Aman Chadha

@i_amanchadha

GenAI at @GoogleDeepMind • @Stanford AI • @UWMadison ECE • Ex-@Apple, #AWS, #Alexa, @Nvidia, @Qualcomm • EMNLP Outstanding Paper Award • #ai #ml #nlproc

SF Bay Area, CA

aman.ai

Joined January 2012

503 Following

4,591 Followers

Replies Reposts Media

Pinned
Aman Chadha
@i_amanchadha
Oct 4, 2025
Thanks, MIT Technology Review (@techreview), for featuring our work! • Our work guides LLMs toward fairness, neutrality, and inclusion. Our approach introduces a custom plug-and-play decoding layer for LLMs, applied only at inference time, without any parameter updates to the
GIF
MIT Technology Review
@techreview
Oct 1, 2025
The Download: OpenAI’s caste bias problem, and how AI videos are made trib.al/MwXaxqU
Aman Chadha
@i_amanchadha
Oct 23
🛠️ [𝐏𝐫𝐢𝐦𝐞𝐫] 𝐇𝐚𝐫𝐧𝐞𝐬𝐬 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 • harness.aman.ai - I’ve increasingly found that improving an AI agent isn’t only about improving the underlying model. The harness around it, including tools, memory, context, verification, and control flow, can
13
Aman Chadha
@i_amanchadha
Oct 18
🤖 [𝐏𝐫𝐢𝐦𝐞𝐫] 𝐍𝐯𝐢𝐝𝐢𝐚 𝐂𝐨𝐬𝐦𝐨𝐬 𝟑 • cosmos-3.aman.ai - I find world models especially interesting because Physical AI requires models to go beyond understanding what they see. They need to reason about what happens next and translate that understanding into
GIF
Aman Chadha
@i_amanchadha
Oct 10
🤖 [𝐏𝐫𝐢𝐦𝐞𝐫] 𝐐𝐰𝐞𝐧𝟑 & 𝐐𝐰𝐞𝐧𝟑-𝐕𝐋 • qwen3.aman.ai * I find Qwen3 especially interesting because it brings reasoning and non-reasoning into the same model while combining inference-time scaling, MoE efficiency, and RL-based post-training. (cf. "Qwen3
2
Aman Chadha
@i_amanchadha
Aug 28
⚙️ [𝐏𝐫𝐢𝐦𝐞𝐫] 𝐋𝐚𝐭𝐞𝐧𝐭𝐌𝐨𝐄: 𝐇𝐚𝐫𝐝𝐰𝐚𝐫𝐞-𝐀𝐰𝐚𝐫𝐞 𝐄𝐱𝐩𝐞𝐫𝐭 𝐒𝐜𝐚𝐥𝐢𝐧𝐠 • latent-moe.aman.ai - LatentMoE (arxiv.org/abs/2601.18089) by @velango, @nidhibhatia23, and @RWaleffe et al. shows that model efficiency is not just about reducing active FLOPs.
GIF

Aman Chadha

Aman Chadha’s posts