AI gets you 80% of the way there.
I used it to scaffold a TCP-over-UDP stack: sliding windows, congestion control, all of it.
Then it started hanging.
Send window full.
App closed.
FIN never sent.
Session deadlocked.
Why?
AI tied FIN emission to can_send().
Flow control was gating lifecycle progression.
The close signal existed, I just wasn’t polling it.
No autocomplete fixes architectural coupling.
So I went old-school:
• Separate flow control from lifecycle
• Add a drain phase
• Move state transitions outside congestion predicates
Deadlock gone.
AI ships code at insane velocity.
First-principles thinking ships systems.
AI + fundamentals?
Dangerous combination.
@flipkartsupport All you guys do is just keep postponing, again and again and again. You do not care about anyone. The amount of mental torture I have to face just because of this one order I can't explain in words. YOU GUYS ARE WORST!!!!!!
Now I filed a new return request after the previous request was cancelled automatically after a few days. The same pickup executive came and again rescheduled it as the laptop was not charged. Like he did not wait even 60 seconds.
Perks of being in a Tier-3 college 🤡:
1.Surrounded by low ambition you’ll start lowering your own bar too.
2. Life’s ultimate mission here: get a banda/bandi.
3. “Skills?” – unheard of. Semester exams are the only religion.
4. Attendance ka Olympic-level competition hota hai.
2022: ChatGPT will take over humanity.
2023: We are close to AGI.
2024: Software developers, beware!
2025: Sorry for the hype...
------
The marketing hype is dying down, and the AI bubble is going to burst.
GPT-5 proves how far we are from AGI.
And the hiring freeze is starting to thaw.
Good riddance.
In 2024, I thought LLMs would quickly replace competitive programmers.
But after reading their papers...
I have a feeling...
That AI companies are lying to us.
Why?
1. All the models have similar generation algorithms.
2. They first generate 50-100 candidate solutions.
3. They then submit these solutions on the contest platform.
4. The platform TELLS them the failing test cases.
5. The solutions are mixed and matched till all test cases pass.
6. Mysteriously, none of the models are released to the public.
Here is what I believe:
The models are achieving these scores with human intervention.
These papers are marketing stunts, with heavily inflated numbers.
If someone from Google or OpenAI wants to explain, I am willing to listen.
Paper links:
Google MLEstar (Kaggle): https://t.co/24RM7B1JRt
OpenAI o3 (Codeforces): https://t.co/d1Skk2Sqkp
AlphaCode (Codeforces): https://t.co/00WuB5NhwH
AlphaGeometry (Olympiad): https://t.co/wh7U6Se8aT