[Gsynch]

Originally Posted by Thad E Ginathom

Before you doubt the words of some individuals that we know on the forum, preferring the marketing material of your favoured companies, you should recognise that these individuals work, or have worked before retirement, at high levels in the high-tech world.

I'd listen to them, rather than Nvidia, Meta, Google, MS, etc etc. any day.

Sir, as I said earlier, "with due respect" i expressed my opinion, I do respect others view who may know more than I do. AI being a relatively new buzz and still evolving, I think it's wise to disagree with each others point of view and still learn.

Originally Posted by Samurai

Simplistic? No, some of us have done deep dive into it before.

I am not questioning their advances in AI, but their frequent prophecies that jobs in industries they don't understand will vanish. Do you think Jensen and Sam has worked in IT services or Call center industries? They don't understand probabilistic outcomes will be not acceptable there. I have actually seen how LLMs fail to address the requirements that customers have. They won't pay for software that works most of the time. It must work all the time.

I am on the same page as you are on the "jobs will vanish" fear mongering part, I don't feel jobs will disappear, their nature however, will change.

On this, I will again quote Jensen AI will not take away your job, the person who knows how to use AI, will.

Thanks for sharing the post link from the other thread, makes lot of sense. But things are changing at a break neck speed around AI development and deployment. Quite a few job functions have become redundant in the marketing department that I work in, because of "Text to Image" Voice to prompt" services but the larger goal here is to drive efficiencies and reduce waste.

[NiInJa]

Ok, I have been doing some reading on AI lately, and here is my understanding rambled below. I'll try to put it in a simple way, like telling a story.



The Origins:
Since time immortal (i.e. the advent of computing and computers), humans have tried to make computers smart, in that sense, the real victory would be when computers could communicate with humans without sounding like computers. They should be able to make cohesive statements, stay true to the context, understand the tone of conversation and at times, even make mistakes like misunderstanding something and apologizing. Computers would need the ability to learn and understand, just like humans, before saying something sensible.

To make this happen, Humans tried to make something similar to a human brain, using computers. Let's call that brain as Artificial Neural Network or simply Neural Network.
For simplicity, a hypothetical computer creature using this 'brain' would be called a 'model'. Since it dealt with processing words, the creature was called 'Large Language Model' or LLM.

Now, A neural network in computers is lot of mathematics, statistics crammed with lot of scientific jargon which we will not go into. But, just know that this Neural Network is the starting point of everything related AI that we talk about.

The Problem:
As time progressed, Neural Networks (i.e. brain) made lot of advancements and so did LLMs. They could learn things on their own and give reasonable answers to humans. But, there was still one problem it would not solve.

All these creatures (models) were not doing a great job at making cohesive statements at lengths. Think about a guy/girl you go on a date where he/she can just go about rambling for hours, or, you sitting in a boring classroom where the professor is just talking continuously, or you are having a conversation with a friend after a long time, sitting in a bar and telling long stories.
These LLMs were just not able to do that. They would falter after two or three statements. Lose context and start sounding dorky.

The Breakthrough
That is until 2017, when a bunch of scientists at Google published a paper called 'Attention is all you need'.
It talked about a new type of Neural Network Architecture i.e. a new computer brain structure, which:

- had the ability to process input (words for example) in parallel, instead of sequence.
- had the ability to learn and adapt fast with better self learning capabilities that could understand multiple contexts at the same time (not getting into the jargon again for simplicity)

Both these qualities were not seen in any neural network architectures before.
They called this new type of computer brain as 'Transformer', simply because they liked the sound of it and their initial paper had an illustration of the Transformers.

The Transformer architecture, used in a creature named LLM, solved the problem of making lengthy conversations, rambling within contexts and giving out solutions based on what it had learnt. It had excellent next sentence prediction and accuracy.

In immediate future, the learnings from these models would go futher, meaning, they could take in not just words, but images, sounds and other forms of data (DNA sequence) as input and generate outputs. The Models would be called as Generative Pretrained Transformers or GPT

In 2018, humans literally created an architecture, a mathematical model, that would mimic the brain with a far greater accuracy compared to earlier models.

The Birth of Foundation Models
To put this theory of Transformers to test, big companies started implementing it in computers.

Now, To use Transformer in a Model, a computer needed:
- Fast parallel processing capabilities (*nVidia CEO smiling in the corner)
- Loads of training data (*companies having user data: Google, Meta, Microsoft popping champagne bottles. OpenAI, not just yet..)

Google put this into a computer which had parallel processing abilities, and came up with a functional model named BERT in 2018. The results astonished and scared them, so Google started incorporating it in their search engine silently without making a lot of noise.

OpenAI put this into theirs and came up with a model named ChatGPT around 2020, where they trained their model using Wikipedia and some free data available on the internet. Even with this, masses were blown away looking at what it could do. Later Microsoft invested in OpenAI, possibly giving OpenAI access to data that Microsoft owned.
Google came out of their shell and made their model available for general public too. Their product is now known as Gemini.

This is the point where Generative AI exploded on the face of general users.

Other companies started coming up with 'Models' which would use some form or the other of GPT or BERT. These models came to be known as 'Foundation Models'. They are used to (but not limited to):
- Text input to Text Output (ChatGPT, Gemini etc)
- Text Input to Image Output (Dall-E)
- Image to Image
- Text to Sound
... and so on

You can create your own Foundation model, or use the ones already available.
For example: Apple integrated ChatGPT instead of coming up with their own.

To create your own Foundation Model, you would need
- Loads of money
- Good legal team (?)

because, firstly, the hardware is exorbitantly priced. GPUs are expensive. You need lot of computing power to train a model
secondly, training data can be proprietary, you will need to pay for it or steal it as OpenAI is alleged to be doing that for ChatGPT.

The advancement
GPTs and these foundation models in principle, work exactly like humans and the brains they come equipped with. The training mechanism too, can be compared with humans.

A human brain samples data based on experience and makes associations that reside in his brain. Example: A baby looks at a ball for the first time, his dad says the word 'Ball'. Baby registers object's name as ball. Then, the dad throws the ball, baby looks at the act, associates the action of 'throwing' with the object 'ball'. He then 'throws' his dad's cell phone (an object he is not aware of). Dad says 'No', baby associates 'No' with not 'throwing' a 'cell phone' but 'Yes' to a 'ball'.

A computer model too works on similar logic. The 'ball', 'phone', 'act of throwing' all are part of training data, the attributes of the training data can be 'object name', 'action', 'color' etc. These attributes and and related information is stored in the form of vectors (or Matrix, an arrangement of numbers on which you can do operations as well)

So a computer brain will become smarter as it gets more attributes for association and more data for training. Sounds similar to humans learning ?

As processing power of computers increase, the ability of computers to process these attributes will keep on rising. ChatGPT initially had a attributes in order of millions and it would make some mistakes while framing output, now, the latest version of GPT has about 165 billion attributes on which it has trained itself. It has now become aware of the 'tone' of sentences and paragraphs, can write songs and compose music nicely. As it keeps getting trained on more and more data, it will get more and more accurate. That is to say, at some point in time, it will become smarter than some humans in general interaction as far as language models are concerned.

Common Pitfalls
People are using the word AI anywhere and everywhere these days. Don't get awestruck by that. Wait. Wait till they show actual, real results on real data. If someone has no idea about Foundation Models, Neural Networks and is making tall claims about AI, you have the right to look at him/her with suspicion

Not everyone can afford GenAI at this point, building own GenAI models for homegrown solutions requires lot of money and time. It is expensive. Training it may need access to data that nobody is willing to give you due to copyright or confidentiality. So if someone is saying we are using AI powered chatbots on our platform, you might want to have a look at the company's finances, the Data science, legal team they have. They might simply be selling small language model based chatbots which we all have frustrating used in banking apps.

My car has a voice command recognition, does not mean I should post that my car is AI enabled on linkedin.

Ending comments

Is AI really a threat to some jobs? Definitely Yes.
Will AI replace humans completely: No, but will drastically reduce workforce in some sectors.
Is Generative AI simple to implement: No, it involves lot of money and resources, legal stuff and all. People have already died in mysterious ways.
Should humans be worried ? Yes

Afterall, this GenAI thing comes very close to human brains.
In a world full of gullible humans who do not have the ability to think and act properly, this Generative AI thing has the ability to do some real damage. Just see how videos are morphed to perfection or how audio clips were 'generated' to spread false propaganda during recent elections.

[searacer932]

All these talk about AI gained momentum around Nov/Dec 2022 when Chat GPT was introduced to the world and Chat GPT is a generative AI i.e. it can generate texts primarily in response to a prompt.

Lot of people are saying that AI can do that and AI can do this. Are there any real world use cases / business problems that have been solved in any industry by any form of AI? If so, please do share it. I am struggling to get this data.

[Samurai]

More than job losses, I think the following should be our chief concern:

https://www.reddit.com/r/ChatGPT/com...r_homework_is/

If Gen Z was cursed with smartphones, Gen Alpha is cursed with LLMs.

[DigitalOne]

Originally Posted by Samurai

More than job losses, I think the following should be our chief concern:

That's sad. Job losses will happen not because AI is becoming smarter but humans becoming dumber.

[Everlearner]

Originally Posted by Samurai

More than job losses, I think the following should be our chief concern:

If Gen Z was cursed with smartphones, Gen Alpha is cursed with LLMs.

Originally Posted by DigitalOne

That's sad. Job losses will happen not because AI is becoming smarter but humans becoming dumber.

Going forward, the more important question is what we use AI for rather than how to use AI. Once the tech becomes mainstream, it will be difficult to control it's usage, like YouTube for example.

[DigitalOne]

Originally Posted by Everlearner

Going forward, the more important question is what we use AI for rather than how to use AI.

Off topic - I don't want to do AI to do even my laundry or dishes. I do a lot of my thinking or listen to podcasts when I am doing mundane chores around the house. It keeps one both physically and mentally active.

[am1m]

Originally Posted by Samurai

If Gen Z was cursed with smartphones, Gen Alpha is cursed with LLMs.

Attachment 2713601

Actually, that sample could use a lot of editing. Perhaps even a ChatGPT summary

This is certainly a issue though, I see often in my line of work. Certainly a lot more people using ChatGPT to generate content, but when used as is, it becomes very obvious that it's AI generated. I had a sample come in for review that was supposed to be a pitch to a big supplier. There was no way a company at that level would fail to notice the bland tone and filler content, and not be insulted that we didn't even care enough to prepare a decent pitch summary! And we'd certainly not get their business. Sad part is, the AI generated summary came in from someone pretty high up. This person subscribes to the view that ChatGPT can basically do anything related to business communication. Hard task convincing them otherwise!

[clevermax]

Originally Posted by Samurai

If Gen Z was cursed with smartphones, Gen Alpha is cursed with LLMs.

I'm GenX, and I'd say, I'm truly blessed with LLMs

One example - I crafted a 2 pager prompt and threw it to GPT4, and it gave me the software design and starter code for one piece of software we wanted to develop. It just took 2 hours from nothing to running that initial codebase spanning across multiple classes, brilliantly addressing the exact needs. Without that, the devs in my team would have sat on the software design task for a week, and would have come up with something inferior for sure.

Of course, foundational knowledge in software engineering helped immensely in crafting that prompt in the first place.

As for the Gen Z people in my house (My daughters), GenAI usage is strictly controlled. In any case, they do not get to ask ChatGPT to solve a problem for them. By default, those domains are blocked on their devices. However, they are encouraged to ask questions to help them think towards a solution, under my supervision.

[Pontiac]

People thinking AI is just about speech and text just shows they have not gone beyond chat GPT. People don’t realize there’s more to it as not all companies are vocal about how they are implementing the same in their field. Countries are ramping up their investments on AI in Military, just to site one of the applications.

https://time.com/6961317/ai-artifici...tary-spending/


As for speech itself and the advancements there, I would strongly suggest one to check out the Gemini live and engage in a voice conversation with the model. You would be surprised to know how far it has gone and in such a short time.

[amol4184]

Originally Posted by Samurai

More than job losses, I think the following should be our chief concern:

https://www.reddit.com/r/ChatGPT/com...r_homework_is/

If Gen Z was cursed with smartphones, Gen Alpha is cursed with LLMs.

Attachment 2713601

Originally Posted by DigitalOne

That's sad. Job losses will happen not because AI is becoming smarter but humans becoming dumber.

Idiocracy movie but just that its being played out in real life.

[One]

Can't understand what the hullabaloo about ChatGPT is. All-knowing entities called "Spouses" and "Teenagers" have existed since time immemorial....

[Gsynch]

> social media is poison to the human brain, pollutes status signaling mechanisms
> AI consumes all social media
> AI proceeds to poison social media with a flood of rubbish interfering with brain's status check mechanisms
> social media unusable, saves human brain
> AI is good :-)

I have a philosophical view that AI content is net good by breaking the stupid status games being played in social media world. Humanity's ultimate play against its own monkey brain.

[Samurai]

Originally Posted by clevermax

One example - I crafted a 2 pager prompt and threw it to GPT4, and it gave me the software design and starter code for one piece of software we wanted to develop. It just took 2 hours from nothing to running that initial codebase spanning across multiple classes, brilliantly addressing the exact needs. Without that, the devs in my team would have sat on the software design task for a week, and would have come up with something inferior for sure.

Is it really designing or stealing from the best without giving credit? How do you handle the copyright issue that would come from this practice.

[RaviK]

Originally Posted by am1m

...Sad part is, the AI generated summary came in from someone pretty high up... Hard task convincing them otherwise!

A higher up giving few AI prompts, noting few points and gurgling them out to business/technology folks ?? The disease is more widespread than you think

At times, convincing/correcting them becomes another bandwidth consumer, on top of actual solution creation timelines.