Akash Kundu

@ariG23498 Unfortunately I'm in Cape Town atm. Will miss this one. Hope it goes well! I'll try to make it to the next one!

[HF ML Club India] I am very excited to announce our very first in-person meetup. We are partnering with the kind folks at Google to bring a code first session on Gemma 4 using Transformers. Date: 18th April 2026 Venue: TBD (in Kolkata) -- any suggestions on a good coworking space would be appreciated. More information in the coming days! Stay tuned.

This is amazing. Deliberate on anything!

Introducing Alchemyst AI. Finally a context backbone for your company that "just works" with ANY AI agent that you throw at it. From regular conversations to remember to thousands of documents you want to refer, we got your AI agents covered.

@cneuralnetwork You should read about sycophancy. Funny how I thought it was only applicable to LLMs

how i got outsmarted by an uber driver today the fare was let’s say an amount x. as soon as the driver accepted the ride, he called me and asked if i could change the payment method from in-app upi to cash, so i could pay him directly via upi instead. it didn’t feel suspicious at all. honestly, it seemed pretty normal, so i agreed. the ride itself lasted about 1.5 hours, and throughout it, he was probably the nicest driver i’ve ever met. he agreed with everything i said. political opinions, random takes, anything. every argument i made, he supported it fully. at the time, it felt great. in hindsight, it felt calculated. you even supported my own points. one of my points was that drivers should not be paid extra if they are already getting the due amount at the correct rate. you agreed with that fact too. nodded along. reinforced it. validated it. i felt intellectually undefeated. and then what the hell. suddenly the tone flipped. it was basically this: i will agree with everything you say so that you like me. then, if you don’t have the courtesy to respect my good behaviour, i will force you to pay an extra hundred rupees. not in those exact words, of course, but that was the message. pay x plus 100 rupees. looking back, it was such an elaborate scam. one and a half hours of psychological alignment, strategic agreement, and social engineering just to extract one hundred rupees at the end. annoying. yes. impressive. unfortunately, yes. i kind of hate how much i loved it. [formatted by gpt]

Can I call myself a researcher now?

I'm going to Cape Town!

@ariG23498 "just use uv"

`pip install panda`

@ariG23498 Your call honestly. Depends on the type of crowd you're trying to find. I think floating the form again may bring in more people, it's just logistically harder.

@Akash190104 Do you think floating the form again for others to participate makes sense? OR keeping the same participants is enough?

Thinking of a meetup after a long haitus. Will email concerned folks with the updates.

@EsbenKC What's wrong now?

Uninstalling chatgpt and not gonna use their API anymore

@opinion Interesting.

QUIZ: Are you an AI Doomer? Or maybe you’re an Accelerationist who can’t wait for machines to out-think us? 👀🤖 Discover what AI future you actually want: bloom.bg/4nQi5PZ

@AnuranRoy @curlysaarthak @chetandoesai Hehe...

@curlysaarthak I already follow @Akash190104 and @chetandoesai

twitter hack: follow all NeurlPS selected researchers

@AnuranRoy @curlysaarthak Lmao I remember.

@curlysaarthak I remember patching Gradio, PyMongo, and Pydantic - broken dependents. @Akash190104

ml engineers please reply have you ever fixed a broken codebase of a library you want to use? how much time did it take?

@0xrguha Ffff

So apparently ants can fuck up your entire macbook motherboard.

@impoulav Damn Lessgo

As my current ASUS laptop loves dying on me for 20 times a day, pre ordered this cute thing on my birthday last Sunday 👀 Can't wait to get my hands on this 😋

@cneuralnetwork use uv.

real talk are you a conda type of guy or pythonvenv kinda guy

@cneuralnetwork There's no reason to use pip when there's uv.

am I the only person who still uses pip install

Went on hibernation for a couple weeks. Came back to inform that neurIPS scores for my paper is 23/30. Naive acceptance threshold is ~3.75 from what I've heard from an SACs tweet. My average is 4.6. Looks hopeful

@prathamgrv It's way too competitive now. In 2019, you could have read all ICLR accepted papers. You can't do it now. The volume of research is huge. The acceptance spots are limited.

Curious trend: I noticed NeurIPS papers in early 2000s often had just one author. Now it’s rare to see less than three. Is research getting more collaborative or more competitive?