Akash Mehra (@akashmehra) 's Twitter Profile
Akash Mehra

@akashmehra

Love math, physics & machine learning. always learning, so I make a lot of mistakes and learn from them. Views are my own and not those of my employer.

ID: 22841327

linkhttp://akashmehra.github.io calendar_today04-03-2009 21:55:18

7,7K Tweet

296 Takipçi

1,1K Takip Edilen

NASA Aeronautics (@nasaaero) 's Twitter Profile Photo

Along with international partners, we're using the same generically shaped wing design to create research models that will help us better understand how air moves around an aircraft during takeoff+landing. More on this innovation for the benefit of all: go.nasa.gov/3XzVHzL

Akash Mehra (@akashmehra) 's Twitter Profile Photo

Absolutely blown away by Brian Cox's Horizons at the Herbst Theater in SF! He has an incredible gift for making complex science not only understandable but utterly captivating. A show packed with information and meaning! And what a thoughtful way to end it! 🙌

Akash Mehra (@akashmehra) 's Twitter Profile Photo

The following is so unlike me to do on Twitter but I got the premium account and it allowed me to ramble on, so 🤷🏻‍♂️... When we invented steam engines, the job of the carriage puller was suddenly at risk, but we didn’t stop there. We kept going. Steam engines powered railroads,

Engineering at Meta (@fb_engineering) 's Twitter Profile Photo

Meta and NVIDIA have teamed up to supercharge vector search on GPUs by integrating NVIDIA cuVS into Faiss v1.10, Meta’s open-source library for similarity search. This collaboration brings groundbreaking performance improvements: 🔹 IVF indexing: NVIDIA cuVS boosts build times

Meta and NVIDIA have teamed up to supercharge vector search on GPUs by integrating NVIDIA cuVS into Faiss v1.10, Meta’s open-source library for similarity search. This collaboration brings groundbreaking performance improvements:

🔹 IVF indexing: NVIDIA cuVS boosts build times
Akash Mehra (@akashmehra) 's Twitter Profile Photo

It looks scary but it isn't. It's a beautiful concept and requires a bit if thinking and then one can derive the equations on their own.

Akash Mehra (@akashmehra) 's Twitter Profile Photo

Yup! a 100% agree. And also, the information content per token is miniscule! Not sure what the appeal is in filling pages with non-informational content. My brain immediately goes into scanning mode and throws out 80-90% of the text.

Akash Mehra (@akashmehra) 's Twitter Profile Photo

The models have gotten better, no doubt and they'll continue to get better but the talk of AGI "next year" or in next 5 years was a bit too much considering we were basing that on LLMs. We're making serious progress in terms of learning and unlocking potential of these models but

Akash Mehra (@akashmehra) 's Twitter Profile Photo

Text is one type of input, in the next step the model could easily be multimodal, followed by speech in speech out. Making the model culturally aware happens when you provide the modality geared towards a culture. Not sure what this 👇🏽random rant is about.

Akash Mehra (@akashmehra) 's Twitter Profile Photo

Everyone's flipping over this, but all Yann is trying to say is (at least I interpret it like this) that you can deploy a dumb system and make it harmful or take a very intelligent system and make it harmless by virtue of guardrails, deployment (to an extent) and an improved

USCIS (@uscis) 's Twitter Profile Photo

President Trump’s new H-1B visa requirement applies only to NEW, prospective petitions that have not yet been filed. Petitions submitted prior to September 21, 2025 are not affected. uscis.gov/sites/default/…

President Trump’s new H-1B visa requirement applies only to NEW, prospective petitions that have not yet been filed. 

Petitions submitted prior to September 21, 2025 are not affected.

uscis.gov/sites/default/…