Krishna Srinivasan

I code stuff. I build models. I work in NLP/DL at Google Research.

[Krishna Srinivasan @krishna2]

Thank you so much WikiMedia Foundation for this great honor and this award! 🙏

[Wiki Workshop 2025 @wikiworkshop]

Congrats to the recipients of the @Wikimedia Research Award of The Year!!

🎉"WIT: Wikipedia-based Image Text Dataset for Multimodal Multilingual Machine Learning" Srinivasan et al

🎉"Assessing the quality of sources in @Wikidata across languages: a hybrid approach" Amaral et al

[Kunal Bhatia @kunalbhatia91]

Superintelligence will be built on Self Improvement.

Today @hexoai, we’re excited to release ‘SIA’ - an open-source Self-Improving AI, to achieve any goal through recursive self improvement.

While trying to solve a problem, SIA doesn't just improve it's abilities by updating it's harness, it updates it's own weights as well.

[Krishna Srinivasan @krishna2]

Doing a startup and organizing taxes ... couldn't help but contemplate between AGI [1] and AGI [2].
AGI [1] = Artificial General Intelligence
AGI [2] = Adjusted Gross Income

[Krishna Srinivasan @krishna2]

@JamesClear Reminds me of this:
How do you make good decisions?
From experience.
How do you get experience?
From bad decisions.

[WikiResearch @WikiResearch]

"WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models", consisting of 384K image-text pairs from Wikipedia with domain labels, along with human-verified in/out distribution test sets.

(Kalyan et al, 2024)

https://t.co/ys7gU3eGO9

[Krishna Srinivasan @krishna2]

@snowmaker @Jacob_Heller Brilliant! 👏👌

[Krishna Srinivasan @krishna2]

@lateinteraction @databricks @DbrxMosaicAI Huge congratulations, Omar!!

[Krishna Srinivasan @krishna2]

@karpathy Eureka we have a lab, Eureka Labs, where we can learn AI the way it is meant to be taught. Fantastic news and couldn't think of a better person to do this. Looking forward to learning. Congratulations and best wishes!

[Krishna Srinivasan @krishna2]

@rasbt Thanks so much for this great work. I am having lots of run reading this and the ML Q&A books. Paraphrasing a popular quote, "What one can't build, do and teach, one doesn't truly understand".

[Krishna Srinivasan @krishna2]

@josh_metcalf @threadreaderapp unroll

[Krishna Srinivasan @krishna2]

@karpathy Ouch, I feel seen! Tsundoku shelf is just too big at this time to even look at it. Oh wait, a shiny new book there!

[Krishna Srinivasan @krishna2]

@keerthanpg Ignorance (avidya) is the root. Which causes unhappiness (apoornathvam). Which causes Attachment (Raga/Dvesha).

[Krishna Srinivasan @krishna2]

@YiTayML Deep respect for you @YiTayML ! Reminded of Nietzsche's “He who has a why to live for can bear almost any how.”. Stay sane and best wishes my good friend!

[Krishna Srinivasan @krishna2]

@XiaohuaZhai @giffmana Sent you email internally. :)

[Krishna Srinivasan @krishna2]

@karpathy https://t.co/WwSo13NWY7

[Krishna Srinivasan @krishna2]

@YiTayML Huge Congrats Yi and team. Well done. Can't wait to see what else you are all up to next.

[AToMiC@TREC2023 @TREC_AToMiC]

We will be presenting our work at a poster session during #SIGIR2023! Can't wait to connect with everyone in person 😎.

A fantastic journey collaborating with the brilliant minds of @cadurosar @rafasamrez @krishna2 @mad_astronaut @sclincha @lintool

👉 https://t.co/hx2nrjKoO3

[Carlos Lassance @cadurosar]

We also have @mattjustram presenting AToMiC our cross modal resource paper at #SIGIR2023 in collaboration with @lintool @mad_astronaut @krishna2 @rafasamrez and @sclincha

[Miriam Redi @mad_astronaut]

For more information the full guidelines are available https://t.co/PLUFIqtwaC and organisers will be more than happy to help with any problem / answer any question (just tweet or send an email to [email protected])

[Miriam Redi @mad_astronaut]

The final test topics will be released soon, but there’s a lot of material to start playing with" such: embeddings of openCLIP models + dev topics with human judgements over the baselines (https://t.co/ha75jGzmFJ).

[Miriam Redi @mad_astronaut]

Interested in Cross-Modal retrieval? Feel that evaluation on caption datasets is limited and noisy?

The @TREC_AToMiC task is here: a real, large scale problem (retrieve illustration for @Wikipedia sections), with human judgements. Deadline in 45 days!

@krishna2 @lintool