Who and what ultimately decides which ML model gets used in enterprises? In what’s often called “classic” ML, most prediction problems end up as linear-regression variants or XGBoost’s cousins. Why not something in between? I often feel it’s because teams lack visibility into

The only book that captures the real story and essence of Data Science and it’s objective; by chriswiggins @nytimes @Columbia just that the field is often seen as data pipeline and dashboard building at many organisations, but the early principles are so ambitious!

If I recall correctly, many projects I've worked on have struggled with what the attached paper calls the inversion problem—predicting user preferences or expertise from behavioral data that's inherently biased, fails to capture mental states, or doesn't reflect what people

Too many nemvid people today

Been writing https://t.co/Ps01LJnKGm

Any LLM benchmarks specific to prediction markets?

Atlanta,GA

Are there any events or meetups happening around Atlanta this week? #atlanta

Saravana Bhavan NYC is so good, real flavours, gonna be a monthly routine !

𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝗶𝘀 𝗶𝗻𝘀𝘁𝗿𝘂𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝗳𝗼𝗿 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀. Let me explain. In my instrumentation engineering days, we used sensors, DACs, and Kalman filtering to bring signals to the control system. Analytics does the exact same thing—just for companies.

big apple brings bigger smiles 😀, let me tell you something what could make us smile bigger, we often choose from the best options available at hand, say most of the MS students choose a course or finalize a University among the best admits they got, I chose columbia over UMich,

everyone’s shipping agents. but who’s training the models the agents call? different game. Applied AI = owning the intelligence Applied ML = owning the algorithm agents are software engineering energy — orchestration, tool calling, workflows models are machine learning energy —

what i have been working on today : - building a simple sandbox for the generative agents simulation - say how one can run parallel executions of multiple studies/experiments - gonna lock-in for MATS Application - continuing the experiments on Counterfactuals + LLMs <hopefully

okay, relax , don't panic if your Social Media feed is filled with Deep Seek's mHyper Connections paper(Manifold-Constrained Hyper-Connections)! and If you haven't trained a Model(even GPT-2 era transformer arch models) in last 100 days, probably this paper is something you

𝗟𝗼𝗼𝗸𝗶𝗻𝗴 𝗳𝗼𝗿 𝗮 𝗦𝘂𝗺𝗺𝗲𝗿 𝟮𝟬𝟮𝟲 𝗜𝗻𝘁𝗲𝗿𝗻𝘀𝗵𝗶𝗽 (𝗼𝗿 𝗘𝗾𝘂𝗶𝘃𝗮𝗹𝗲𝗻𝘁 𝗘𝗻𝗴𝗮𝗴𝗲𝗺𝗲𝗻𝘁) I’m exploring Summer 2026 opportunities where I can work on hard, real problems at the intersection of AI/ML, applied science, and product execution. I’d love to

if you are learning Deep Learning , and wanna bet on that... Don't aim for getting an AI engineer role, you are not at going to train any models, or say touch any architectures, in most of the places the role is not at all demanding test time compute/optimization work too!!

𝗧𝗵𝗲 𝟮𝟬𝟮𝟲 𝗗𝗮𝘁𝗮 𝗦𝗰𝗶𝗲𝗻𝗰𝗲 𝗥𝗼𝗮𝗱𝗺𝗮𝗽 — 𝗕𝘂𝘁 𝗡𝗼𝘁 𝘁𝗵𝗲 𝗢𝗻𝗲 𝗬𝗼𝘂'𝘃𝗲 𝗦𝗲𝗲𝗻 𝗕𝗲𝗳𝗼𝗿𝗲 I have a blog, a podcast, and a deck to share with you. But more than that — a refreshed perspective. Most roadmaps you get from LinkedIn or WhatsApp forwards

Convergence at Midnight... + 5 Years Before the Next Leap Wednesday, December 31st This is going to be an exciting year ahead. So before that, consider this my step zero of something. I want to call this the convergence point. This is where I just want to start converging

2025 has been a year of emergence and convergence in tandem, and personal experiments that showed me all kinds of limits and interests. From being a Senior ML Engineer to now being in a position where I need a summer internship, but with a lot more clarity about frontier AI/ML,

Sunday brunch!