[Samurai]

Originally Posted by ZenMaster

With what I am hearing about how good the AI copilot on GitHub already is, it wouldn't be far fetched to think that companies will no longer need "Programmers" and only mangers since the coding will be done by AI.

What you are hearing is wishful thinking. We are very far from replacing programmers with AI. This has been discussed in many threads ever since ChatGPT burst into the scene.

People who are making these statements don't realise that most programming is about deterministic output and not probabilistic output.

[Jaggu]

Originally Posted by ZenMaster

With what I am hearing about how good the AI copilot on GitHub already is, it wouldn't be far fetched to think that companies will no longer need "Programmers" and only mangers since the coding will be done by AI.

We wouldn't want to find ourselves in a future where, no one except the AI's know how to Code.

I saw a demo of what can be possible, I am not a techie so this is my understanding of the same.

Step 1: UI / screen layout was hand drawn on a sheet of paper. Scan it and put into UI AI tool. Voila 3 mts UI screen comes out as image, design which a UX guy would have taken about 4 hours done (estimated).

Step 2: Use the output of AI Tool was fed into to another code generating tool. 5 mts and code is ready. This would have taken about 6-8 hours.

The guys who did were not coders and they said setting up the tools took few hours since they were doing it for first time and figuring it out on the go. That was a onetime activity.

Now to connect this front end code to back end it will take some work or there maybe a tool which they were trying to figure. But in short UX designer and UI developer job was kind of eliminated, atleast it did not require experts. Second what a human needed hours was reduced to under 15 mts.

That is HUGE if you ask me.

[warrioraks]

Originally Posted by ZenMaster

With what I am hearing about how good the AI copilot on GitHub already is, it wouldn't be far fetched to think that companies will no longer need "Programmers" and only mangers since the coding will be done by AI.

We wouldn't want to find ourselves in a future where, no one except the AI's know how to Code.

That is far fetched. But what's going to happen is - programmers who don't use GENAI will be more or less redundant soon. Primarily because the productivity of someone not using these tools will be similar to a company manufacturing cars by hand. They will not stand a chance against companies manufacturing in state of the art factories.

[greyhound82]

Originally Posted by warrioraks

That is far fetched. But what's going to happen is - programmers who don't use GENAI will be more or less redundant soon. Primarily because the productivity of someone not using these tools will be similar to a company manufacturing cars by hand. They will not stand a chance against companies manufacturing in state of the art factories.

The technology is capable of generating codes with high accuracy and optimisation for specific SoC’s as the prior distribution / model for that kind of language can be easily learned from data.
Existing approaches mostly fails for hierarchical tasks like architecting a business solution and creative tasks like posing a hypothesis but openAI, meta and deepmind is working on solving these but with different approaches. We can expect it to be there within another few years and there comes research assistants for scientists, medical assistants for doctors( with multimodal data - symptoms as text, sound of cough, ecg data, a snap of your face, CT and MRI together to one system. I am of the opinion that doctors can be completely replaced by AI )

[warrioraks]

Originally Posted by greyhound82

I am of the opinion that doctors can be completely replaced by AI )

I am all for the power of AI. But even the makers of these advanced models are not claiming as much (AI completely replacing doctors).

It's your guess against mine but I don't see this happening even in the next 10 years.

[Tutenkhamen]

Originally Posted by 84.monsoon

It is an interesting proposal, but I have a few questions on the proposal - while answering them, I hope we get reminded of the true purpose of a forum like this.

For some of us who research topics and write long posts, the sheer reward is in knowing if it was useful or interesting to people. We often find ourselves checking, sometimes multiple tomes a day, how many views are there of the post, how many likes are there etc. Will AI bots reading and summarizing the post create any satisfacton for authors? Will it count towards the view count of the thread? Can they start liking threads? If so, on what basis?

Today, mods decide on the popularity of a thread and designate "hot threads" by looking at multiple factors like views, replies, likes etc. Would many AI bots reading, liking and summarizing the text count towards this process? How do we know if real readers read a post, or, if it is useful to them, if the bot takes over this process?

In many cases, what makes the forum interesting is the diversity and sometimes the divergence of ideas and opinions. Members can often go on for dozens of.posts respectfully and logically arguing contradicting opinions. An AI bot will likely summarize this in one sentance - "most readers have the opinion that X is true" - while in reality you need to come to your own conclusion on the topic, as it may be based on your own opinion combined with the rationale laid out in the thread. You can only come to your conclusion if you go through the nuances laid out in the entire thread.

• If we cannot know what our audience really wants or likes, can we continue to be relevant and viable as a community?
  
• Why would the members contribute if goes into a black hole and there is no measurable feedback?
  
• How would members reach correct conclusions - as the car is a very personal and emotive subject - without going into the details?

There are many more such questions, but hopefully we can stay sane and preserve this forum's uniqueness and relevance by keeping it human.

Thanks for the detailed reply.
First, I would have the summariser to act as an add on. Users can opt to use it but by default it’ll be opt-out for everyone. This is to ensure end users’ original experience doesn’t degrade.

We call it ramping a feature at LinkedIn. Ramp by x% and correlate it with few different metrics - m% change in mean time a user was on a thread, n% thread reactions, o% probability that a user would interact (in few different ways), etc.

Further, the summariser is a set of model(s) at it’s heart. Context specific fine tuning is required based on your intentions, as forum moderators, for how best to put content to a user which overall, saves their time spent on a thread but also peeks out all relevant content. This may also mean that in the same time a user was spending on this forum, they are now able to get hold on k% more threads.

This discussion also requires talking about various different facets, for example, where a user’s forum history can help personalising the summariser for them.

My meta point is new age tech helps in productivity. And if it is configured thoughtfully, I don’t think it can hurt. It can only potentially aid the audience and enrich their experience

[Dry Ice]

Originally Posted by SmartCat

Ola Krutrim (beta) is now open to public

Originally Posted by warrioraks

Very similar to Gemini, this model has already shot to fame because of some interesting responses.

It’s probably just a wrapper over OpenAI’s offering. Maybe that’s why they only compare it to Meta’s Llama 2. No comparison to the first mover ChatGPT? How can ChatGPT compare it to itself? But hey, who cares, the investors certainly don’t

Company says, it has been trained extensively on Indian languages. But, it still manages to screw up a simple hindi sentence! This is what you get if you literally translate Sure, here is the recipe of Dhokla from English to Hindi.

Funny, that it translates a much harder word like fermented to किण्वित but leaves a simple word like recipe alone which can be rephrased as बनाने का तरीक़ा, नुस्ख़ा etc.

[SmartCat]

Originally Posted by Dry Ice

It’s probably just a wrapper over OpenAI’s offering. Maybe that’s why they only compare it to Meta’s Llama 2.

Bhavish says it is not just a wrapper:
https://swarajyamag.com/technology/i...ast-indic-data

“This is not just a wrapper on some existing API. This is not just a little bit of fine-tuning done… taking an existing model and putting a little bit more of a data set into it. This is deep foundational work, starting from the science layer, changing the math and the algorithms of the models to make it more relevant for Indian languages, putting in the right mixture of data, and generating this outcome,” said Aggarwal.

No comparison to the first mover ChatGPT? How can ChatGPT compare it to itself?

But suspiciously, when somebody asked if it was based on GPT 3.5, this is what it said:
https://analyticsindiamag.com/why-ol...s-its-creator/



After this answer went viral, it no longer answers that question (I just checked):

But hey, who cares, the investors certainly don’t

$1 billion valuation for a probable tophat on GPT 3.5/4.0. Hmm!

[warrioraks]

Originally Posted by Dry Ice

It’s probably just a wrapper over OpenAI’s offering

That was my initial thought as well. But it seems they did some finetuning here for Indian languages. Which again is not very hard to do.
So the truth could be somewhere in between - 'this is a home grown deep learning model built from scratch' && 'it's a wrapper on OpenAI'.

Originally Posted by SmartCat

Bhavish says it is not just a wrapper:

Could be marketing talk. Till date, there is no research paper or documentation around their approach to build this model or the underlying model architecture.

Never the less, no reason to ding them without any factual information on what's under the hood.

[Samurai]

Originally Posted by Samurai

One day you quit your Big Tech job and start a company. You are no more sitting on the 10ft tall elephant. Now you are standing on ground or worse

Yesterday night when I was writing the above in another thread, I wondered if AI can generate this visual.

So I asked Co-pilot to generate the image with the following description: A man riding an elephant talking to someone on the ground.

After a very long time, the following image was generated:



He is talking to someone on the ground using a mobile phone??? Never mind the intent of my instruction that was totally lost.

[am1m]

Originally Posted by Samurai

Never mind the intent of my instruction that was totally lost.

Look at it this way. For a paid project/editorial where a designer needed an image of a man riding on an elephant talking to someone on the ground, what would work out faster/cheaper- spending a couple more iterations getting the prompt and image correct, or searching through image repositories like Getty images, finding a suitable image (might take some time) and paying for that individual image, or arranging a model shoot with an elephant?

[Samurai]

Originally Posted by am1m

Look at it this way. For a paid project/editorial where a designer needed an image of a man riding on an elephant talking to someone on the ground, what would work out faster/cheaper- spending a couple more iterations getting the prompt and image correct, or searching through image repositories like Getty images, finding a suitable image (might take some time) and paying for that individual image, or arranging a model shoot with an elephant?

Yes, I get your point. However, I was looking at the defaults it chose. A Caucasian guy in traditional Indian attire riding an African elephant using mobile phone.

And regarding the point about Getty images, let me repeat the question I asked before:

Originally Posted by Samurai

BTW, can you guys copyright these images as you own?

I recall Midjourney having major issues due to copyright issues.

https://www.analyticsvidhya.com/blog...t-infringement

[WindRide]

world is also moving towards a highly subjective area like content/forum moderation -

https://xdaforums.com/t/revolutioniz...rator.4571165/

[am1m]

Originally Posted by Samurai

And regarding the point about Getty images, let me repeat the question I asked before:

Yep, copyright infringement is always something to watch out for. Most of these AI image generators allow you to use the images you generate as you like, especially if you have an upgraded (paid) subscription. But I suspect a lot of them will run into issues. This is still an emerging area and the legal battles will determine how this evolves. Some professional research organizations have already started including "this content cannot be used to train AI models" in their copyright text.

As of now, the way I'd look at it would be based on usage. For a small, limited visibility project, I wouldn't worry at all. if I needed an image that was going to be front and center of a national campaign, conference, etc. I'd pay for completely original content with clear copyright. (I'd arrange to shoot with an actual elephant! Just an analogy btw, my job doesn't involve anything that interesting!)

[PreludeSH]

Are there anyone who just does coding or programming ? Like someone tells what exactly to do and you write code to implement it ?

You need to come up with a solution which you then implement in code ? If you are the first type, then you will be replaced