of you that might've heard of the Dead Internet Theory, it hypothesizes that the internet died about a decade ago, once the number of interactions between Bots exceeded the number of human driven interactions across the internet; eventually almost all interactions on the internet will occur between Bots.

Where this theory becomes conspiracy (theory) among its adherents is their belief that "the U.S. Government is engaging in an Artificial Intelligence powered gaslighting of the entire world population." The exact method? "A.I. powered Bots that have subtly subverted and steered culture towards nefarious ends."

Aside from the theory's more absurdist claims trafficked on 4Chan, "the internet does feel empty now, doesn't it? Nothing to see, nothing to do, like a hot air balloon with nothing inside."

"...the evidence that the internet may have slipped out of human control is everywhere...[As of 2016] Bots are responsible for over half of all web traffic...Google is currently worried that its world changing algorithm will drown in the current tidal wave of A.I. generated content and become far less useful, if at all."

In my opinion, Google search results already suffer from a glut of A.I. driven content, much of which is Google's own fault - see my last post.

"If I were to add anything to the Dead Internet Theory it would be that the public release of ChatGPT was a watershed moment...it's release and record breaking adoption both created enormous economic incentive for every knowledge-work industry to create and incorporate generative A.I. into their businesses, turbocharging the technology and allowing anyone to create bottomless pits of lifeless digital corpses..."

"The internet 'died' not because people can be fooled by manipulated media...it died because synthetic media like that created by Sora, ChatGPT, Midjourney, and others can emulate human creativity at a scope and scale that is impossible for humans to engage...after its release ChatGPT was creating more synthetic texts for its users than appears in all physical books ever written, every two weeks."

"...like history can be meaningfully split into the time before nuclear weapons and the time after, Generative AI and the death of the internet seems like it will be another historical delineation. They are both technologies that instantly changed the world. It mattered who had them and who developed them...Generative A.I. is the nuclear bomb of the Information Age."

MY THOUGHTS: The internet is an increasingly stale experience in several ways. For nearly all of its users, including myself, the internet is about a half-dozen websites we willingly visit daily (I say willingly because I'm excluding websites used for specific chores like banking and paying the power bill). My repeat websites are YouTube, Twitter/X, Wikipedia, Google Maps, and two sports message boards I've frequented for years.

Of these, only on the message boards, which came online before the advent of social media, do the interactions feel organic. The rest of the more trafficked internet largely looks the same thanks to Google's SEO promoting websites that promote a flatter appearance. The most visited websites (Facebook) are overrun with spammy adverts and predictably targeted content, and it's all turning into the same "optimized" experience.

Gone are the days of highly individualized profiles or personalized blogs or Chat Rooms. The older, 'developmentally un-zoned' internet didn't work as efficiently and wasn't as organized or navigable but that's the point; the internet back then wasn't made for us to live out our lives on like it is now. Back then the internet wasn't meant to be an alternate reality for us like the internet is increasingly becoming. My hope is that rather than us all being slowly lulled into an internet taken over by AI content, we'll reach a point where eventually we'll learn to start tuning it out, unable to keep both feet in, and unable to relate to or keep up with its Hyperreality.

More details in the Video:

[Merkin]

Kind of related I suppose.

[dovecanyoncat]

When AI is Just Badly Paid Humans!

https://www.youtube.com/embed/huu_9rAEiQU

Patrick Boyle is mostly about things financial, and his Brit dryness can be hilarious. But I found this interesting, if not completely in line with thee spirit of this thread.

Embed doesn't seem to be showing up.

<iframe width="853" height="480" src="https://www.youtube.com/embed/huu_9rAEiQU" title="When AI is Just Badly Paid Humans!" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen></iframe>