"Life is short. You guys have won the intellectual lottery. You're privileged and smart enough to do anything you want. Make the most of it." Parting words from Phillip Compeau at one of my last lectures at CMU.

meanwhile I’m still on free Colaboratory

Lots to look forward to Google IO tomorrow 👀

Open source is pretty amazing... Spent the last week exploring Sinhala/Tamil models - one of the biggest blockers for production AI use cases in Sri Lanka. For low resource languages like these, only the largest closed models(opus/gpt4o) seemed to do an acceptable job. These

If you tuned in to WWDC to see what Apple is doing with AI, we're all probably thinking the same thing around now 50 minutes into it... 🫠

Free ChatGPT with 4o without any logging - but you can sign in with your openai account to access paid features... Is apple paying openai or vice versa 🤔 Hard to see the case for openai paying apple (as with Google on safari) given no data collection, and no usage based

"In this World Cup, no fewer than five other teams (Sri Lanka, South Africa, Scotland, Ireland, and the USA) are sponsored by Indian companies that barely sell products in foreign markets. Which means that for even established sides such as South Africa and Sri Lanka, no

Crazy what Apple is capable of. This year, they managed to do what Geoffrey Hinton and Yann LeCun have failed to do for years. They have managed to make MNIST relevant again!!

UPI is really everywhere... But it's made transacting as a tourist harder - permanently out-of-order ATM machines, (anecdotally) lower credit card accepting merchants than in SL, etc. Makes sense that UPI doesn't extend to tourists without the aadhar/kyc layer. But why not

Figuring out how to price something like context caching seems like such an interesting problem that requires a pretty deep understanding of the underlying infra and model architecture. Even more interesting to try and infer/reverse engineer the high level infra and model

The tech Twitter and cricket Twitter parts of my feed have converged to the same topic today.

You can now use your voice to add context to searches in Google Lens! Press and hold on the shutter button in Lens, and it'll say "speak now to ask about this image." After speaking your question, let go of the button and Google Gemini will attempt to provide an answer.

♊️

Using whisper is so 2023. Just use gemini, pass in the raw audio directly and prompt the model directly with all the questions you have. With instructor, we can get - The exact mispronounced word - The timestamp when we did it - Advice on how to do better Flash truly is the

Probably the biggest use case unlocked by native audio - asking it to go slower, or check pronunciation just breaks without it.

Gemini 1.5 002 beats OpenAI o1-preview on MATH, and it does it at 1/10th the cost and no thinking time. When 2024 started, lot of people were critical of Google falling behind OpenAI. However, since then they have gathered themselves to pull the right strings. Whether it is

Gemini is now accessible via the OpenAI libraries! Update 3 lines of code and get started with the latest Gemini models : ) developers.googleblog.com/en/gemini-is-n…

1/ Two new features are coming to Lens starting today that let you search with your camera in new ways. These features use our latest AI models in Google Search which are especially good at multimodal tasks. 🧵