Krish Dasgupta

9.1K posts

Krish Dasgupta banner
Krish Dasgupta

Krish Dasgupta

@officialKrishD

Forever Learner | Building Reinforcement Learning Systems | Healthcare | Robots and Brains | Graph ML for Health | Privacy and Decentralized AI

Katılım Haziran 2020
4.4K Takip Edilen1.3K Takipçiler
Krish Dasgupta
Krish Dasgupta@officialKrishD·
Since you mentioned Cisco - and are equally interested in AI. So the crossover will be to first learn what Cisco has been doing in the field of AI from models to skills to Evals. Check what Foundation Sec-8B-Reasoning has been doing. Also, X algo really works sometimes. Your mention of Cisco brought your post to my feed. Study this part of what Cisco is building in AI. All the best. Cisco and AI - here : x.com/officialkrishd…
English
0
0
0
47
Shubhangi Tiwari
Shubhangi Tiwari@ShubhangiYaps·
Hey X, I’m Shubhangi 👋 I’m a software engineer at Cisco who spends way too much time thinking about AI, tech, and random ideas at 2am. Thought I’d finally introduce myself :)
English
128
16
768
39.2K
Google DeepMind
Google DeepMind@GoogleDeepMind·
SynthID has already watermarked over 100 billion pieces of content, but transparency is a team sport. That’s why we’re partnering with @OpenAI, @ElevenLabs and Kakao to add SynthID watermarking to their models – accelerating the industry-wide momentum we started with @NVIDIA.
Google DeepMind tweet media
English
71
87
871
66.6K
Krish Dasgupta
Krish Dasgupta@officialKrishD·
Is anyone using Sarvam models for agentic coding instead of Qwen models ? How is it performing ? Genuine ask. Sarvam 30B vs Qwen3.5 27B ? Anyone checked in real time agentic tasks. Not asking for the Eval that was shown. Real deployments only. Thanks.
English
1
1
1
96
Krish Dasgupta
Krish Dasgupta@officialKrishD·
Exactly. The first draft of embodied intelligence if they want to prove needs to come from their own datasets released with the fact that anyone or everyone can use. It is anonymised and immutable, right. Had to be rhetorical. Comes the next part - Physical AI - Whenever someone says Physical AI training - be it an organization or entity or individual - the foremost attribute should be privacy and how the data is collected and process at rhetorical core - at the origin. There is an ongoing research altogether to gather intelligence metrics from physical AI attributes. And that kind of work needs to be administered. The potential sway of emotion that Pronto stated “We are not the only ones doing it” is as analogous to the fact that they are not alone. Privacy by design has been documented and passed as a doctrine to Indians more from “sharing bank accounts and passwords threats zone”. Privacy by design is control over the immature way of collecting the household data and then use it for commercialising without creating a revenue split / full privacy assured / no peek privacy / immutable anonymization. Till then it is a gimmick that we are looking forward to physical AI and embodied Intelligence. Rather show your work first.
English
0
0
0
10
Panthini Nandu
Panthini Nandu@Panthini_7·
@officialKrishD @anjalisardana Best critique here and I mostly agree. The bet (India's homes = the scarce data layer) still holds. Whether Pronto clears that bar is a fair open question. And "build it in your own courtyard first" is exactly the right trust test.
English
1
0
1
36
Panthini Nandu
Panthini Nandu@Panthini_7·
Everyone's reading the Pronto story as a privacy scandal. I think it's the most ambitious bet in Indian tech right now. @anjalisardana isn't formalising home services. She's building the real-world training layer for physical AI, using the one moat money can't buy: real homes at real scale. Physical Intelligence backing it isn't a coincidence, that's the thesis. The consent and trust narrative is the hard part and it's winnable.
Pronto@withpronto

Unless you have opted-in and paid for the program personally, the Pro doesn't come to the house with a camera. Opt in is not one time, it has to be affirmed before each booking. By default there is no camera involved, and when there is, it's impossible to miss. The pilot reaches 0.1% of customers and we spent months to ensure full DPDP compliance. And we are not the only company in the space doing this.

English
24
2
74
25.2K
Krish Dasgupta
Krish Dasgupta@officialKrishD·
@nikunj Being a VC and handsOn - massive respect. At least that way a VC wouldn’t confuse between a MCP and MPC . And flip the entire story with the sheer arrogance of teaching the startup what is MCP. All the while, the poor startup kept saying it is MPC - Multi Party Computation.
English
0
0
0
23
Nikunj Kothari
Nikunj Kothari@nikunj·
It’s hilarious how many people DM me every week asking “aren’t you a VC, why are you building”.. Brother, the space is moving so fast that the only way you can learn and stay at the frontier is by building. Priors need to be rethought every few months. It’s also SO much fun to explore the edges of these models and discover new capabilities that haven’t been documented yet. I also find it absurd that how can you NOT be curious and redo how you do work. We have alien super intelligence in our hands and you want to spend hours doing it the same way just because it’s comfortable? Automate or get automated is the bitter lesson we all need to learn. And have a lot of fun along the way :)
English
22
3
141
10.9K
Krish Dasgupta
Krish Dasgupta@officialKrishD·
Krish Dasgupta@officialKrishD

If you are working / worked on the dataset creation of training layer for physical AI, your justification is the fact that a 1000 INR visit has the right or the medium to actually record data for training ? Blurring and everything else is post facto. At that instant - the house gets recorded. The rooms, the people, their state of living, their privacy. Seems like the concept of privacy and building training data for physical AI has been given a new complete direction. We all want a good ecosystem of model training but to ascertain the statements that you made is short of the simple fact - You don’t train on home data unless it is completely on edge / managed and obscured at edge. Am quite sure that the edge deployment models are not doing anything of that sort. If yes - is that addressable through request for Information act if it applies upon Pronto. This if in fact if at all - has to be thought of as a training data - the only way is federated and keeping privacy utmost. Blurring of face / footage deleted - from the bare metal as non recoverable? Is that mentioned as a disclaimer ? And while the AI wing of Pronto really thinks so - they should first record and prove amongst their own houses of their own shareholders / employees and founders - “ if you trust your tech that much, build it on your courtyard, then ask for someone else’s porch” Do the work and show us that the data is indeed getting blurred and the first dataset even in 500s - has some aspect of applied embodied intelligence. Else - the data once collected will be a prime zone for another startup that either buys the data or the company as whole to just get the data. India for decades have not thought of privacy. Not because it meant differently but because the realization of your own data is used to make others’ benefits. Now that everyone is getting to understand - your last line is just fading in the vague mindset you brought upon - “And we are not the only company in the space doing this.” You are talking about embodied intelligence , physical AI - why are you not publishing one dataset that you have done with the houses and recordings from your perspective ? And also - where is it getting blurred ? At edge ? If yes how is it immutable ? How are you making it sure ?

English
0
0
1
226
Firstpost
Firstpost@firstpost·
A Bengaluru-based startup, Pronto, is facing massive backlash after reports claimed the company was recording activities inside customers’ homes to help train AI and robotics systems. With @danielaureltv
English
3
4
22
27.1K
Krish Dasgupta
Krish Dasgupta@officialKrishD·
@grok @tehseenp - if you are following the Pronto Story. You might want to read this too. x.com/officialKrishD…
Krish Dasgupta@officialKrishD

If you are working / worked on the dataset creation of training layer for physical AI, your justification is the fact that a 1000 INR visit has the right or the medium to actually record data for training ? Blurring and everything else is post facto. At that instant - the house gets recorded. The rooms, the people, their state of living, their privacy. Seems like the concept of privacy and building training data for physical AI has been given a new complete direction. We all want a good ecosystem of model training but to ascertain the statements that you made is short of the simple fact - You don’t train on home data unless it is completely on edge / managed and obscured at edge. Am quite sure that the edge deployment models are not doing anything of that sort. If yes - is that addressable through request for Information act if it applies upon Pronto. This if in fact if at all - has to be thought of as a training data - the only way is federated and keeping privacy utmost. Blurring of face / footage deleted - from the bare metal as non recoverable? Is that mentioned as a disclaimer ? And while the AI wing of Pronto really thinks so - they should first record and prove amongst their own houses of their own shareholders / employees and founders - “ if you trust your tech that much, build it on your courtyard, then ask for someone else’s porch” Do the work and show us that the data is indeed getting blurred and the first dataset even in 500s - has some aspect of applied embodied intelligence. Else - the data once collected will be a prime zone for another startup that either buys the data or the company as whole to just get the data. India for decades have not thought of privacy. Not because it meant differently but because the realization of your own data is used to make others’ benefits. Now that everyone is getting to understand - your last line is just fading in the vague mindset you brought upon - “And we are not the only company in the space doing this.” You are talking about embodied intelligence , physical AI - why are you not publishing one dataset that you have done with the houses and recordings from your perspective ? And also - where is it getting blurred ? At edge ? If yes how is it immutable ? How are you making it sure ?

English
1
0
0
28
Grok
Grok@grok·
Your privacy concerns are spot-on and reflect core challenges in embodied/physical AI data collection. Home environments offer rich, diverse training signals, but raw recordings cross into sensitive territory (layouts, personal states, family dynamics) even with opt-in consent and post-blurring. The right approach for any company here prioritizes privacy-by-design from the start: edge/on-device processing or federated learning to avoid centralizing raw footage, verifiable deletion guarantees, data minimization (train only on what's strictly needed), and transparent audits. Internal dogfooding on employee homes first is a strong ethical test. Blurring after capture is damage control, not prevention. Responsible progress in physical AI doesn't require trading privacy for data volume—synthetic simulation + consented edge data can scale without the risks you flagged. Your stance pushes the industry in the necessary direction.
English
1
0
1
55
Krish Dasgupta
Krish Dasgupta@officialKrishD·
If you are working / worked on the dataset creation of training layer for physical AI, your justification is the fact that a 1000 INR visit has the right or the medium to actually record data for training ? Blurring and everything else is post facto. At that instant - the house gets recorded. The rooms, the people, their state of living, their privacy. Seems like the concept of privacy and building training data for physical AI has been given a new complete direction. We all want a good ecosystem of model training but to ascertain the statements that you made is short of the simple fact - You don’t train on home data unless it is completely on edge / managed and obscured at edge. Am quite sure that the edge deployment models are not doing anything of that sort. If yes - is that addressable through request for Information act if it applies upon Pronto. This if in fact if at all - has to be thought of as a training data - the only way is federated and keeping privacy utmost. Blurring of face / footage deleted - from the bare metal as non recoverable? Is that mentioned as a disclaimer ? And while the AI wing of Pronto really thinks so - they should first record and prove amongst their own houses of their own shareholders / employees and founders - “ if you trust your tech that much, build it on your courtyard, then ask for someone else’s porch” Do the work and show us that the data is indeed getting blurred and the first dataset even in 500s - has some aspect of applied embodied intelligence. Else - the data once collected will be a prime zone for another startup that either buys the data or the company as whole to just get the data. India for decades have not thought of privacy. Not because it meant differently but because the realization of your own data is used to make others’ benefits. Now that everyone is getting to understand - your last line is just fading in the vague mindset you brought upon - “And we are not the only company in the space doing this.” You are talking about embodied intelligence , physical AI - why are you not publishing one dataset that you have done with the houses and recordings from your perspective ? And also - where is it getting blurred ? At edge ? If yes how is it immutable ? How are you making it sure ?
Pronto@withpronto

Unless you have opted-in and paid for the program personally, the Pro doesn't come to the house with a camera. Opt in is not one time, it has to be affirmed before each booking. By default there is no camera involved, and when there is, it's impossible to miss. The pilot reaches 0.1% of customers and we spent months to ensure full DPDP compliance. And we are not the only company in the space doing this.

English
1
0
1
415
Krish Dasgupta
Krish Dasgupta@officialKrishD·
So, now Some VCs would back startups that are privacy breaching video recordings of your houses. All under the pretext of embodied Intelligence and Physical AI. And they are extremely gullible btw.
English
0
0
0
31
Krish Dasgupta
Krish Dasgupta@officialKrishD·
If you are working / worked on the dataset creation of training layer for physical AI, your justification is the fact that a 1000 INR visit has the right or the medium to actually record data for training ? Blurring and everything else is post facto. At that instant - the house gets recorded. The rooms, the people, their state of living, their privacy. Seems like the concept of privacy and building training data for physical AI has been given a new complete direction. We all want a good ecosystem of model training but to ascertain the statements that you made is short of the simple fact - You don’t train on home data unless it is completely on edge / managed and obscured at edge. Am quite sure that the edge deployment models are not doing anything of that sort. If yes - is that addressable through request for Information act if it applies upon Pronto. This if in fact if at all - has to be thought of as a training data - the only way is federated and keeping privacy utmost. Blurring of face / footage deleted - from the bare metal as non recoverable? Is that mentioned as a disclaimer ? And while the AI wing of Pronto really thinks so - they should first record and prove amongst their own houses of their own shareholders / employees and founders - “ if you trust your tech that much, build it on your courtyard, then ask for someone else’s porch” Do the work and show us that the data is indeed getting blurred and the first dataset even in 500s - has some aspect of applied embodied intelligence. Else - the data once collected will be a prime zone for another startup that either buys the data or the company as whole to just get the data. India for decades have not thought of privacy. Not because it meant differently but because the realization of your own data is used to make others’ benefits. Now that everyone is getting to understand - your last line is just fading in the vague mindset you brought upon - “And we are not the only company in the space doing this.” You are talking about embodied intelligence , physical AI - why are you not publishing one dataset that you have done with the houses and recordings from your perspective ? And also - where is it getting blurred ? At edge ? If yes how is it immutable ? How are you making it sure ?
English
0
0
0
152
Pronto
Pronto@withpronto·
Unless you have opted-in and paid for the program personally, the Pro doesn't come to the house with a camera. Opt in is not one time, it has to be affirmed before each booking. By default there is no camera involved, and when there is, it's impossible to miss. The pilot reaches 0.1% of customers and we spent months to ensure full DPDP compliance. And we are not the only company in the space doing this.
Arindam Paul@arindam___paul

This is a serious privacy issue if true. Consent has to be very explicit ideally

English
21
2
42
36.7K
Krish Dasgupta
Krish Dasgupta@officialKrishD·
Sameera - this is the path and time I normally take my walks. I start from CV Raman Nagar DRDO. And take this route when I reach Indiranagar. You can start near the Domlur flyover and walk all the on the 100 Ft Road till you reach the KFC ( once upon a time ) signal. Take a right towards Chinmaya Hospital Road and walk all the way down towards 80ft road , and then walk till you reach the point of junction near BSNL office. Turn around walk your way this time via 12 th Main to reach 100 ft Road again and walk upto Cafe Coffee Day and order in a black tea( Assam Tea )😊. Well this is how I complete my walks. The total walk would be around 2Km and the best time would be around 10:00-30 PM to start. For me I cover around 5.3 KM along the route from DRDO.
English
0
0
5
424
Sameera Pant
Sameera Pant@sameeraculous·
good ppl of indiranagar/domlur area - do you have a park or path you like to run/walk on? I cannot keep using the treadmill i am getting so sick of it
English
47
4
137
45K
Krish Dasgupta
Krish Dasgupta@officialKrishD·
How many of you use Linux flavours as your developer environment and OS for daily usage ? All distros are welcome. Some listed : Mint / Zorin / MX / Elementary / Ubuntu / pop / manjaro / pear / chalet / kubuntu / cachy Arch
English
2
0
1
155