Rahil Parikh

AI Scientist at Amazon working on LLMs and Responsible AI #NLPProc, Speech, Trustworthy ML | Prev- @UofMaryland, @bitspilaniindia | Views are my own

[Intology @intology]

Can coding agents do research?

We release NanoGPT-Bench, an internal eval we’ve used to test agents on an AI R&D problem with months of human progress

Codex, Claude Code, Autoresearch recover only 9.3% of human progress, mostly tuning hyperparams & ignoring algorithmic research

NanoGPT-Bench is built on the NanoGPT Speedrun, a popular LLM pretraining competition to minimize the training time of a GPT-2 style model. Existing human submissions constitute nearly 2 years of work. To control for dependencies and contamination in frontier models, we standardize evaluation to a 5-month window of world records. Evaluation is fully autonomous and end-to-end, with no human intervention or internet access. 🧵

[Rahil Parikh @ParikhRahil]

@forwarddeploy Hell yeah!!

[Rahil Parikh @ParikhRahil]

@gargi_balasu @GoogleDeepMind Congratulations!!

[Rahil Parikh @ParikhRahil]

@drparikhpurnima @MumbaiPolice @mybmc Please look into this issue that has not been resolved despite of years of complaining

[Rahil Parikh @ParikhRahil]

@ngsankha @lambda_calculus @michael_w_hicks @LeoLambro Congratulations!!

[Rahil Parikh @ParikhRahil]

@drparikhpurnima @mybmcwardKW @mybmc @MumbaiPolice

[Rahil Parikh @ParikhRahil]

@drparikhpurnima @mybmcwardKW @mybmc This has been a constant problem here. Please look into this

[Rahil Parikh @ParikhRahil]

@yadapruksachatk @deliprao @rachel_l_woods Thanks— that’s useful information :-)
I’ll keep a lookout for communities of researchers and builders working on trustworthy AI

[Rahil Parikh @ParikhRahil]

@lorenlugosch Congrats!!

[Rahil Parikh @ParikhRahil]

@nathanbenaich I’d love to join 👋

[Rahil Parikh @ParikhRahil]

@_josh_meyer_ Awesome! Exited to see and use your work :D

[Kartik Parikh @kartikparikh]

Burning bushes every night is a risk to the environment; not to mention the risk of fire to the vehicles around. This is at Tilak Udyan area opposite Arya Vidya Mandir School. Juhu sceme. @MumbaiPolice @AmeetSatam @mybmcwardKW

[Rahil Parikh @ParikhRahil]

@TerribleMaps Good for you, Chile!

[Rahil Parikh @ParikhRahil]

@drparikhpurnima @mybmc

[Rahil Parikh @ParikhRahil]

@stephanedascoli Wow, there���s a seal behind you!

[Rahil Parikh @ParikhRahil]

@shreya8pp @mybmcWardN @mybmc This is located around multiple hospitals and hinders people from using the footpath @ParagShahBJP @manoj_kotak

[Kartik Parikh @kartikparikh]

@FastechT Looking forward to meet you at @exploreIMC in Hall 5, Booth 5.23 where we will show our range of test instruments and network visibility solutions for 5G NR.
Do reach out to us at @FastechT
to book an appointment! 1-4 October 2022 in IPTO New Delhi! #IMC2022

[Rahil Parikh @ParikhRahil]

Looking forward to share our work at @eceumd linking speech production and the auditor cortex at @ISCAInterspeech in session- Thu-P-VR-9-3, on Thursday at 10AM KST
https://t.co/Oq1i4t2n0o

[Rahil Parikh @ParikhRahil]

3. In 'Acoustic To Articulatory Speech Inversion Using Multi-Resolution Spectro-Temporal Representations Of Speech Signals' we demonstrate a correlation between the brain's representation (cortical features) of speech and the vocal tract parameters using speech inversion

[Rahil Parikh @ParikhRahil]

I'm excited to present this work with @AmazonScience on Wednesday 4PM KST @ISCAInterspeech
https://t.co/iZHVU5MYOp

[Rahil Parikh @ParikhRahil]

2. In 'Impact of Acoustic Event Tagging on Scene Classification in a Multi-Task Learning Framework' (with @AmazonScience), we show that joint-training with AET improves ASC mainly due to the regularization effect of event tagging and not because the model has learnt the 'events'