Science Festival - 2026 - Notes


Whats next in AI, Spatial Intelligence and World Models


How ai is moving beyond data ?

Next frontier - spatial ai

Generative ai - Generate a photograph/image in the horizon hall

Ai does not understand 3d space. Example - The length canal in front of TajMahal from the image

spatial reasoning
Ability of machines to generate and act, perceive, reason in 3d space and time

3d reconstruction (something like panorama view of physical world in an image/photograph)

VR/AR automated dacing video
how we digitally represent things happening in physical world (ai generates a video). Spatial structure of the world. Give a list of commands to follow, to generate the video

Data for generative AI
1 to 100s of images
These days less data to do the same thing

AI keywords:
multimodal AI
general intelligence
super intelligence
could spatial ai, be one next field ?
4d reasoning is critical
spatial reasoning ability
depth sensor
plenotic function
3d/4d gaussian splatting
dynamic radiance field
understanding/reasoning
action models
special models
Spatial Understanding

Generative AI generates/deals with
- language (generate a poem)
- image (generate an image with required specification)
- video (Generate a dancing video, which later may be on youtube)
- audio

there is no  need of embodiments


From model to market, building scable ai setup startups in India



Panel Discussion - Questions about AI
 - Turning ai ideas into innovative products?
- How ai startup grow from concept to market?
- LLM is cheaper but how to extract value out of it? What part of LLM are we going to use? What if the LLM does not help to solve your problem? (If you are working on the edge case which the platform does not solve. For example - Build smaller SLMs and finetune for Indic language. Get compliance in every conversation and audit trail.)
- What a model might do and startup might require?
- The open source are cheaper how to generate revenue?


Some Points:

*  Regardless of what problem you want to solve ai will help
*  Some indian companies are also developing foundation models, as they may be finding some insufficiency in the out of box models for their requirements
* Foundation models give you better starting point
* Don't say I am building foundation model. Its a huge work. How do you build the models well ? How will we differentiate their wellness?
* Models are widely available, very hard to build. By understanding the problem in deeper way.
* Starting today, building a model is not easy. Is the problem large enough? Do you really wish to solve it?
* Why won't google do it tomorrow? It is often a misconception. Large companies like Google, Microsoft do not focus on narrow markets (white spaces). Bigger companies, broader platforms. The companies working on narrow markets should look to differentiate their products. If you have an idea, is problem relevant?
* Even the large companies were initially small. That is the advantage. What matters is vision.
* Get deeper insights about the customer.
* From B2C, working together for "lockscreen",
inmobi made it into a business
* superstars ai - you can create your own virtual cricket match
* Where are indian startups doling well ? Multiple industries (education, healthcare, deeptek, security, general pharmaceutical etc). AI solution exist at lower cost.
* Large models, pushing the boundary towards the big companies
* How do we bring solutions to Indian market?Lot of problems to solve in india. Indian entrepreneurs are doing well in b2b.
* Its nice to see prof. starting startups academics joins. No shortage of talent.
* Indian companies do not value research enough
* Truth - If roti, kapda, makan is sorted, go for entrepreneurship, Entrepreneurship involves delayed gratification. I need to have a support system at home. India is celebrating entrepreneurs. Team, investors to manage. Every moment you hit a new milestone.
* AI as an enabler/tool. Don't think you are in competition with ai. The question is, what to make?

Q and A
Q - BFSI (Banking, Financial Services and Insurance) Quantify benefit which ai will offer in case we migrate it from legacy system?
A - Energy and money is reducing, watch the trend. Check the cost curve, Future proofing, Have sufficient generalizability

Q - Is there any study if ai startups fail %
A - Very small % startup succeed. That's the nature of the game. Failed entrepreneur is oxymoron. Be willing to take on ambitious goal. The only way to succed is to take risk. Be open.



Inaugration Speeches

IISER is non profit
Goal - Make India among top 3 science and technology sector
Fast India - Is it a start up ? - Yes

- We do not celebrate science enough
- We need to answer the questions like how science happens
- We need to bring scientists face to face with young crowd

We think that science has power. Power to know. Power to answer how intelligence work. Science push economics.

debate - Is the world an illusion
prizes - Young NRI who did exemplary work in science
How ISRO came into being ?
Operation Sindoor - Defence research
The man who knew infinity - Book


Prof. Karandikar

Last two years initiatives
Fast India -
Establishment on Anusandhan NRF
Research and development fund
Technology development and commertialization
Translate research into technology
Commertialize technology
Boundaries between the research is disolving with interdisciplinary research
AI,
Enhance the public trust in science and scientist

Prof. Vijay Raghvan
Between solution to problem and answer to a question, answer is simple, solution is hard
Issues to be addressed: waste, pollution, traffic

Research institutions to be connected to world

Prof. Dhavan
- Bajaj - Scooter is sold in Latin America and ... -

Bajaj Representative
- Believe in people
- Bajaj facilitates research in their own premises


Building AI that powers and protects the digital workspace


Girija Narlikar

What is Google workspace?  "meet + drive + email + ... "
How do we keep users safe in the era of ai?
Too much email - AI as personal assistant?
Assists to create email replies better.
The current generation communicates in a short video.
What you can do with ai in email?
Example - Consumer email video. Echo friendly cleaning product. Launch the product
Use foundation model (LLM) - Gemini
Get improved productivity space with AI
AI stack of Gemini looks like
(
1. foundation model
2. google cloud
3. hardware (TPU hardware specialized to run the AI
)
what happens behind the scene when people build foundation model? - Look for an answer.
multimodality (video, audio, text, images) slides build by gemini

Explore
- genie3
- video 3m1
- gemini robotics  (instruct the robot)

AI
- multimodal understanding
- deeper reasoning
- powerful agentic features
- superior coding and development

Some Points:
* With a powerful foundation model, we can improve all the products
* Emotions are being replicated in different language
* Agentic Features - automated ways of connecting different app
* Foundation model to generate the question
* Foundation model to generate the answer

how do we manage risks
 1. content safety, unsafe cotent
2. security

AI security vocabulary:
- prompt injection attack
- lab's user

Security Steps for a Task - summarize the email
1. input data
2. classifier (LLM of smaller size)
3. runtime instructions
4. Red teaming (experts in security)
5. model resilience (deepmind team improves resilience to security attacks)

Who are bad guys ? - state sponser actors to get intellectual property.
Is the email safe to keep in users inbox?
Lot of efforts goes into identifying spam.


Collaborative robots (Cobots)

Shared tasks
Human machine interactions

Application:
prothetic hand
survilance
surgery
drones collaborating in shapes
















Comments

Post a Comment

Popular posts from this blog

Robotics Ideas

Robotic Cleaning - Observe the drain blockage from outside of the pipeline - Need to develop something like X-Ray machine. Which will monitor the drainages for any blockage from outside. There should not be any road digging required, to find the blockages in the underground pipelines. - Robot to clean small (1 foot wide) as well as big (10 meters wide) drainages from the blockages or garbage thrown into it - Water treatment plants wherever the water gets drained into sea or river - Garbage collection, categorization, recycling and disposal system across the countries - Robot to clean the existing garbage thrown in the river/sea - Cover any open drainages - Educate people over garbage collection, categorization, recycling and disposal system. May be in the schools. Project Description A robotic arm which is able to clean a wide (50m x 50m) dry area, with dry garbage A robotic arm which will pick up and segregate the below matter    Plastic bags         ...
Read more

Lego Mindstorm - Experiences

  When I first started working on Lego, I never felt "lacking help", although I was not "fully equipped" per say. I figured out my answers on my own as I continued working. I worked on various experiments on my own first. Then I started teaching young children. But never did I feel lack of help or doing work alone. The help on the internet was very good, the forum helped and to some extent it was my thinking. I think I started working on it from 2012. I continued working on it till 2023...11 years ...What a creation "Lego Mindstorm" is ! Ideally, we can use it from a young age of 10+ years till your Masters or so. I started working from 36 years of age ! But none the less, it was fun. Some experiments like "Maze Solver", "Wall Follower" or "Printer" are really challenging. The Building Instructions are ready. What I did was mainly programming. Ideally I could have appreciated even more, if some "building specifications...
Read more

My Life As a HouseWife - Part 2

I spent the initial few weeks after lay-off at home. Preparing for interviews. Soon Hiren got me a laptop and a Lego Robotics kit. I did not understand the use of Lego Robotics kit at all. Now, after 11 years of working on the Lego Robotics kit,  I realized I must be naive to say that. However, this kit helped me in my life to educate myself with basic Robotics and also to judge my potential to teach young students. Even though I got the kit in 2008, I started working on it only from 2012. For these 4 years in between, I was literally desperate about following other options like, getting a job and earning some money, applying for GRE and getting further education, writing my own "music" application for mobile and publishing it on Google Playstore (as mobile was new to me that time). But somehow nothing clicked for 4 years. And then I started taking Lego seriously. A lot of its creation material is on the internet. Once I got a little hang of following their instructions and a...
Read more