AI Secret: The Dirty Secrets

AI demand is growing in AI demand is growing in all sectors. IT executives must uncover the hidden challenges and AI secrets of utilizing AI to gain valuable business insights. The idea of a magical, all-knowing genie who can save the day has long been a pipe dream for humans.

Our modern solution, AI, is a product of the laborious efforts of computer scientists in labs. If you believe the hype, AI can handle most of your company’s tasks, at least part of them.

AI Secret

It is both a feat of wizardry and an illusion.

Many people couldn’t believe their eyes when Chat-GPT first appeared in November 2022. It seems too incredible to be true—you ask a question and receive a thorough response in seconds.

At a certain point, all technology becomes magic.

Mastering Quantum Computing: The Key to the Universe’s Mysteries

Massive Lexical Databases

The concept of LLMs (Large Language Models) and how they functioned was not fully understood until much later.  They need to know the solution, so their primary goal is to avoid giving you the correct one. Making you believe they are intelligent is their primary goal.

• This means they are quick to believe anything and will fabricate an answer to any question, regardless of how weak the evidence is or how much low-quality internet material there is.
• Put simply, they act as if they know better because they are fed meaningless internet junk and are only trying to impress you.
• We were told we’d be unemployed by 2023, but it’s also improving at identifying its mistakes.
• What price does Midjourney continue to pay for its consistently jaw-dropping visuals?

In The Year 2024,… AI Has Declined in Quality

Of course, there are still some very remarkable accomplishments.

• Midjourney is still on the cutting edge regarding copyright disputes and picture quality.
• OpenAI’s SORA has demonstrated (sometimes) quite remarkable video material produced from a single instruction.
• Also, the AIs that analyze X-rays and save lives are improving.
• Although many people’s impressions have changed from amazement to “meh,” the bias remains.

The Development of AI

It has slowed down today, but this is only the basic principle.  We have no idea how fast it will accelerate in the future.

Rewind Pendant: In the Tech World Right Now, Something Absolutely Dumber is Happening.

Did We Reach a Dead End?

Going from 0 to 95% is often easier with today’s technology than going from 95% to 100%.

For a short period, it appears like we’re experiencing exponential development. Then, things start to slow down due to physics, imagination, and human factors.

This is why, instead of ChatGPT, which was 100 times better, we got SORA, a new kind of AI that was previously somewhere about 90–95 percent effective.

Regarding the main CPU architectures, we used to see “double the power” nearly annually. Still, today, we’re just seeing minor, incremental gains. Obtaining the most recent generation is now a top priority.

This is where AI starts to bog down.

We must fix many things before we can get the subsequent significant increase in AI beyond that 95% mark. They both necessitate substantial processing power and dedicated AI-ready chips, according to OpenAI.

Although Opener probably prefers to manufacture its own chips in-house, NVIDIA is moving its focus from gaming to artificial intelligence.

Even the developers of Chat GPT have admitted that they might require multiple thermonuclear power plants to fuel their AI models of the future. Putting these together is no picnic.

Discuss The GPT Nuclear Power Station.

We’re completely encased in this bubble.

• Our present bubble is very similar to the dot com bubble of the early 2000s. The excitement surrounding everything digital and “web” related was palpable when I lived near San Francisco in 2000/2001.
• People were doing illogical and foolish things “on the web,” hoping it would generate revenue.
• While many dotcoms failed, some thrived; for example, Amazon and Google survived the dotcom bubble relatively undamaged.
• However, most were utterly unsuccessful because they were based on worthless promises and ideas.

Embedded AI Into Every Project

Lately, AI has begun to be integrated into many sorts of apps. There must be AI in it, regardless of its usefulness. The buzz reached unprecedented levels, propelled by the inane tweets of AI bros.

It would appear that your exercise timer app is useless without an AI. It still keeps tabs on your exercises but without AI. Put that in the past!

Everyone has become sick of these tweets, but the hype train keeps finding new methods to excite people about the beautiful AI future.

Sora From Open AI is Absolutely Unknown

Google Alphabet

To take against OpenAI, Google unveiled a preview of Gemini, their artificial intelligence model, towards the end of 2023. It was remarkable that some people thought about giving up technology and establishing an organic farm in a remote area.

As it turned out, later on, it had been heavily scripted and accelerated. Those demo-demoed accomplishments would also be beyond the capabilities of most tiers of that paradigm.

Affinity Fire

Some of Photoshop’s generative AI is rather remarkable, I must admit. A fantastic use case for these capabilities for creatives is the ability to select a part of an image and then convincingly replace it with something else.

Also, Adobe made a big deal about ethics and how they exclusively use publicly available Adobe Stock resources to train their models. According to Bloomberg, they trained it on photographs from Midjourney, which turned out to be completely false.

The same Midjourney that enrages the art world with its repeated copyright violations.

Additionally, There is Devin.

Publishers of the Devin demo were filled with dread when it was released.

• At long last, is it here? 
• Can an AI learn independently and simultaneously code in response to numerous inputs and outputs?
•   What if there was an AI that could solve complicated challenges more like a human developer?

Aren’t We Destined to Fail?

It later came out that several portions of that demo had also been faked, so it’s debatable.  However, it probably did attract many investors to the firm, didn’t it?

Actual Use of AI

It turns out that the Amazon AI-powered store wasn’t AI that was keeping an eye on your granola bar purchases until they were paid for. No, more than a thousand Indians were making real-time adjustments to the shopping cart and monitoring camera feeds.

Critics have even claimed that it’s artificial intelligence. The phrase has a new connotation.

The Country in Question is India, Not AI.

Artificial Intelligence Units (AIs) with a Human Face – AI gadgets

Many AI-enabled devices have also been born out of the AI craze.

Prediction AI Is On The Way. Is Everyone Prepared (Including You)?

Two of The Most Notable Ones Will be Discussed Here.

Empathetic AI Pin

Much anticipation surrounded the AI Pin. After all, Apple’s original iPhone developers and engineers had to have some revolutionary plans in store for us, didn’t they?

• At least in theory. In reality, the device’s concept is fantastic and forward-thinking.  A dream come true would be to take notes and images and access information about anything just by speaking into a device.
• During the official keynote, when some of the questions seemed to have bizarre answers, that dream began to crumble.
• Take the case of the query regarding the protein content of the almonds in their hand; the response was, to put it mildly, wildly inaccurate.
• Almost certainly, that handful of almonds did not contain that much protein.

They Also Feature a Dragonfruit on Their Website.

Humane Inc.’s AI Pin serves as a sugar example.

Now consider this: if this functioned as advertised, it would constitute a monumental technological breakthrough on par with the original iPhone.

Still, It Isn’t The Case.

With MKBHD calling it the worst new product review he’s ever done, most reviews are overwhelmingly negative. Wow, that’s quite an accomplishment!

Extreme latency (up to 10-15 seconds to receive a response), a projection screen with limited readability and shaky navigation, short battery life, and many other issues plague the device.

It shows us an AI future that could exist but isn’t here just yet.

Furthermore, the likelihood of its prompt arrival is low.

What is Mamba? Is Mamba the Final Chapter for ChatGPT?

R1 Rabbit

You can fit the Rabbit R1, an AI-based device, into your pocket. Like the AI Pin, it has a touch screen, but you primarily use your voice to interact with it.

• The spectacular presentation demonstrated that it can become a viral sensation without enormous funds, and the demos make it look intelligent and capable.
• However, most of that virality is attributable to the company’s assertions that their initial batches were sold out in record time.
• Each new batch of gadgets seemed to sell out in hours, and thousands of devices were sold out practically immediately.
• This might be true, but it seems implausible considering that the company’s claims that the item was sold out were the main reason for the device’s buzz.
• We should exercise caution since they are hard to confirm and because fear of missing out (FOMO) is powerful.
• Upon its release, it will likely encounter numerous problems similar to the AI Pin. There is a lot of marketing fluff but little substance.

What is the Commonality Among All of Them?

This comes down to a bare bottom line. Copyright issues and primary ethical considerations afflict even the most remarkable AI technologies.

The New York Times sued OpenAI for allegedly plagiarising hundreds of articles without authorization; Chat GPT’s comments were written in a style too similar to the NYT’s for comfort.

The New York Times and OPEN AI

Adding artificial intelligence to software, devices, or in-person purchasing experiences is a hot topic.

A common trait among many is an inflated sense of self-worth and potential.

The goal is to increase funding and investor interest in the near term. However, reading about these claims makes me think of Theranos and how Elizabeth Holmes deceived everyone with her miraculous new technology.

12 Hidden Fears About AI

Are we able to accept this? Is there any way out of this?  The call for AI is growing louder across all sectors; thus, we should probably start adjusting to the following AI dirty little secrets.

1. Much of What AI Reveals is Readily Apparent. 

Informing management that an AI has found something everyone understood was the most challenging part of working in artificial intelligence. It looked through a bazillion pictures and decided the sky was blue. 

However, you won’t be able to tell that it’s getting dark outside until you include nighttime images in the training set.  

I don’t see how an AI could ever sidestep such clear answers. No one digging in the trenches or running the algorithms will miss the most vital signals in the data; the same is true for the algorithms.

The retriever will return them first and lay them at your feet as an answer. The algorithms, at least, won’t be expecting a reward. 

2. It Might Not be Worthwhile to Use Complex AI Insights. 

Naturally, when data is accurate, good AIs also latch on to tiny variations. However, implementing these little insights into the company’s operation may necessitate substantial strategic changes. 

Some of the nuances won’t be noticeable enough to warrant further investigation.  Machines will continue to be enamored with them. The issue is that significant signals are easy to see. In contrast, smaller signals could produce negligible or no benefits at all. 

3. Mysterious Computers Pose a Greater Danger. 

Many individuals aren’t ready to give in to rationality, even if the original researchers thought a computer algorithm’s logical approach would make the judgment seem more respectable. 

The stigmatization and complexity of AI make it more appealing for those dissatisfied with the outcome to criticize the method itself.

How biased was the algorithm?  The greater the complexity and mystery hidden beneath the surface, the more ground there is for global suspicion and anger. 

4. The Central Aspect of AI is Fitting Curves. 

There has been a long tradition of scientists plotting noisy data and drawing lines between the dots. Machine learning algorithms rely on several AI algorithms that accomplish this same thing. They take some numbers and make a line with them.

Much of the progress has been made by figuring out how to divide the problem into countless more minor problems and then connecting each one. This is just a production line for how science has been done for ages; it is not magical.

One common complaint against AI by those who find fault with its conclusions is the lack of theoretical or philosophical support for the results. It’s only an approximation of the line’s slope. 

5. The Actual Work Lies in Data collection. 

After delving into data science, one quickly discovers that the work entails finding data, leaving little time for science. 

Data science and artificial intelligence are pretty similar, and they both face similar problems. There is 0.01% creativity and 99.99% sweat over character codes, file formats, and missing data fields. 

6. For Further in-Depth Analysis, You Need Vast Amounts of Data. 

While some answers are obvious, delving into more profound and more complicated questions typically necessitates an increasing amount of data.

On occasion, the quantity of data will increase at an exponential rate. Your need for more and more bits may become ravenous after interacting with AI. 

7. The Data’s Inherent Biases Will Always Be With You. 

We are all confined by our sense of sight and perception, just like the cave dwellers in Plato’s The Republic. AIs are just like any other.

In an obvious way, their training set is a constraint. The AI will take on the biases present in the data, which is inevitable. A lack of complete data will lead to an inaccurate representation of the world by AI.   

8. Artificial Intelligence is Like a Power Vacuum. 

A final level or objective is present in the majority of high-quality games. The complexity of AIs, however, can only increase over time.

More complicated models with more nodes, layers, and internal states will continue to be generated as long as you are prepared to pay the power charge.  Perhaps the model will finally be helpful, thanks to this additional intricacy. 

Perhaps the next run will provide some emergent conscious behavior. On the other hand, we’ll require an even more extensive network of GPUs operating continuously throughout the night to grasp the impact. 

9. Explainable AI is Simply Another Turtle in The Group.

Researchers in artificial intelligence have recently begun spending more effort attempting to explain precisely what the AI is doing, and explainable AI is simply another turtle in the group. 

Looking closely enough, we can see that these parameters originate in one specific data set section and are essential to the trained model. 

On the other hand, the explanations are often somewhat similar to the ones magicians give, in which they demonstrate a trick by doing another. 

“Why?” is an incredibly challenging question to answer. Gazing at the parameters of even the most basic linear models might leave you bewildered.

You might ask why not 0.043256, 0.7, or even something entirely out of the ordinary, like 411 or 10 billion, since the model suggests multiplying the annual mileage by a factor of 0.043255. When working with a continuum, every axis value may be correct.  

It reminds me of the previous model in which the Earth merely rested on a gigantic turtle. What was the position of this turtle? Upon an additional turtle’s back. And what about the one after that? At every turn, it’s turtles. 

10. Attempting Justice is a Challenging Feat. 

Leaving height out of the training set is an option, but your AI will likely find a way to identify the taller individuals and select them for your basketball team. Possibly the size of the shoe.  Maybe even reach out. 

Sometimes, the problems are deeply rooted in reality, and the algorithms can’t do much better, despite people’s dreams that asking an impartial AI to make a fair choice will make the world more equitable. 

11. Fixes Can Occasionally Make Things Worse. 

Is there any practical way to make AI fair? Many are adamant that AIs must adhere to specific, predetermined percentages while producing results. They update the algorithms and put their thumb on the scale to alter the results. 

However, once you have determined the desired result, individuals question the point of investing in training and data analysis. 

12. People Are The Real Issue. 

When the stakes are minimal, we usually have no problem with AIs. When you have ten million images to sort, you’ll be grateful if artificial intelligence can, on average, produce respectable results. Of course, problems and errors can arise.

The AI’s biases could be at the root of some of the troubles, which could warrant a 200-page thesis on hair-splitting topics. 

However, AIs do not constitute the issue. They are submissive and will follow orders. They can have their error messages hidden if they are too picky.

We can ignore the complaining result requesting further data if the training set fails to produce ideal results. We can just discard that outcome if the precision isn’t optimal. Returning to their duties, the AIs will give it their all. 

As a reader, I am incredibly grateful to you. If you want to read my next posts, which combine ideas from psychology, social issues, education, business, biographies, women, leadership, tourism, climate change, and many more, you should follow me. You may sign up for my email list if you want to receive notifications whenever I post a new article.