[jomyboy]

My earlier review was based on the free versions out there, which are not really reliable, clunky, and barely worth the effort. But once you step into the paid zone -- Boom! Mind blowing. The power and accuracy of AI are just next-level.

Forget memorising a thousand Excel and Word functions—AI’s got you covered. Need to code? No problem. I’m cranking out Python scripts in minutes that would leave seasoned devs scratching their heads and the best part? this thing never quits. Throw errors at it and it just keeps going until every last issue is crushed.

I’ve levelled up so much that I’m even building programs to convert voice to text and vice versa. Honestly, I’m beyond impressed—this tech is an absolute game-changer!

[Thad E Ginathom]

Originally Posted by RiderZone

Like any techbro he looks at all the world's problems, and thinks that the only way to solve them is through more technology. AI will solve war, AI will solve world hunger, AI will solve climate change. I do not know what that means, I don’t think Mo Gawdat knows what that means.

"Business Manager?" I strongly suspect that he is not so much a techbro as a marketing bro! It may well be that he doesn't know much tech at all: it certainly sounds like that.

I salute you! You must have worked really hard to not only finish, but cogently summarise and rebut the content of a book you hated reading. Well done!

I don't think I'd be able to cope with more than a few minutes of marketing man with big-tech-co brainwashing. Not even on paper

Tangentially... I remember a sketch by a very brilliant American comedian. She mention that she had worked for Google, and had loved working for Google. She knew that she loved working for Google because they kept on telling her that she loved working for Google. And if she had stopped loving working for Google they would have sacked her.

[searacer932]

Originally Posted by jomyboy

My earlier review was based on the free versions out there, which are not really reliable, clunky, and barely worth the effort. But once you step into the paid zone -- Boom! Mind blowing. The power and accuracy of AI are just next-level.

Forget memorising a thousand Excel and Word functions—AI’s got you covered. Need to code? No problem. I’m cranking out Python scripts in minutes that would leave seasoned devs scratching their heads and the best part? this thing never quits. Throw errors at it and it just keeps going until every last issue is crushed.

I’ve levelled up so much that I’m even building programs to convert voice to text and vice versa. Honestly, I’m beyond impressed—this tech is an absolute game-changer!

If I may ask, and if you don't mind, which paid version of the AI software do you use? I can largely see the below AI tools that seemed to have gained a bit of traction for single user purpose for their day today tasks

1. Open AI's ChatGP
2. Microsoft's Co-Pilot (Again the underlying LLM is Open AI's)
3. Perplexity (Uses multiple different LLMs)

[jomyboy]

Originally Posted by searacer932

If I may ask, and if you don't mind, which paid version of the AI software do you use?

I’m using the Copilot Pro trial because they gave me a trial version, but I don’t plan to continue with it.

[Samarth 619]

Originally Posted by RiderZone

A look inside the AI bubble

I’m part of a book club at work where we gather together every month to discuss something we’ve all............... that he thought his thoughts on AI are valuable enough to be compiled into a book is fascinating to me, but clearly this man’s infinite lack of modesty didn’t hinder his rise in the world of tech, it might even have helped.

Your point makes perfect sense, for long as we firmly believe that AI is not self aware, sentient, alive, its just like any other machine out there.
We, the human beings are also machines, and supposing that they could create an entire replica or teach an artificial system to behave like a human, what's stopping them? Please understand that a person can't write it all in a book, due to legal, employment related complications and controversies and other reasons.
Or suppose that everything Android is just a huge research project and its listening to their own users, is it tough to train an AI for the same? I think its probable.

It reminds me of a Google R&D employee who gave a speech in media that he was scared by how the AI was responding, and then Google sacked him.

Source: https://www.theguardian.com/technolo...ot-is-sentient

The Speech Transcript is given here: https://cajundiscordian.medium.com/i...w-ea64d916d917

Given the randomness of how AI behaves in conjunction with the violent history about human beings it has learned, could an AI actually take a destructive route? It will be everywhere, so will it collaborate with each other and create a power that might remove all digits from our life altogether?
Once again, I don't know, nor can I claim anything. But, just read what the speech transcript says.

Why are these guys at Google so damn shocked?
I'd be convinced he's a lunatic if he's alone, but as of today, he's not.

These guys are not MBA's, they're from the IT and data world, they know how computers work. It's not an easy job shocking them.

[SmartCat]

Originally Posted by Samarth 619

Given the randomness of how AI behaves in conjunction with the violent history about human beings it has learned, could an AI actually take a destructive route? It will be everywhere, so will it collaborate with each other and create a power that might remove all digits from our life altogether?

Even if one AI system becomes malevolent, it is fallacy to think that different AI systems will form an alliance against humans. As we have seen with generative AI, each system is trained differently and will have different 'values'. Hence, when they become self-aware, very likely we will have benevolent AI systems working with humans to keep malevolent AI systems in check.

I shall now use chatgpt to create a script for the above movie storyline/concept and send it to Warner Brothers!

[digitalnirvana]

Originally Posted by Thad E Ginathom

Tangentially... I remember a sketch by a very brilliant American comedian. She mention that she had worked for Google, and had loved working for Google. She knew that she loved working for Google because they kept on telling her that she loved working for Google. And if she had stopped loving working for Google they would have sacked her.

This gives me Gavin Belson and Nelson "Bighead" Bighetti vibes. Courtesy of Silicon Valley. Not sure if I'm bemused or creeped out. The world is stranger than we often imagine.

[NiInJa]

Originally Posted by RiderZone

A look inside the AI bubble

I’m part of a book club at work where we gather together every month to discuss something we’ve all read together. Our next meeting is scheduled for this coming Wednesday, and the book we’ll be talking about is “Scary Smart, The Future of Artificial Intelligence and How You can Save Our World”, by Mo Gawdat.

I am confident that the summary in this post is way better than the contents of the book itself Brilliantly written as always @RiderZone

Reading books -of authors who have dedicated their lifetime to something esoteric and then taken up writing to promote their opinions under the garb of imparting valuable 'gyaan' to general public- these days is becoming a risk, frankly. I hate to see unfinished books lying around in my house just because somewhere in the middle you realize that was not worth the time and efforts.

[Mr.Ogre]

Originally Posted by jomyboy

I’ve levelled up so much that I’m even building programs to convert voice to text and vice versa. Honestly, I’m beyond impressed—this tech is an absolute game-changer!

Originally Posted by jomyboy

I’m using the Copilot Pro trial because they gave me a trial version, but I don’t plan to continue with it.

That's really encouraging. Please don't mind asking, but did you achieve all that with Copilot Pro? If you are reaping the benefits, then why would you not continue.

I would be more than happy to pay for something like that which as per what you said actually makes life easier for people in technology. Would be really great if you could share the paid versions that you tried which helped you form that opinion.

[jomyboy]

Originally Posted by Mr.Ogre

That's really encouraging. Please don't mind asking, but did you achieve all that with Copilot Pro? If you are reaping the benefits, then why would you not continue.

Honestly, it’s pretty simple. You just throw out 'I want this' and 'I want that,' and boom—code comes flying at you from all directions. When you hit a creative roadblock, GitHub’s open-source playground is right there to mess around with. Been there, done that, got the T-shirt

There are some parts of AI that really matter to me, but I’d rather not dive into them in an open forum—too much noise, not enough solid insights.

[RiderZone]

So the book club meeting was yesterday, and it was quite interesting to see the opinions people gave. Important context, more than half of the people in the book club were from the marketing team, so it shouldn't come as a surprise that they were all ready to take Mo Gawdat's words on face value and were convinced that superintelligent AI is coming. I had to really be the downer guy and keep reminding them that even though we do not know what the future of AI is, this book and this man are the least trustworthy place to learn that answer.

I did not want to push my negative opinions too much however, I think my anger at the book was hard to hide, but I did a good enough job staying in control. I don't like the fact that this stupid little book was convincing enough to some people who've spent decades in the tech industry, even if it's in a non-technical department. They should know better in my opinion, but it's hard for me to put myself in their mindframe.

Ultimately this book is so bad that a large part of our discussion was about the philosophical implications of AI, wholly disconnected from the book itself, and that was good. Overall it was an interesting experience talking to people about the book, but it makes me a feel a bit more concerned about the broader public and their ability to see through obvious propaganda.

Originally Posted by Thad E Ginathom

I salute you! You must have worked really hard to not only finish, but cogently summarise and rebut the content of a book you hated reading. Well done!

I don't think I'd be able to cope with more than a few minutes of marketing man with big-tech-co brainwashing. Not even on paper

It was quite a task finishing this book, most of the sticky notes I've put in there are just random expletives and screaming. I had to push through because I did really want to engage with the discussion at the book club, but it was torture.

Originally Posted by SmartCat

Even if one AI system becomes malevolent, it is fallacy to think that different AI systems will form an alliance against humans. As we have seen with generative AI, each system is trained differently and will have different 'values'. Hence, when they become self-aware, very likely we will have benevolent AI systems working with humans to keep malevolent AI systems in check.

Agreed. I think it's very important not to let ourselves be scared by this tech hype. Us being scared is profitable for these tech giants, scared people make irrational decisions which may not always be in their own self interest, and this FOMO based marketing is nothing new.

Originally Posted by NiInJa

Reading books -of authors who have dedicated their lifetime to something esoteric and then taken up writing to promote their opinions under the garb of imparting valuable 'gyaan' to general public- these days is becoming a risk, frankly. I hate to see unfinished books lying around in my house just because somewhere in the middle you realize that was not worth the time and efforts.

I agree, it's quite a pain. Especially with the whole self-publishing business the quality of reading material out there is on a downward trajectory. That's why nowadays I just read and re-read the old books that I like, seems like a much better use of my time.

[shashanka]

Originally Posted by RiderZone

So the book club meeting was yesterday, and it was quite interesting to see the opinions people gave. Important context, more than half of the people in the book club were from the marketing team, so it shouldn't come as a surprise that they were all ready to take Mo Gawdat's words on face value and were convinced that superintelligent AI is coming. ....... the whole self-publishing business the quality of reading material out there is on a downward trajectory. That's why nowadays I just read and re-read the old books that I like, seems like a much better use of my time.

Thank you RiderZone for using plain english (mostly) in a post bashing a supposedly tech-dense region as AI. Being a 74 yr old nuts & bolts car buff it was a pleasant experience seeing the Emperor's new clothes being de-robed skillfully. AI means little to me. I prefer believing in Baba Vanga's prophecie and hope Mr. Chikumbutso's zero energy EV is proved viable. And thanks once again - we need a few more folks like you to balance out the landscape.

[Samurai]

Originally Posted by clevermax

In the example I mentioned, my prompt was almost 2 pages it had a lot of specifics on what needs to be built to solve my problem. The overall code structure and what it solves is unique. However, while deeply examining this code, one may find similarities in smaller segments that handle common routines within the larger solution, matching (very similar to) pieces of code that exist on the internet. This holds true even for code written by human programmers in some cases, as they too implement standard solutions for common programming tasks.

Now that I have a paid Github Copilot Pro license, I tried the following using o1-preview model:

1) I asked Copilot to review one page of my C++ code dealing with cryptography. I was very impressed with the results. It appreciated the good parts of the code and also pointed out some best practices I could adopt. It made me feel that if the code had bugs, it may have pointed out. It is just my feeling, it didn't point out any bugs yet. So I don't know for sure...

2) I gave a bunch of sql tables, and gave a detailed description of those tables and what kind of query I want it to write. It did a good job, but also told me to add a few indexes to make the queries fast.

This is where I got concerned. It was asking me for create indexes for columns that were already primary keys for that table. I wondered what am I missing, and I asked "Is there a need to create indexes where the field is already a primary key?"

It replied "You're absolutely correct. When a column is defined as a primary key in a relational database, the database automatically creates a unique index on that column to enforce uniqueness and improve query performance. Therefore, there is no need to create an additional index on a field that is already a primary key."

Then it gave a revised list of indexes without involving any primary keys.

I also did many other queries, and I am getting an impression that Copilot is like a very knowledgeable advisor, whose answers have to be verified every time. It can be very useful to experienced programmers who are capable of verifying the answers before using. But it can be very dangerous in the hands of rookies who are not capable of verifying the answers, and they would end up using it as is and end up in big trouble.

We know such things happen routinely even with stackoverflow, where people lift code chunks without verifying.

[BullettuPaandi]

Originally Posted by Samurai

a very knowledgeable advisor, whose answers have to be verified every time

I remember being reluctant to try these AI assistants at first, only to finally give in and try ChatGPT (don't remember the version; it was free and the default one we face when the website is entered and we choose to not login/signup).

I was trying to switch from Nobara to Arch Linux and had a couple of manual install attempts gone wrong for reasons I couldn't understand. So, I finally gave in to try ChatGPT, and the quoted excerpt from your post is exactly how I felt.

It was really good at explaining things- why we do what we do, what does each part of the command means, etc. I had the Arch Wiki and couple of other articles opened on other tabs, as I didn't completely trust the AI. And there were a couple of parts ChatGPT got wrong; but when confronted (just as you explain doing), it corrected the mistakes quite well, and also explained why it was wrong earlier and what precisely it has done in the new answer to fix it.

After a couple of weeks or so, I was looking for a text editor to take notes, that also allows hyperlink. But the important part is, it has to be linked within the text and not beside it (like typing [LINK] and entering the website beside in parentheses). And the link should only appear if I want it to, say by hovering over the text. As far as I know, the text file an editor operates on has to be rendered in some sort, to allow this. I was basically looking for more of a word processor without much of its added features to serve as a simple text editor.

Having had a mostly positive initial impression with ChatGPT, I decided to ask it. Especially given how complicated this is for a Google search query. I tried my best to be precise and peculiar about what I'm looking for. It basically swore that a text editor called Obsidian does that. I went through all the trouble on an unfamiliar Linux distro to install a snap product. Only to find out that it doesn't do hyperlink as such. Again, when confronted, ChatGPT was quick to point out that how my expectations make no sense given how text editor works- in a polite manner as it does- which to me was basically repeating my initial prompt to it back to myself.

So, I feel as if these systems try too hard to be knowledgeable. Sometimes, a simple 'I don't know' won't be a bad answer. But these systems seem to feel the need to try and make sense of even outlandish and unreasonable prompts, which'd only results errors inevitably.

[clevermax]

Originally Posted by Samurai

But it can be very dangerous in the hands of rookies who are not capable of verifying the answers, and they would end up using it as is and end up in big trouble.

You're right. GitHub Enterprise Co-Pilot (claims entire codebase understanding) isn't doing great, in general. It has mixed opinions and many teams are finding the tool no as good as how they expected it to be.

Understanding an entire large codebase and being able to act as a true co-pilot (or virtual co worker who knows everything about your project) - GH Enterprise Co-pilot (paid) is yet to reach there. In fact, we have the co-pilot support guys telling us that co-pilot isn't the right tool to fix your memory leaks in a large codebase, and they went on to say that it is ideally the job of an experienced developer who is well versed in that codebase.

Yes, it will be good for code suggestions & generations, review, and refactoring and so on - if you prompt with the right questions. But does it really possess the knowledge about the entire codebase? Well, not at a level where we expected it to be.

You cannot simply have an LLM understand the entire codebase with just a RAG on a vector database where it stored the chunks & their vector embeddings of the files of the codebase. Instead, one needs to build a complex graph of lexical, domain and semantic relations between many different entities in the codebase, such as such as functions, classes, dependencies, and domain concepts. Vector lookups can then be performed on different elements of this graph-based knowledge representation, enabling more accurate retrieval and contextual understanding for the LLM.

Building such a good knowledge graph requires human assistance, because only experienced people understand their codebase in which they’ve worked on for years. An automatic graph construction may not get you good results. GitHub Co-pilot automatically does this (indexing they call it), and the effectiveness remains limited by the lack of deep, structured context that only human expertise can only provide.

I would like to also point out that this is not the fault of the LLM behind the scenes, but lack of effectiveness of the lookup and retrieval system that helps the model to understand the entire codebase.

IMO, a useful and effective solution around this is yet to come. There are several solutions claiming this, but something has to get popular by user opinions.