Aneesh Aithal

I gave nine closed-source language models the brand-new JEE-Advanced 2025 paper—the toughest under-grad Physics-Chem-Math exam in India—and many of them out-scored the human AIR-1 (332/360)!! Why JEE-Advanced is a clean STEM benchmark - Fresh content → Paper was held on 18 May 2025, so it isn’t floating around training corpora. - No biology, no trivia → Pure PCM reasoning across 96 high-pressure questions. - Uncontaminated → GPQA and annual AIME are excellent, but their problems can leak into model training sets. JEE-Adv 2025 avoids that. Knowledge Cut-offs - OpenAI o3/o4-mini models → 1 Jun 24 - Gemini latest models → Jan 25 - Anthropic latest models → Mar 25 - JEE Advanced 2025 Paper date → 18 May 25 Only Gemini 2.5-Pro/Flash released after the exam date, so 7/9 models had zero chance to memorise these questions Implications - Deterministic PCM Q&A at JEE level is essentially **solved** for frontier models.   - Latency and unit-economics are now the real bottlenecks. - I'm expecting inference routers -- that hit a cheap, fast model first and fall back to a stronger model only when its confidence score is low -- to become the norm

665

68 Comments

🍀 6 months at Google.... Today I have completed 6 months at Google after having joined back in January 27th, 2025. These 6 months have been a rollercoaster of learning, exploration, meeting some exceptionally skilled people, and growth. Summarizing my journey at Google, here's the top 6 learnings for me over the period of the last 6 months... 1️⃣. Imposter Syndrome is real - Working at a place like Google, the talent density is the real deal. With being surrounded by some the smartest folks in the industry and working alongside them, its easy to feel lost and out of place. 2 things kept me going here - a. The support from the team in my rampup. b. The confidence that I've done harder things in the past - this too shall pass. 2️⃣. Ownership matters - At a place like Google, with highly skilled people all around, its kinda obvious that people like to own their work and expect you to do the same. How to own your work end to end, and to deliver it with the highest quality is something I've greatly learnt in the past few months. 3️⃣. Communication >>> Code - At a place, where any code change you make affects billions, communication and alignment regarding the same becomes something of utmost nessecity. Skillful communication within the team, cross teams, and with multiple stakeholders to effectively deliver work is something I've seen matter more than ever here. 4️⃣. Asking for help is a strength - In companies like Google and Microsoft, asking for help is looked on as a strength, not a weakness. The culture fosters that. However, to get help, you've to ask for it. Getting over the thought of judgment, to reach out for help, is a key learning which helped me a lot in the journey. 5️⃣. Make meaningful connections - Walking in any journey, what makes it worthwhile are the friends you make along the way. Making intentional connections, and going out of the way to help others move ahead in their journey, is the way to move ahead in the journey. This is something I also try to incorporate in my Linkedin journey by trying to give back to the community. 6️⃣. You're never too good at something, Keep learning - Tech is a fast moving space. The thought of "I already know enough" becomes obsolete as fast as the tech of today. Its a journey of constant of learning and upskilling. Parallely, make your basics stronger, that's what allows you to keep up the learning pace, with the industry. The last 6 months have been really exciting, fun, full of learnings and challenges all along the way. As its said in Google, being a Noogler is like "trying to drink from a firehose". Really excited for what the next 6 months unfolds. Stay tuned for more such content!!✌🏻🚀

909

85 Comments

This resume helped me land my role at Qualcomm as an AI/ML Engineer — and eventually my current position as a Sr. ML Scientist.🎉 But getting it right wasn’t a one-day job.I struggled just to get it shortlisted at first. Thankfully, with the help of some amazing mentors, I kept improving & eventually started landing interviews at some big tech companies. ⭐ Here’s what I changed after tons of feedback👇 ✅ I stopped just listing tasks. Instead, I showed real outcomes: “Reduced compression artifacts by 10%” sounds way better than “Worked on CNN models” ✅ I made it easy for recruiters to scan my stack: Python, PyTorch, Diffusion Models, LLaMA, LangChain, Generative AI — no guesswork. ✅ I highlighted patents, papers & internships, anything that showed depth beyond coding. ✅ I kept the format clean and simple — no fancy designs, just solid content. And most importantly, I made every bullet start with impact. It took me multiple revisions to get here. But once I did, things started moving. 👉 If you’re applying for AI/ML roles, focus on clarity, outcomes, and skills. The market is crowded. Your resume needs to work hard. But your Resume is just the first step, you need to be fully ready when it gets shortlisted. So, if you’re planning your journey in the Data Science, and want to be fully prepared, I’d suggest taking help from Bosscoder Academy. Checkout here: ​https://bit.ly/4kd8jVj They teach Data Visualization, Machine Learning, Generative AI, etc. as part of their curriculum to become a Data Scientist. You can choose them because they provide: ✅ A complete structured curriculum to get you interviewed & job-ready at top product companies. ✅ Personal mentorship from Data professionals at Google, Amazon, and Microsoft. ✅ Live classes, project-based learning, and access to a talented community of people with similar goals of upskilling. And, you’ll get dedicated placement support after upskilling as well. #machinelearning #resume #AI #sponsoredpost #interview #paidpartnership

1,306

24 Comments

🎉 Exciting milestone: my first research paper is published! The Aryabhata 1.0 Tech Report is now live on arXiv and Hugging Face A few weeks ago, we open-sourced Aryabhata 1.0, a compact 7B math reasoning LLM built for Indian academic exams like JEE. Today, we share the full technical journey behind it. What’s inside: 🔀 Model merging ⚙️ Data curation 🎯 SFT with curriculum learning 🤖 RLVR (A2C + group relative advantage, with our novel tweaks) 📊 Evaluation pipeline 📄 Read the paper: https://lnkd.in/gSPvKk8G 🛠️ Try the model: https://lnkd.in/de3aJyFF Grateful to my amazing team Sandeep Varma Sachin Dharashivkar Tejas Chaudhari VISHAL SINGH for making this possible. 🙏 It’s been a rewarding journey, from data curation to model training to evaluations, and seeing it documented as my first paper makes it even more special. #AI #MachineLearning #LLM #Research #OpenSource #NLP #MathReasoning #JEE #PhysicsWallah #HuggingFace

429

38 Comments

$100,000 for an H-1B visa. What does this mean for Indians? Every Engineer dreams of studying in the US and secure a job in the Silicon Valley. But Trump’s new H-1B fee changes that. Students may be forced to return earlier than planned because companies will now think twice before sponsoring visas. But you don’t need to go to US to earn in dollars. With the right skills and network you can earn in dollars while living in India. And this H-1B visa fee might change the hiring process in MNC’s too. Remote jobs will now get a spotlight. So here are some of the best portals where Indian engineers can land US remote jobs (links to all portals are in comments) ✅ Flexiple: Curated freelance network for top engineers and designers. ✅ Turing (dot) com: AI-powered platform that matches Indian developers with US companies. ✅ We Work Remotely: One of the oldest and most trusted boards for global remote roles. ✅ AngelList Talent (Wellfound): Startup-first hiring platform, many remote roles listed by US founders. ✅ RemoteOK: Job board focused on tech and product roles across geographies. ✅ Contra: Great for independent creators, developers, and freelancers targeting US startups. In 2025, opportunity isn’t tied to one country. It follows skills that prove their worth anywhere.

210

6 Comments

Right now, it’s not just H1B holders under pressure, the ones really feeling the burn are 𝘀𝘁𝘂𝗱𝗲𝗻𝘁𝘀, 𝗳𝗿𝗲𝘀𝗵 𝗴𝗿𝗮𝗱𝘂𝗮𝘁𝗲𝘀 𝗮𝗻𝗱 𝗮𝗹𝗹 𝘁𝗵𝗲𝘀𝗲 𝗨𝗦 𝗳𝗼𝗹𝗸𝘀 𝗮𝗿𝗲 𝗰𝗼𝗺𝗶𝗻𝗴 𝗯𝗮𝗰𝗸 𝘁𝗼 𝗲𝗮𝘁 𝗜𝗻𝗱𝗶𝗮𝗻 𝗜𝗧 𝗷𝗼𝗯𝘀. So, the bigger hit is actually on:  • Students studying here in US with crazy loan amounts on their head.  • Folks who graduated but still don’t have a job. Many students are already stressed about getting a job, and now this adds on. Dealing with rent, family pressure, loan interest, and living expenses all at once, so just imagine. And don’t think this stops in the US, it’s gonna bounce back to India too. The first impact? 𝗜𝗻𝗱𝗶𝗮𝗻 𝗜𝗧. Bcz the people heading back from the US will be looking at those jobs, and let’s be real, they already have an edge. They’ve worked with 𝗠𝗔𝗔𝗡𝗚/𝗙𝗔𝗔𝗡𝗚, on 𝗿𝗲𝗮𝗹 𝗽𝗿𝗼𝗱𝘂𝗰𝘁-𝗯𝗮𝘀𝗲𝗱 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀, working on 𝗹𝗮𝘁𝗲𝘀𝘁 𝘁𝗲𝗰𝗵 𝗹𝗶𝗸𝗲 𝗔𝗜/𝗠𝗟. Sure, India has a lot of potential and a big gap to fill in tech. But the real question is, 𝗮𝗿𝗲 𝘄𝗲 𝗿𝗲𝗮𝗱𝘆 𝘁𝗼 𝘁𝗮𝗸𝗲 𝗶𝗻 𝘁𝗵𝗶𝘀 𝘄𝗮𝘃𝗲 𝗼𝗳 𝗿𝗲𝘁𝘂𝗿𝗻𝗲𝗲𝘀? Truth is, 𝘁𝗵𝗲𝘆’𝗹𝗹 𝗲𝗮𝘀𝗶𝗹𝘆 𝗼𝘂𝘁𝗰𝗼𝗺𝗽𝗲𝘁𝗲 𝗮 𝗹𝗼𝘁 𝗼𝗳 𝘁𝗵𝗲 𝗲𝘅𝗶𝘀𝘁𝗶𝗻𝗴 𝘁𝗮𝗹𝗲𝗻𝘁 𝗶𝗻 𝗜𝗻𝗱𝗶𝗮. Why? Bcz most niche technologies emerge first in the US and only later get outsourced to Indian firms like Infosys or TCS. 𝗦𝗼𝗺𝗲 𝗵𝗮𝗿𝘀𝗵 𝗿𝗲𝗮𝗹𝗶𝘁𝗶𝗲𝘀:  • 𝗗𝗼𝗻’𝘁 𝗿𝘂𝘀𝗵 𝘁𝗼 𝘁𝗵𝗲 𝗨𝗦 𝗿𝗶𝗴𝗵𝘁 𝗻𝗼𝘄. There are already too many students stuck in the job hunt, some graduated, some still studying, and many more on the way.  • The numbers are insane, 𝗼𝘃𝗲𝗿 𝟭.𝟭𝟮 𝗺𝗶𝗹𝗹𝗶𝗼𝗻 international students in the US (2023-24, record high), 𝟯𝟯𝟭𝗸+ from India—𝟮𝟯% jump in a year. Another report puts Indians at 𝟰𝟮𝟬𝗸 (𝟮𝟳% of all students). Massive pressure on the market.  • A big reason? 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝗶𝗲𝘀 𝘄𝗮𝗶𝘃𝗲𝗱 𝗚𝗥𝗘 𝗮𝗻𝗱 𝗧𝗢𝗘𝗙𝗟/𝗜𝗘𝗟𝗧𝗦 during covid, which made it way easier for people to get in US, even if they weren’t really ready for the competitive job market here.  • On top of that, 𝘀𝗶𝗻𝗰𝗲 𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗯𝗹𝗲𝘄 𝘂𝗽 𝗶𝗻 𝟮𝟬𝟮𝟯, 𝗰𝗼𝗺𝗽𝗮𝗻𝗶𝗲𝘀 𝗲𝘃𝗲𝗿𝘆𝘄𝗵𝗲𝗿𝗲 𝗰𝗵𝗮𝗻𝗴𝗲𝗱 𝗰𝗼𝘂𝗿𝘀𝗲 𝗮𝗻𝗱 𝘄𝗲𝗻𝘁 𝗮𝗹𝗹-𝗶𝗻 𝗼𝗻 𝗔𝗜. Now tell me, 𝗵𝗼𝘄 𝗺𝗮𝗻𝘆 𝘂𝗻𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝗶𝗲𝘀 𝗶𝗻 𝗜𝗻𝗱𝗶𝗮 𝗮𝗿𝗲 𝘀𝗲𝗿𝗶𝗼𝘂𝘀𝗹𝘆 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗔𝗜/𝗠𝗟-𝘀𝗸𝗶𝗹𝗹𝗲𝗱 𝗴𝗿𝗮𝗱𝘀? And even if they are, 𝗵𝗼𝘄 𝗺𝗮𝗻𝘆 𝗰𝗼𝗺𝗽𝗮𝗻𝗶𝗲𝘀 𝗮𝗿𝗲 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘄𝗼𝗿𝗸𝗶𝗻𝗴 𝗼𝗻 𝗻𝗶𝗰𝗵𝗲 𝗔𝗜 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀 𝗶𝗻 𝗵𝗲𝗮𝗹𝘁𝗵𝗰𝗮𝗿𝗲, 𝗳𝗶𝗻𝗮𝗻𝗰𝗲, 𝗼𝗿 𝗼𝘁𝗵𝗲𝗿 𝘀𝗲𝗰𝘁𝗼𝗿𝘀 on their own? Not many. So yeah, 𝘁𝗵𝗶𝘀 𝘄𝗮𝘃𝗲 𝗶𝘀 𝗴𝗼𝗶𝗻𝗴 𝘁𝗼 𝗵𝗶𝘁 𝗜𝗻𝗱𝗶𝗮𝗻 𝗜𝗧 𝘁𝗵𝗲 𝗵𝗮𝗿𝗱𝗲𝘀𝘁. And if anyone in India is feeling happy about US layoffs or immigration issues, 𝗷𝘂𝘀𝘁 𝗯𝗲 𝗰𝗮𝗿𝗲𝗳𝘂𝗹, 𝗯𝗰𝘇 𝗶𝗳 𝘁𝗵𝗲𝘀𝗲 𝗨𝗦 𝗳𝗼𝗹𝗸𝘀 𝗮𝗿𝗲 𝗰𝗼𝗺𝗶𝗻𝗴 𝗯𝗮𝗰𝗸... 𝗮𝗻𝗱 𝘁𝗵𝗲𝘆’𝗿𝗲 𝗰𝗼𝗺𝗶𝗻𝗴 𝗳𝗼𝗿 𝘆𝗼𝘂𝗿 𝗷𝗼𝗯𝘀.

111

18 Comments

H-1B lottery 2025 is happening right now. If you don’t get chosen, here are 6 alternatives. (Save this 📌) In 2024, a total of 442,000 unique beneficiaries submitted H-1B registrations, out of which 135,000 were selected. Every year, tens of thousands don’t get picked in the H-1B lottery. But that doesn’t mean your U.S. career is over. There are a few alternatives that don’t rely on luck. ✅ Here are 6 solid options you can explore: 1. F-1 Visa (OPT & STEM OPT) – Go back to school for a higher degree & get OPT. Cost: → Visa fee (MRV): $185 → SEVIS fee: $350 2. O-1 Visa – For individuals with extraordinary ability. Cost: → Filing fee: $1,655 (includes $1,055 + $600 Asylum Program fee) → Legal fees: $5,000–$15,000 3. L-1 Visa – Transfer abroad & return after a year. Cost: → Filing fees: $1,655 (for most applicants; smaller employers may pay less) → Legal fees: $3,000–$8,000 4. H-4 EAD – Work authorization for spouses of H-1B holders. Cost: → Filing fee: $410 → Legal fees: $500–$2,500 (if using a lawyer) 5. International Entrepreneur Parole (IEP) – For startup founders with U.S. investor funding. Cost: → Filing fee: $340 (plus additional documentation costs) → Legal fees: $5,000–$15,000 6. Other Countries – Canada, UK, and India offer solid career paths. Cost: → Canada: Work permit ~ CAD $155 (~₹9,300) → UK: Skilled Worker visa ~ £719–£1,639 📌 Swipe through the carousel for a breakdown of each option—including costs, benefits, and challenges. 🌿 Please re-share this post so it helps more people. 🎁 Don’t miss out on the free guide on O-1: https://lnkd.in/d8fNeYUi

152

22 Comments

🚨 SWE in 2025 and still ignoring LLMs? You are not just behind …. You are replaceable 💣 But the good news? It’s not too late to catch up — and you don’t need a fancy degree or paid course to start. Note: This is not promotional! I found a goldmine of a FREE resource by Victor Dantas that takes you from: 🔸 Basic Machine Learning ➡️ to Neural Networks ➡️ to RNNs ➡️ to Transformers ➡️ all the way up to LLMs like GPT & LaMDA. It’s literally a zero-to-hero roadmap that breaks down how these complex systems actually work — in plain English. 📚 Whether you’re a SWE, student, or AI-curious engineer — this is the kind of stuff that will future-proof your career. #AI #LLM #MachineLearning #NeuralNetworks #Transformers #SoftwareEngineering #GPT #LearnForFree

190

37 Comments

At 3 AM on a Wednesday, Amazon needed me. I wasn’t even on that team anymore. Back when I was an SDE at Amazon, a Sev-2 had hit one of our services. Many customers were impacted. The team had been chasing it for hours. Still, they called me. I rolled out of bed, opened my laptop, and joined the bridge. The tricky part was that the fault started a week earlier. A code change had shipped behind a flag, traffic ramped slowly, and the failure signature was noisy. Logs were chatty in all the wrong places and silent where we needed them most.Dashboards showed errors, but no clear line to the root. We did unglamorous things. We wrote a minute-by-minute timeline. We diffed deployments, configs, and flags across regions. We checked queue backlogs and retry storms. We sampled logs with new filters. We bisected the traffic ramp. We rolled back the smallest safe thing first. Then we added one log, in one hot path, and the picture snapped into focus. By sunrise, we had the root cause and a fix. The impact was contained. Customers recovered. What stayed with me was not the fix. It was why they called. They did not call because I am smarter. They called because, over months, I had done the boring work. I wrote runbooks when no one asked. I paired with juniors on their first on-call. I shared context so others could make decisions without me. I showed up when it was not my ticket. Trust is not built in one night. It is built in all the small days that come before it. Ownership is simple. If it affects your customer, it is your problem. Org charts are for payroll, not for 3 AM. If you are early in your career, remember this: → Document the weird corners when you discover them. → Add one useful log where it hurts, not twenty where it is easy. → Keep a living checklist for rollback, not a static wiki page. → Teach one person the thing you just learned. → Read your dashboards when things are calm, not only when they are red. At 3 AM, you will not rise to the occasion. You will fall to the level of your habits. Build the right ones while the house is quiet. That’s how you earn trust that lasts beyond teams and ownership that outlives org charts. A few days later, this moment turned into an accolade from the team..  a reminder that trust is built long before it’s tested :)

334

45 Comments

🇮🇳 #AACL2025 is coming to India and I couldn’t be more excited! 🚀 Alongside serving as Student Research Workshop (SRW) Chair, I’m glad to share that WiNLP will be collaborating with the Diversity & Inclusion Chairs —(Tirthankar Ghosal, Sriparna Saha, Sudeshna Sarkar), to help create a more inclusive, diverse, and memorable conference experience for all participants. 💜 WiNLP (Widening NLP) is dedicated to highlight the contributions of researchers from underrepresented groups and fostering access, equity, and inclusion across languages, disciplines, and backgrounds. We’re here to promote and support ideas and voices of underrepresented groups in Natural Language Processing (NLP). Our goal is inclusivity for people of all identities, abilities, and perspectives in the field. At AACL 2025 in Mumbai, we’re looking forward to building on this mission and ensuring the conference becomes a welcoming space for everyone in the NLP community. 👉 Stay tuned for updates on upcoming activities and initiatives — more news soon! #WiNLP #AACL2025 #Inclusion #Diversity #NLProc

57

2 Comments

When math draws a hard line for large language models A recent research paper by Vishal Sikka and his son Varin Sikka takes a very different approach to evaluating large language models. Instead of benchmarks and demos, it uses mathematics. The authors show that LLMs operate within a fixed computational limit. Once a task requires more computation than the model can perform during inference, two things happen: the model cannot reliably solve the task and cannot verify correctness either. Incorrect output becomes unavoidable. This applies directly to many “agentic AI” scenarios - long planning chains, multi-step decision making, global optimization, and autonomous workflows. The paper does not argue that LLMs are useless. Quite the opposite. Within their computational domain, they are incredibly powerful tools. What it shows is that some classes of problems sit permanently beyond what transformer-based models can handle, no matter how much data or training we add. AI can become a phenomenal accelerator - for code, analysis, automation, and knowledge work. At the same time, treating it as a universal reasoning engine or a fully autonomous problem solver creates real risk. The research is a good reminder that understanding the limits of a tool is just as important as admiring its strengths. You can read about in more detail here: https://lnkd.in/dqQBQZxd

12

You are in a system design interview at Amazon for an SDE-3 role and the interviewer has given you the question to Design Netflix. He then asks a follow up to you: How does Netflix knows when to show: Are you still watching? Is it just time-based? If it were time-based, you would see it every time. So what is actually going on under the hood? Both look like a simple pop-up. Underneath, it is a mix of product thinking, client logic, and backend events. Btw, if you’re preparing for system design/coding interviews, check out our mock interview tool. You can use it for free here: https://lnkd.in/gpCn7t2T [1] Start from the product goals The feature is not only about nagging people. - Save bandwidth and CDN cost if the viewer has slept or walked away - Avoid autoplaying potentially sensitive content in an empty room - Protect kids if parents start a show and leave the TV on - Do all this without annoying active binge watchers So the design must answer one question: "When is the user probably not here any more" [2] Naive design - pure timer Simplest idea: - If playback has been running for 2 hours, show Are you still watching - If user clicks "Yes", reset the timer Why this is weak: - Someone can binge 5 episodes in a row and gets interrupted in the middle of an intense scene - Someone can start a show, walk away after 5 minutes, and the platform will keep streaming for the next 2 hours - It ignores how many episodes were auto played, device type, time of day, user habits  [3] Realistic design - session and engagement signals Think in terms of a "watch session" and "engagement events". Signals the client can track: - Play, pause, seek, volume change - Episode finished and next episode auto started - Remote or keyboard input, UI navigation - Screen on or off events from the device - For mobiles: app background or locked state A common heuristic could be: - Count how many episodes have auto played without any user interaction - Track how long it has been since the last button press or navigation - Only trigger the prompt when both are high enough Example rule: - If 3 episodes in a row have auto played - And there has been no interaction for 45 minutes - Then, before starting episode 4, show Are you still watching This feels much smarter: - Active viewers who pause, skip intro, change volume keep resetting the "engagement clock" - Sleeping viewers do not touch anything, so the next episode is blocked by the prompt  [4] Client heavy vs server heavy design You can talk about two design choices. Client first: - All logic runs inside the app on TV, mobile, or web - Client keeps an in memory session model and decides when to show the popup - It still sends telemetry events to backend for analytics and future tuning Pros: - Works even with flaky network - Highly responsive, no extra server round trip Cons: - Logic must be implemented and updated across many platforms - Harder to quickly roll out rule changes Continued ↓

184

12 Comments