neural nets.

unemployed | prev: @ai4bharat @cisco(returning jan '27)

- (35k users)
- (7.5k users)
- (4.5k users)

[neural nets. @cneuralnetwork]

Every paper begins as an idea. And ends at a deadline.

https://t.co/zywXhEg3r3

[Pulkit @puulkit]

I built an open-source experiment so people could see how it worked learn from it and build on it. my posts and repo are still public. I’m here answering for what I actually did!!

You can disagree with the project. but dragging my company (even my colleagues) into a personal project makes no sense.What exactly does my employer have to do with this?

And the threats and racist comments about Indians that followed? What part of that is criticism of my work? (If you think these are just the comments, no. Michael himself was also racist. I'll attach the screenshot below)

If making something open source makes me a thief, then I’m robin hood and claude is my partner in crime, I guess :)

But seriously if you think I copied something, show exactly what I used as a reference. Even though I removed Lexington’s templates, you might want to refer to the forks. Making a repo public doesn’t automatically make everything in it fair game, but calling me a thief doesn’t prove your case either

When I realized I’d made a mistake by including their proprietary media, I removed it. I still own that mistake

And I still stand by my point AI is changing how software gets built. If your entire livelihood depends on selling generic UI templates that's going to be a difficult business to sustain in the coming years. that doesn't give anyone the right to copy your work but it doesn't mean we should pretend the industry isn't changing either

I’m happy to discuss the actual work. My nationality and workplace have nothing to do with that discussion

PS: this is dictated by Wispr Flow not AI written🙂

[neural nets. @cneuralnetwork]

check them out :)
https://t.co/r9nJcsMV2l

[neural nets. @cneuralnetwork]

I once spent hours debugging an AI workflow, only to realize a single number was incorrectly extracted from a PDF. Turns out, parsing accuracy matters a lot.

Just came across @nace_ai's Drex, which checks extracted data against the original document and corrects errors.
Already #1 on the parsing leaderboard.

[neural nets. @cneuralnetwork]

a friend of mine was recently applying for jobs.
most of their experience was in defence projects, so they couldn't publicly share their work. when a founder asked for proof of work, they even offered to solve any assignment to demonstrate their skills.

then came the stipend: ₹10,000/month. in Mumbai.
my friend politely declined the assignment altogether.

imagine expecting someone with actual industry experience to prove themselves for ₹10k/month in Mumbai. what do y'all think about this?

[neural nets. @cneuralnetwork]

@trikcode @nace_ai yess

[neural nets. @cneuralnetwork]

I once spent hours debugging an AI workflow, only to realize a single number was incorrectly extracted from a PDF. Turns out, parsing accuracy matters a lot.

Just came across @nace_ai's Drex, which checks extracted data against the original document and corrects errors.
Already #1 on the parsing leaderboard.

[Nace AI @NaceAI]

Introducing Drex for Document Processing 📑

Decision-powered parsing that catches errors before they reach your agents.

→ #1 on our parsing leaderboard
→ Beats GPT-6 Astra, Claude Opus 4.7 & every parser tested
→ 38% faster corrections on RealDocBench

Try it: https://t.co/eTUlO9j0wU

[neural nets. @cneuralnetwork]

@NaceAI Used the application. It's actually very awesome

[neural nets. @cneuralnetwork]

@trikcode @firefliesai yes!!

[neural nets. @cneuralnetwork]

i've always thought launching one product is stressful enough. imagine launching TEN at once.

sounds like absolute madness, but @firefliesai somehow made me question that. they've basically turned their AI notetaker into an AI teammate that handles meetings, emails, voice typing, scheduling, search, and more.

the interesting part? instead of juggling 10 different tools, what if you just had one AI teammate doing it all? not sure if this is genius or insane, but i'm genuinely curious to see how it plays out.

[neural nets. @cneuralnetwork]

@trikcode the approach is actually innovative

[neural nets. @cneuralnetwork]

check them out :)
https://t.co/lyPLhcTSlV

[neural nets. @cneuralnetwork]

i've always thought launching one product is stressful enough. imagine launching TEN at once.

sounds like absolute madness, but @firefliesai somehow made me question that. they've basically turned their AI notetaker into an AI teammate that handles meetings, emails, voice typing, scheduling, search, and more.

the interesting part? instead of juggling 10 different tools, what if you just had one AI teammate doing it all? not sure if this is genius or insane, but i'm genuinely curious to see how it plays out.

[neural nets. @cneuralnetwork]

ML papers become a lot easier to read once you understand these 5 symbols:

∇ (gradient): which direction should the parameters move to reduce the loss?
𝔼 (expectation): what's the average value over all possible outcomes, weighted by probability?
⊙ (element-wise multiplication): multiply corresponding entries, not entire matrices.
‖·‖ (norm): how big is a vector or error?
argmax: which input gives the highest value?

[neural nets. @cneuralnetwork]

@aidancramer Europe mainly!

[neural nets. @cneuralnetwork]

check them out :)
https://t.co/u3YZURD8Az

[neural nets. @cneuralnetwork]

a friend was job hunting recently and spent more time tweaking resumes and filling out forms than actually preparing for interviews.
AI Apply would’ve saved her weeks. It writes the resume for each job and handles the applications for her

wish she’d known about it sooner

[neural nets. @cneuralnetwork]

some of the most influential papers in AI/CS are surprisingly short.
papers you can read in one sitting:

MapReduce (2004): 13 pages, how Google processed massive datasets across thousands of machines.
The Google File System (2003): 15 pages, how distributed storage works at scale.
Adam (2014): 9 pages, the optimizer behind countless deep learning models.
LoRA (2021): 8 pages, fine-tuning large models using low-rank updates.
Dropout (2012): 2 pages, a simple trick to reduce neural network overfitting.
A Mathematical Theory of Communication (1948): foundational information theory, but substantially longer.

[neural nets. @cneuralnetwork]

a friend was job hunting recently and spent more time tweaking resumes and filling out forms than actually preparing for interviews.
AI Apply would’ve saved her weeks. It writes the resume for each job and handles the applications for her

wish she’d known about it sooner

[Aidan @aidancramer]

We just killed job hunting.

Introducing AiApply: a career agent, in your pocket.

Tell it what you want. It scans millions of jobs, skips 99.9% of them, and helps you apply to the few that perfectly fit, with a resume and cover letter written for each.

• No Spray and Pray. Our matching learns from millions of applications and interview outcomes.

• Accurate. Every application uses your real skills and experience. Nothing invented.

• Swipe to Apply. Every swipe teaches your agent.

• Full Control. Review every application, or put the strongest matches on autopilot.

• Already Applied. Other apps alert you when the right job goes live. AiApply has already applied.

Link below.

[neural nets. @cneuralnetwork]

@electro_pppp D

[ppp. @electro_pppp]

Be honest, how did u choose your engineering branch?

A. Genuine interest
B. Placement opportunities
C. Rank decided it
D. Family suggested it

[neural nets. @cneuralnetwork]

Average user writing their research papers after they find out https://t.co/zywXhEg3r3 has MCP support for Claude Code, Codex, and all your coding tools

[neural nets. @cneuralnetwork]

one idea that changed deep learning forever: attention.

2014: Bahdanau introduced soft attention
2015: Luong refined attention mechanisms
2017: Transformers made self-attention the foundation
2020: Linear attention tackled quadratic complexity
2022: FlashAttention made exact attention faster and more memory-efficient

which breakthrough do you think had the biggest impact?

[neural nets. @cneuralnetwork]

@trikcode let's go!
nice launch

[ppp. @electro_pppp]

A MrBeast clipping campaign was offering creators $1,000 for every 1 million views their clips generated.

And meanwhile i used to think uploading a video was the end of it.
If you've got a demo, webinar or any long video lying around, Kaleidly can turn it into short clips and posts without any of your intervention.

[neural nets. @cneuralnetwork]

5 things I wish someone told me before I started doing research:

1. Read papers backwards. Start with the results, figures, and conclusions. Only dive into the methodology if something interests you.
2. Reproduce before you innovate. Implementing one paper teaches you more than passively reading 20.
3. Keep a research log. Document your experiments, failed hypotheses, and observations. You'll forget more than you think.
4. Use coding agents for research. Tools like https://t.co/vcv1vDMWRC have MCP support, so you can connect your research notes and tools directly to coding agents instead of constantly switching between reading, note-taking, and coding.
5. Negative results are still results. Understanding WHY something failed is often more valuable than getting something to work.

what would you add?