Nice I'm on the same path rn. Been watching online lectures and practicing on the weekends. Still going through the fundamentals but I'm liking it so far. How long did it take you?

1.5-2 years

Atp im literally just chatgpting all my code even editing it jus f*** it up a little then throw the code back in. Software engineering is cooked as f***

im starting to rly think like a data scientist

machine learning is life. f*** deep learning

ive watched like 3 videos trying to understand VAEs and im still confused

Machine learning is the only way to stay ahead. Learning how to make the AI smarter is where you can get jobs now. Code monkies are officially extinct.

I actually want to learn how to develop and use the AI systems people are creating. The future is learning how to optimize the AI models to their best performance.

Long story short it doesn’t matter what you are coding anymore.

What matters is how you are doing it.

arxiv.org/pdf/1906.02691

Here’s the research paper from the creator themselves of the technology . I’m sure you can use DeepSeek as a research assistant to help explain some of the technical aspects of the technology.

I use DeepSeek whenever I’ve exhausted my own effort into learning or trying to solve something

(Usually I ask for hints where I am stuck at)

I couldnt disagree more ngl. Ive never found an ai that can solve problems on its own. It still needs a lot of human oversight. For coding its only good for simple template code, it never gets it right the first few tries unless youre working on something very simple. And people who cant code are COOKED if they think they can just use deep learning models for coding

Deep learning is a subset of machine learning that contains modern "AI". AI is a huge bubble the likes of with we havent seen since the crypto craze or maybe pre dot com bubble. AI is as useful as its able to solve problems, which its mediocre at. Everyone is assuming that deep learning will continue to evolve rapidly which just isnt true.

People will try to replace humans with AI. When it inevitably fails all the people who dedicated themselves to ai are shafted. If it DOES continue to advance then it will replace the people who develop it. Its a lose lose for everyone except CEOs and stock holders.

I understand people want to try to use ai to replace labor. Ive worked on multiple teams that tried this and i myself have developed ai models for work. It fails every time if u dont have intimate knowledge of your domain beyond AI. And even then the answer is often "fuck ai lets just do things the old way"

Yea i am familiar, i understand everything but the reparameterization trick

I use ML/deep learning for work. But now i dont have a job. Im actually realizing now that not only is math in general not interesting to me, i generally dont want to return to technical work, not because its hard, but because i genuinely think itd a bad idea and a waste of time

Being human>>>everything else

F*** the job, f*** the computer, go outside, find love, be happy

As much as i wanna do the right thing im broke asl and if i gotta return to some bullshit math job im down

What do you think about DeepSeek’s new update on learning compared to other models? It seems to be outcompeting all of the AI models in America .

The Chinese are engineering ai on a different paradigm compared to the ChatGPT like models. And outperform them on reasoning skills while using less power

And here’s a video of physicist explaining how DeepSeek in particular is very good at reasoning when you prep it with a good foundation of information.

These American AI algorithms are s***.

we gotta move past the denial stage and get crazy with claude / mcp servers/ agentic engineering
Ive literally spent the last week building an app from scratch with s*** i never used before

I think MCP servers are gonna be obsolete soon. Honestly, everything we’re doing now will be obsolete come next year. There’s been rumors about recursive self improving models, but this will probably be a reality before 2030 arrives. Hell, probably before then.

Humans will eventually become the bottleneck. Nobody really knows what’s going to happen in the future, and we’re all trying to create workflows and techniques to optimize for it, but I think most people are going about it in the wrong way.

I don agree humans are the bottle neck though, the tools are
You are basically saying you can one day type in a prompt "hack the aws servers and bring it down"
Or ask it to make an even better model that can do it

Theres always going to need to be a human to guide it and create solutions