Mann Acharya

@goyalaman03 0.5x H100 is real

Hiring: Research Intern @ MaruthLabs We are looking for a Research Intern to join us for a 3-month internship focused on pushing the boundaries of high-performance Small Language Models (SLMs). The Role: • Research & Experimentation: You will be given access to 0.5x H100 GPU compute to test and iterate on your own research ideas. • Scaling Up: Upon reaching your research milestones, you will be granted access to an 8x H100 node for a full-scale training run. • Integration: Successful experiments and optimizations will be integrated directly into our core model training pipelines. Requirements: • Strong proficiency in Python and a deep understanding of Transformer architectures. • A research-oriented mindset with an interest in SLMs, efficiency, and context-length expansion. • Degree is not a barrier: We value proof of work, GitHub contributions, and technical curiosity over formal credentials. Details: • Stipend: ₹15,000 per month. • Duration: 3 Months (Extendable). • Location: Remote. How to Apply: Interested candidates should send their CV and a brief outline of a research idea they would like to explore on an H100 to contact@maruthlabs.com. #MaruthLabs #LLM #Research #Hiring #MachineLearning #SLM

@RisingSayak Congratulations 👏 an awesome paper!

Made it to ICML. Bye.

Really enjoyed this workshop on the current state of late interaction. Great to get caught up on what's been happening in the field with some nice call to actions: 1. More efficient late interaction 2. Stop beating single-vector IR 3. Scaling laws for late interaction (# parameters vs # vectors) 4. We need a late interaction community 5. Infra: (Open models and OSS for training, hosted search, OSS feature-complete native late interaction search engines, PyTorch-like interface)

A friend of mine just dropped some wisdom on a call "Don't be a judge to yourself, be a coach"

How to win best paper awards? Find the idea - have good research taste: know what to work on - get good collaborators: catch mistakes early (how to get them? spend time into writing cold emails) - read papers: why? 1. know what is possible, 2. what to do to get some result, 3. build on the paper - ignore papers: don't be biased by bad ideas, hard to know what to discard and what to keep - work on the most important problem in your field - be fast: the exact same paper written 6 months earlier is the difference between groundbreaking and boring - find your comparative advantage: e.g., I am good at data work but bad a modeling work - get lucky: 1. you stumbled upon a good idea, 2. be uniquely positioned to solve the problem Do the research - get lucky again: the idea actually works - kill papers that are not working: start on the hardest subproblem to derisk (I find this the hardest!) - kill low impact papers: finish fast as blog posts or workshop papers - reprioritize ruthlessly: immediately drop what you're working on and pivot to higher impact project (this is super hard because you have to be confident in your taste) - put in unreasonable amounts of effort: good papers don't have poorly performed experiments - have focus: on the core idea, nothing else - the paper should be in a local optimum: no obvious improvements visible Write it down - have focus: on the core idea, nothing else - know your reader: pick a particular one and write tailored to that reader - your title is not so important: convince with the science, not clickbait - your abstract is: convey sense of importance - write a good intro: your results don't matter if the reader doesn't get the problem - each figure must stand on its own: because most reader skim through the paper - write a good conclusion: moment of reflection, not just abstract in past tense - write well: read it out loud to catch things you would not otherwise After writing - get lucky one last time: 1. it's not too early, 2. it's still a hot topic, 3. nobody else did it first, 4. the committee likes it - be persistent: often multiple submissions are required Obviously don't just have the award as a goal, just use this to write impactful papers! Great post by Nicholas Carlini! nicholas.carlini.com/writing/2026/h…

@anulagarwal @fildotai Agreed. But isn't this then the perfect feature to add on a weather app though?

@fildotai Nooo what What does a weather app do???

Million Dollar app idea: create an app that uses this api and ping users in advance. Charge $29/yr, 10s of thousands of users living near risky areas will be happy to be notified of potential upcoming disasters.

@claudeai Oh no my tokens 😭

Introducing Code Review, a new feature for Claude Code. When a PR opens, Claude dispatches a team of agents to hunt for bugs.

@levelsio MCP has made tool calling and agents super accessible. It kickstarted widespread adoption of AI based tools. It is obvious that it may be 'universal' but not really the best for all scenarios.

Thank god MCP is dead Just as useless of an idea as LLMs.txt was It's all dumb abstractions that AI doesn't need because AI's are as smart as humans so they can just use what was already there which is APIs

after a certain number of LLM tokens burnt, you should just stop and get a technical intern

huh? senior openclaw developer???

Super impressive. I believe setting up coding infrastructure for agents in large codebases is going to be a practice as standard as making docker configs Minions: @stripe's one-shot, end-to-end coding agents stripe.dev/blog/minions-s…

Perplexity cannot beat @Google at search. The AI mode in Google is amazing. Try this query out: "Find me tech events in X city on Y date" And you'll know the difference

How web design/software design evolved through history Was trying to find some nostalgic flash games, but discovered something more interesting. @WebDesignMuseum exhibits thousands of screenshots and videos of websites, apps, software, and Flash games from the 1990s to the 2000s. Big kudos to @KovarPetr81 ,Ondřej Letocha and others who are maintaining this amazing timecapsule of the internet! webdesignmuseum.org

This was not in my 2025 Bingo Card

a ci/cd pipeline which we wrote months ago, which was not working suddenly just worked. In too big of a time crunch to even care at this point

Graduated from @bennettuniv 🎉 - Found the most amazing people, to be partners for life! - Went deep into tech, building across domains and stacks - Led groups, mentored people, gave talks! - Did internships, projects, research, hackathons from all across the globe! - Had fun through it all! Looking forward to learn even more. (considering posting a showcase/timeline of it all)

We have arrived at the era of slop machines: destination nowhere

@abhi1thakur @saibuilds i knoww!!!! pretty hilarious 🤣

@manndacharya @saibuilds it was a joke. dude took it seriously, so did i. obviously i had research backing me, he has poop backing his pants when i was working in recsys