Asha Vishwanathan

AI / Data Science / ML + Product + Startups , Artist, Sustainable Living, Blogger,

[Asha Vishwanathan @ashavish]

@SaiDrGaddam Haha! Good one. Looking forward to see the list of AI generated signs you look for :)

[Asha Vishwanathan @ashavish]

@sai is doing some great work redefining education from the way we know it!

[Sai Gaddam @SaiDrGaddam]

College as we know it is most likely going to be obsolete in about 5 years.

The early signs are here in terms of employability.

If you are in tech, you already see what's happening to cognitive work.

Everything that is neatly measurable and done on a computer is going to be automated in a few years. And yet our entire schooling  pipeline, from kindergarten to board exams to coaching dungeons to college, is built to implant exactly that kind of MEASURABLE learning into our children.

But what is measurable is about three years away from automation.

The great news is that this could be a generational opportunity to let kids get their back and free them from obsolete classrooms and coaching dungeons. It is finally time to go back to MEANINGFUL learning.

And bring back the natural order of learning, the way we learn best.

MEANING > MOTIVATION > MECHANICS > MEASUREMENT

https://t.co/yrN3jl6ABX

[Asha Vishwanathan @ashavish]

Very good intiative in education. Definitely gives us pause to rethink our current education system

[Sai Gaddam @SaiDrGaddam]

Here's a story of how the Bhagavad Gita led to us setting up an apprenticeship program.

We at Comini, along with https://t.co/OEaWUU5vw8, are going to fund bright, passionate students and enable meaningful real-life learning across domains.

PLEASE SHARE THIS WITH COLLEGE STUDENTS YOU KNOW 🙏

In effect: Cheaper, better, faster real-life learning from the real world.

You can read the story here:
https://t.co/r7jJZHbfRB

Interested in apprenticeships or mentoring? Please fill this form: https://t.co/6GSS73dw3B

And here’s the story! 🧵

https://t.co/r7jJZHbfRB

[Asha Vishwanathan @ashavish]

Doing RAG on PDFs and other gnarly document formats is no walk in the park.

Using markdowns in our ingestion pipeline has been a game-changer for many of our use cases — clean, simple, and surprisingly effective.

https://t.co/0r7aLhkoPJ

[Sai Gaddam @SaiDrGaddam]

We just received 2,500 job applications. Fewer than 100 filled a form with open-ended questions. About 15 went on to complete a reasonably simple real-world challenge.

We are a nation with the cheapest internet and a billion internet users.

Where are all the self-directed learners? The eclectic Ekalavyas?

PS: Here's the link with the form and challenge https://t.co/eUT4ONCm3W

[Asha Vishwanathan @ashavish]

Cool stuff ! This sounds like LLM powered RPAs
https://t.co/I7zyDD0DSD

[Asha Vishwanathan @ashavish]

Jaw droppingly good !

[ElevenLabs @ElevenLabs]

Here’s an early preview of ElevenLabs Music.

All of the songs in this thread were generated from a single text prompt with no edits.

Title: It Started to Sing

Style: “Pop pop-rock, country, top charts song.”

[Siva Narayanan @K2_181]

You can find the recording of this talk by @ashavish here. It is a must-watch for anyone looking to build production-grade features using LLMs.

https://t.co/MftpvYP2Eb

[Asha Vishwanathan @ashavish]

@K2_181 Thanks @K2_181 !

[Asha Vishwanathan @ashavish]

@K2_181 Siva runs an awesome talk series on Saas Engineering ( Check out https://t.co/tuglXwSDcR)

Join me for a talk on his channel on Dec 21, where I talk about our learnings as we deployed LLMs to Production for https://t.co/diJtzPo4TV !

[Siva Narayanan @K2_181]

In the next SaaS Eng talk on Dec 21, @ashavish, who conjures up LLMs at @verloopio, talks about her experience taking LLMs all the way to production in a SaaS product.

Register here to get an invite lest you are made obsolete by AI!

https://t.co/jNJ58fLomS

[Asha Vishwanathan @ashavish]

This looks like a really neat idea !

[elvis @omarsar0]

LLMs as Optimizers

This is a really neat idea. This new paper from Google DeepMind proposes an approach where the optimization problem is described in natural language.

An LLM is then instructed to iteratively generate new solutions based on the defined problem and previously found solutions.

It was first tested on linear regression and the traveling salesman problem. Leveraging LLMs with simple prompting match or surpass hand-designed heuristic algorithms. This shows good potential for using LLMs as optimizers.

The idea is then applied to prompt optimization that aims to maximize task accuracy on different tasks like math word problem-solving.

The first piece of the proposed meta-prompt takes in previously generated prompts along with corresponding training accuracies. The second piece includes the optimization problem description with samples obtained from a training set representing the task.

At each optimization step, the goal is to generate new prompts that increase test accuracy based on the trajectory of previously generated prompts.

The optimized prompts outperform human-designed prompts on GSM8K and Big-Bench Hard, sometimes by over 50%!

For math word problem solving, one of the most effective instructions found begins with "Take a deep breath and work on this problem step-by-step".

https://t.co/GsF8fzjevX

[Asha Vishwanathan @ashavish]

This is awesome 🔥. Llama 2 seems to be doing better on the benchmarks for 7B models as well ! https://t.co/CsOUnj87mP

[AI at Meta @AIatMeta]

We believe an open approach is the right one for the development of today's Al models.

Today, we’re releasing Llama 2, the next generation of Meta’s open source Large Language Model, available for free for research & commercial use.

Details ➡️ https://t.co/vz3yw6cujk

[Asha Vishwanathan @ashavish]

This is some crazy speed ! OpenAI just announced Functions today

[LangChain @LangChain]

🚨🤖New Agent Release🤖🚨

We can @OpenAI's new function parameter to create a new type of agent (`openai-functions`) now available in Python and JS

Links to documentation a thread on what went into it below 👇

[Asha Vishwanathan @ashavish]

Good stuff . Vicuna is definitely coming on top. We found 13b unquantized model outputs are quite good for fact based outputs, but dont do so great for "creative" outputs in comparison to chatGPT. Latency however is still an issue.

[Graham Neubig @gneubig]

There are so many chatbots nowadays, it’s hard to keep up!

To help out, we made an open source tool for automatic comparison of chatbots, and created a report on LLaMa, Alpaca, Vicuna, ChatGPT, Cohere, etc.!

Report: https://t.co/n3JfKYcIXB
Browser: https://t.co/kOK4YsNHBq

🧵⬇️

[Gaurav Singh @sgaurav]

Pause and think for a second, 689%!

LLMs will change the way industry manages customer support.

[Gaurav Singh @sgaurav]

CBSE board exam admit card now has a mention of ChatGPT 😳

[Asha Vishwanathan @ashavish]

*Correction - Multi tenant optimization for few shot task oriented FAQ retrieval

[Asha Vishwanathan @ashavish]

Was awesome to present our paper - "Multi tenant optimization for few shot talk oriented FAQ retrieval" for Industry track at EMNLP 2022 ! Thanks to all for stopping by at our poster session !
@rajeevuwarrier @gauthamsuresh09 @LazyLearner_

#EMNLP2022

[Asha Vishwanathan @ashavish]

Thrilled to have our paper on "Multi-Tenant Optimization For Few-Shot Task-Oriented FAQ Retrieval" accepted to EMNLP 2022 Industry Track !!!
Preprint coming soon ! #EMNLP2022

@rajeevuwarrier @gauthamsuresh09 @LazyLearner_

[Asha Vishwanathan @ashavish]

@emnlpmeeting Have the notifications of acceptance for the industry track been sent out ?