This is INSANE!!
People with no coding background are making $69k/month with apps or their SAAS.
You can also build an entire working MVP in just 2 days
And start making $ from the very 3rd day
🧵Here's How:
(Bookmark this)
AGI will not be a thing... anytime soon.
For all those people working in AI might have a different take on this, but here is mine:
I keep hearing strong claims that LLMs will replace many jobs end-to-end, and even get DMs from people asking if this part of their work can be automated or not and if so, please tell them how to do it.
Some are scared, and some are still trying to leverage it as their personal replacement and get the work done fast and with efficient accuracy.
I am having a hard time buying that based on my experience as an end-user of these LLM models, of these state-of-the-art models, and even local Chinese models are really that great.
Mind it, I use all these models daily to ship products, and still I am on this track.
My impression is that these models are powerful assistants, but they still struggle with long-horizon tasks and consistent execution, mainly factoring in context retrieval and hallucination.
I'll not get into the numbers for this post, but let me see if I can extend it into a thread with actual data figures and numbers.
Some things I keep noticing:
- They can be impressive on short tasks, but degrade over longer multi-step work
- They make basic mistakes that a careful human would not make
- They can sound confident while being wrong
- They need constant checking, which makes full autonomy feel unrealistic
- Reward hacking tendencies - It wants to achieve the goal - even if it means a suboptimal solution or cheating (by hardcoding variables or overfitting). Barely any design chops or long-term thinking mindset.
Because of that, I see LLMs evolving into something like a very advanced coding and knowledge tool, not a full replacement for people.
More like increasing productivity and raising competition in the workforce, rather than fully removing the need for humans in a job.
For people who are actually working in AI research or building these systems, what is your take?
I have a few questions linked below. If you can help me answer that, then that will surely make my day....
@0x_Kapoor This is actually insaneeee.
The fact that you built a clipping pipeline by using cheaper models and still getting the desired output says a lot about where thus "AI
replacing jobs" conversation is heading.
I did this way back with my minimax coding plan on a custom AI pipeline I built myself.
Saved like ~$20 for this same thing.
Here's the workflow:
> Give YouTube video link (Long form)
> YT-dlp fetches the video and downloads it locally
> FFmpeg extracts the audio from the video in .wav
> Using DeepGram CLI, it gets the transcription of the entire .wav audio file
> Sends full transcript with custom-written prompt to minimax API
> Minimax analyses and picks up the required clips as per the custom prompt.
> Using vision MCP + FFmpeg clips those timestamps that minimax gives.
> Clips are generated in .mov or .mp4 format
All this is done in less than 15 minutes per video.
Cost: 10x cheaper than any SOTA model.
I can literally build a clipping platform and sell a subscription for $20 a month, and this automation will take care of everything while you have a cup of coffee or have some free time with your girlfriend.
Would anyone like to see this whole pipeline in action?
Comment "Make" and repost this.
I'll make a detailed video showcasing this entire automation and even give away the entire thing for FREE with a small plus GIFT on top of this from my side>>>