Artificial Intelligence thread

O

Overbom

Brigadier
Registered Member
tphuang said:
so question for fellow members, I'm going to have an upcoming episode on AI regarding to just the infrastructure part of things. But after that, it gets a little harder. I have used a lot of AI so plan to do a Q&A on that. But I need people to actually give me some questions to answer, so if you can reply to me and pm me with some questions/topics to explore that would be great.
Click to expand...
Maybe something about what kind of infrastructure SMEs could use for AI that needs to be ran locally (sensitive data).
How much would that investment cost (hardware, potential software costs, implementation, maintenance), and how would that process look like


Techniques or ways to improve reliability of AI systems (prone to hallucinations) in a production environment.

Your opinions on why MoE architecture is getting so popular lately



Small note: would love if you could mention some of the lesser known Chinese AI companies and their products which are not known in the West



tphuang said:
Also, I'm seriously weak on the part with regard to training/building models and transformer architecture, so I'm looking for someone who can talk about that. If you know this stuff, please PM me.
Click to expand...
Is it even worth/feasible for new small companies or individuals to train new models?
I think fine-tuning is far more relevant and more widespread for application building



Good luck with your AI episode. Please be sure to link it here, will def listen to it
 
Engineer

Engineer

Major
tphuang said:
so question for fellow members, I'm going to have an upcoming episode on AI regarding to just the infrastructure part of things. But after that, it gets a little harder. I have used a lot of AI so plan to do a Q&A on that. But I need people to actually give me some questions to answer, so if you can reply to me and pm me with some questions/topics to explore that would be great.
Click to expand...
Close to a year ago, your stance on AI taking over software development was that it would happen in the distance future. What's your take on this now? How did you think AI change the nature of software development and what would be the impacts to the white-collar jobs moving forward?

Aside from DeepSeek, Chinese models are not well-known in the West. What models are available for us living in the West and which particular (with API access) do you recommend?

tphuang said:
Also, I'm seriously weak on the part with regard to training/building models and transformer architecture, so I'm looking for someone who can talk about that. If you know this stuff, please PM me.
Click to expand...
The people who could successfully build and train models belong to a rare breed. Among them, the people who could communicate their intuitive understanding to laymen are even fewer. My suggestion would be to watch videos from 3Blue1Brown and Welch Labs.
 
tphuang

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
Engineer said:
Close to a year ago, your stance on AI taking over software development was that it would happen in the distance future. What's your take on this now? How do you think AI would change the nature of software development and impact the white-collar job market moving forward?


The people who could successfully build and train models belong to a rare breed. Among them, the people who could communicate their intuitive understanding to laymen are even fewer. My suggestion would be to watch videos from 3Blue1Brown and Welch Labs.
Click to expand...
well the first one I can answer now. It's pretty clear that at least at this time, software developers are the first industry to get automated away, especially junior devs. This is a huge problem. I will talk about more I guess in Q&A
 
O

Overbom

Brigadier
Registered Member
I maintain my stance that by end of 2026 we would have something similar to AGI.

Maybe not real AGI, but for majority of common people it would look, function and act as AGI.

Can you also talk about the progress to AGI in your episode
 
C

chlosy

Junior Member
Registered Member
european_guy said:
This costs 6000 USD!!!


I was confident that effective and useful AI in office work will come before AI on factory floor....but now I am not so sure anymore.
Click to expand...
Looking forward to the wash dishes, chop food, cook meals, load laundry, hang laundry, fold laundry, sweep floor video- or is that already achieved?
 
P

PCK11800

Just Hatched
Registered Member
tphuang said:
well the first one I can answer now. It's pretty clear that at least at this time, software developers are the first industry to get automated away, especially junior devs. This is a huge problem. I will talk about more I guess in Q&A
Click to expand...

Throwing in my two cents as a software dev myself:

- Front end devs are screwed. All our internal tooling, main app and our brand new customer facing portal is built pretty much 90% by AI. JS/TS + React apps and websites, and web-based UI in general is a solved problem. LLMs are excellent in producing short snippets of code with a defined parameters and goals, which fits widgets and UI components to a tee.

- Backend devs like me are still chugging along. The biggest hurdles and challenges with writing backend code is dealing with how that specific backend is structured and how the business logic works. It's unique for every backend there is, especially when it gets more and more specialized, with different requirements and modules and what not. You need to feed in essentially the entire relevant parts of the backend (if possible), and then test and review whether or not the changes are robust and correct. All these still require manual input. Also god help you if you want to integrate a less known new payment processor or libraries, the LLM will just hallucinate code from thin air.

- The really complex software engineering, such as writing graphic pipelines and renderers, compilers and OSes, are safe and sound. Try asking a LLM to update your personal OpenGL renderer written in C to handle multi-textures and see how well that goes (spoiler, it doesn't).
 
tphuang

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
PCK11800 said:
Throwing in my two cents as a software dev myself:

- Front end devs are screwed. All our internal tooling, main app and our brand new customer facing portal is built pretty much 90% by AI. JS/TS + React apps and websites, and web-based UI in general is a solved problem. LLMs are excellent in producing short snippets of code with a defined parameters and goals, which fits widgets and UI components to a tee.

- Backend devs like me are still chugging along. The biggest hurdles and challenges with writing backend code is dealing with how that specific backend is structured and how the business logic works. It's unique for every backend there is, especially when it gets more and more specialized, with different requirements and modules and what not. You need to feed in essentially the entire relevant parts of the backend (if possible), and then test and review whether or not the changes are robust and correct. All these still require manual input. Also god help you if you want to integrate a less known new payment processor or libraries, the LLM will just hallucinate code from thin air.

- The really complex software engineering, such as writing graphic pipelines and renderers, compilers and OSes, are safe and sound. Try asking a LLM to update your personal OpenGL renderer written in C to handle multi-textures and see how well that goes (spoiler, it doesn't).
Click to expand...
it's hard to argue with that.

I think there are certain development which are easier like python coding, which are quite easily replaceable. Or at least you don't need junior devs to do. I mean I would see it as productivity gain tool.

There is just less junior devs needed, which means less mentorship and that's not great for the next generation.
 
You must log in or register to reply here.
Top