Researcher (ex DeepMind, Adobe Research) Sharing research, opportunities and obvio random stuff. X Scholar: https://nitter.gq/t.co/ZO6wgrE2BS
Joined December 2022
- Tweets799
- Following617
- Followers1.9K
- Likes5.2K
Pinned Tweet
Two "road less traveled" decisions that shaped my career more than anything else:
1. In 2024, I had an on-campus SWE offer from Google in hand. It seemed a safe choice and a good role. but I knew I wanted to do ML research, which wouldn't really happen in a standard software role. I remembered my Adobe Research internship mentor mentioning the Pre-Doctoral / AI Residency program at Google DeepMind India. I applied without overthinking it and got in.
The growth over those 2 years was insane. Getting to work on frontier models like Gemini and Veo fresh out of college, the network, crazy offsites, traveling for top ML conferences, and the doors it opened afterward (and yes, money follows my brother, money follows lol). Definitely had imposter syndrome early on, but having incredible mentors and collaborators made all the difference and the best part is doing all of this while eating clean and putting in 1-2 hours a day toward fitness.
2. After JEE, based on my rank, the conventional move at IIT Roorkee was BTech in ECE. Instead, I chose the 5-year BS+MS in Applied Mathematics, a branch that was less popular and had a lower cutoff than CS, ECE, or EE at the time. I didn't know exactly what I wanted to do yet, and my gut said math would give me the best foundation and flexibility. Looking back, it aligned with what I ended up doing next.
Sometimes stepping off the default track is the best bet you can make.
Reminder to use /compact in Antigravity to free up your context window and avoid context rot in your conversation:
1. Use /compact when you're in the middle of a single long task and want to clear out large exploration or debugging history.
2. Run /learn before wrapping up if you solved a tricky thing or corrected the agent and want that knowledge to persist into reusable rules/skills across conversations.
3. Start a fresh conversation when switching to an unrelated task.
Sahil Goyal retweeted
I’m hiring a Student Researcher at @GoogleDeepMind.
I am looking for PhD students working on agents, open-endedness, self-improvement or scientific discovery.
📍 Location: London
🗓️ Duration: 24 weeks, 100%
🚀 Start date: June or July 2027
Apply now using the links below👇
My Xscholar analytics: share research papers, resources, and interview experiences.
My brain: shitpost.
Sahil Goyal retweeted
See what I found. First patent award from Google. Its from 2025. My parents think I did some groundbreaking invention. tbh I am not going to correct them lol.
See what I found. First patent award from Google. Its from 2025. My parents think I did some groundbreaking invention. tbh I am not going to correct them lol.
Seems a path to RSI
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Sahil Goyal retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Reviewer 2: Proposed method lacks theoretical novelty and mathematical rigor.
Camera-ready revision:
Sahil Goyal retweeted
I am hiring multiple PhD positions at AMLab at the University of Amsterdam to work on deep learning from a physically inspired point of view.
come work with me!
(link below)
1/n
Sahil Goyal retweeted
I'm on the industry job market this year!
I'm a PhD student at CMU LTI. My work spans post-training, evals, alignment and multimodality!
Please reach out (DM) if you are hiring and think I'd be a good fit, I'd love to chat :)
Website: simran-khanuja.github.io/
CV: simran-khanuja.github.io/ass…
I was little skeptical if it'll refresh the feed daily through some X's sync api as I completely vibe coded it. It seems it does. It has my latest tweet there.
Moreover based on trending topics and my previous tweets, it also shows daily post suggestions.
Sahil Goyal retweeted
📢 Our team at Qualcomm is hiring Research Interns for 2027. We're looking for candidates with a background in deep learning and computer vision to work on generative AI, video generation, and neural rendering.
🔗 Apply here: qualcomm.wd12.myworkdayjobs.…
#Qualcomm #Internship
Sahil Goyal retweeted
Our team at @GoogleResearch is looking for a PhD research intern for 2027 at Mountain View.
The research topics include (but not limited to) multimodal RL and agents/RSI for healthcare.
You will be directly working with me. Please check details at:
forms.gle/tGrb1v2sF89AN1fV6
Sahil Goyal retweeted
A ‘hello world’ milestone for Project Suncatcher 🚀
After launching our prototype satellite a few weeks ago, we’ve now successfully sent a command up to orbit, booted up the TPU, ran the prompt, and received our first complete response back on earth - a haiku!