Devang Sachdev

I’m excited to support @govclab initiative with the Mensarius Oath, fostering a more ethical and conscientious investment landscape.

By adopting this oath, investors pledge to uphold integrity, fairness, and responsibility, ensuring that technological advancements benefit humanity as a whole.

In 2024, global venture capital investment in AI startups surged to $131.5 billion, a 52% increase from the previous year, highlighting the immense potential of AI to enhance productivity and transform industries.

@barrald @adamlisagor @sandwich This is one the best things I’ve seen in a while!!

I still can’t believe we actually made this. Thank you @adamlisagor and @sandwich for the amazing work!

We are honored to be named one of America's Best Startup Employers by @Forbes 😊. This award recognizes startup companies that carry out groundbreaking work, invest in employees, and demonstrate strong growth. Read more 👇

After all, data and language are quickly becoming the interface for programming.

A few key capabilities for LLMOps are: 1. foundation model fine-tuning, 2. iterative training data development, 3. programmatic labeling, 4. weak supervision 5. prompt experimentation, 6. prompt chaining, 7. data synthesis and augmentation, 8. knowledge distillation

It would take 355 years to train GPT-3 on a single NVIDIA Tesla V100 GPU. Fine-tuning LLMs doesn't take the same amount of compute or data but is by no means a lightweight task. Unsurprisingly, this has led to the emergence of #LLMOps.

2022 was a big year for team Snorkel: we tripled our customer base, introduced programmatic labeling and foundational model innovations, and significantly expanded integrations across the ML stack -- accelerating AI development by 10-100x. 📈businesswire.com/news/home/2023…

1/ ‘22 was a big year for @SnorkelAI! We - Introduced new foundation model & data-centric AI paradigms - More than tripled our customer base, incl. 5/10 top US banks + top-10 in 7 other industries, accelerating their AI dev by 10-100x+ - Welcomed 40 new Snorkelers And much more!

This week @eileenguo @techreview wrote an in-depth story on how photos of a woman on the toilet recorded by Roomba to train ML models ended up on Facebook. Here is how programmatic labeling can minimize the data exposure risk. snorkel.ai/how-programmat…

True data privacy is not possible with legacy outsourced manual labeling- as this unfortunate example accentuates... technologyreview.com/2022/12/19/106… Programmatic labeling is not just faster and more adaptable- it's private! @SnorkelAI

Also, register for a free virtual summit that will bring together speakers from @Google @CohereAI @StabilityAI @huggingface @Stanford @SnorkelAI and more to discuss enterprise applications of foundation models and large language models (LLMs). snorkel.ai/event/foundati…

In 2022, Snorkel researchers and our academic partners published seven papers pushing the boundaries of foundation models and LLMs for enterprise use. Here is a short summary of each: snorkel.ai/seven-research…

In an environment where every organization is trying to do more with less, it’s exciting to hear a @SnorkelAI customer say, “Snorkel Flow will literally help us 2x-4x the number of projects we can take on in 2023!!” More on the impact of Snorkel Flow: snorkel.ai/case-studies/

Delighted to announce @SnorkelAI’s partnership with @AimpointDigital to build and scale data-centric AI solutions for biopharma, manufacturing, retail, and other major industries👇snorkel.ai/snorkel-ai-par…

Read more about Foundation Models entering the #datacentricai era by Chris Re and @simran_s_arora: hazyresearch.stanford.edu/blog/2022-10-1…

Snorkel AI’s view of leveraging data is the way to go--you need a programming abstraction, a way to express, combine, and iteratively correct different sources of data and supervision signal, & ultimately, a way to train a smaller, faster, cheaper deployable model for your task.

The rise of foundation models like #gpt3, CLIP, #dalle2, #imagen, #stablediffusion, and many more has been amazing and fun. But getting models into production is about your data. Data is the only thing that doesn’t commoditize.