Some Amazing startups trying to solve the Data Quality/ Data Governance/ Data Cataloging issue -
1)First, Made in India -
@AtlanHQ
.
@prukalpa
wrote an amazing blog on data catalog 3.0. -
they have inMobi as a client 🧐
Karnataka charges the highest road tax (2.5x the national average),so that bbmp can dig up the same road 5 times a year.
Bangalore’s infra is worse than a tier 3 Indian city.
@vaibhaw_vipul
model building has always been easy. value lies in turning models into data products that drive economic impact. It's about understanding the business context, integrating with systems, and iterating based on real-world feedback. You will always be paid handsomely for that.
@sourav_bz
They have tried in coimbatore. It works well for long distances but not for short distance. Getting to a atomic network takes a lot of resources. Also if it is open-source, who will be the customer support? Who will check for driver’s authenticity? It’s not that uber and ola are
Small career update: Joined
@SarvamAI
last month and have been heads down building. The chance to work with
@pratykumar
and
@vivekrag
was too good to pass on.
We are thrilled to announce our latest $41M funding round by our partners-
@LightspeedIndia
,
@peakxvpartners
, and
@khoslaventures
.
Our mission is to build full-stack Generative AI for India.
Press Release:
@sourav_bz
are extremely profitable firms. Even after all these years they have been loss making. eventual goal was to remove the need of having a driver through self driving. Which looks unlikely to happen in the near future (uber atg is done, and tesla looks the most likely to pull this)
@ganeshunwired
The solution is simple - neutering and vaccination. Please donate to the right NGOs in your area. Killing or putting them in shelters is not the solution.
People who support stray dogs do much more than people who just shit on twitter.
Finally finished Chip wars. Only 2 takeaways, go long TSMC and if anybody ever mentions anything about India becoming a chip manufacturing super power, ignore them.
@maheshperi
30% of 20L is year 1, year 5 it would be less than 5%. Nobody is going to IIM for year 1 salary, they are going for the trajectory it puts you on.
@nishanth_who
The loans for IIMs are extremely cheap (6-7%), which is as good as it gets in the current market.
As for where the money goes - Professors salary(check salary diff at iim vs iit), PhDs, guest lecturers (expensive), and infra (Top IIMs have solid infra).
Started embracing Occam's Razor a lot.
Best credit card: the one easiest to use, Amazon ICICI.
Best investment strategy: the simplest one, index.
Best tactics for a team in FIFA: just use the default.
Best K value for retriever: use grid search.
Got caught in a road rage incident today. Stopped the car right infront of the sarjapur police station. Guy was a native. He came and started shouting 🤣🤣 first got scolded by the lady constable, and then I showed the dashcam footage, got bashed by the officer also.
IIUC, this new Indian regulation only applies if you’re deploying your own models. OpenAI wrapper wale bhaiyya’s product that’s been in the making for 4 years, remains unaffected by it. IDK why he’s protesting so much? 🤣
@gopikm0
100% aligned with RBI on this. I still have a random loan on my credit history (still open), which I have no clue how it got in. Have asked the bureau as well and they said it is still active. Might be one of these shady fintech apps.
Langchain was sitting on one of the best ideas in the AI space for more than a year. Completely fucked up the user experience and will most likely now lose to OAI.
Avoid dating those with dysfunctional families. Whatever little I’ve seen, difficult childhood & troubled relationship with parents always reflects in people’s adult relationships. Not their fault they are victims of bad parenting, but it’s a repair project you can avoid.
Always fun working with friends. The base model is also out now on huggingface -
@chheplo
still waiting for the NDA from you🤣 thanks for the blind trust 🙏
Dhenu 1.0 - Pioneering domain-specific agriculture model, trained on high-quality conversational dataset, specially focused on Indian agriculture practices. Designed to be bilingual, trained on 300k instruction sets in English and Hindi.
OK, folks, have I got a treat for you today!
A guest post by
@larsiusprime
comparing web3 culture to the culture of Amway!
It's very long (but fun), so enjoy!
India needs more compute. Every college kid should have access to A100 compute available for free. Big part of ML engineering is just making the gpus go brrrr
@adadithya
I literally convinced a person yesterday not to buy ola scooter and today this happened. My simple pitch was it’s been 10 years and the ola app barely works, how would you trust them with an electric scooter/car.
Random dump of cool ML papers I read in the past 1 month -
Multi token prediction - why do we do only next token prediction? We can do multi-token. Training this way helps with better long range predictions, faster inference. We should expect this to become the norm soon
Since we got our last super intern in
@Kanakmi
, trying out here again. MPL is looking for multiple interns for the data science team. You will get to work on improving our Matchmaking system, CLM, and Fraud. We filed multiple patents also last year, and we do a mix of research +
The highly paid professionals staying in high-rises of Gurgaon have ruined the domestic help market in the city. Maids don't want to work in kothis/builder floors, they enjoy working in condos where all their friends work. They say pay is better there and there is lesser work!
Excited to announce we've raised 73.6M$ at 520M$ valuation, led by IVP, along with our seed and Series A lead investors NEA, Elad Gil, Nat Friedman. Honored to have NVIDIA, Jeff Bezos,
@tobi
, Databricks,
@naval
,
@rauchg
,
@balajis
partner in this round, among others.
@ravihanda
@GabbbarSingh
Sorry, he is a hate spewing account. Just because you are a founder doesn’t make it ok. Twitter makes the karma feedback cycle pretty short.
@aviralbhat
Solana mainnet was down for a day. ETH had a bug because of which a fork had to be down as half the network didn’t update. Atleast here facebook was able to intervene.
Finetuned a GPT-2 base model on Indian Languages.
IndicGPT -
It understands the language but throws out gibberish right now because I ran out of GPU credits. 🤗
Loaded the dataset and the training script as well if someone has the GPUs and time to burn.
Thrilled to unveil Sesame by
@Setu_API
in collaboration with Sarvam 🚀 India's 1st domain-specific LLM for financial services. It’s set to redefine banking—offering personalized advice, multilingual onboarding, and AI-driven insights. A new era of inclusive finance begins today!
So krutim started at the same time as mistral, trained a multi-modal foundational model, on their own hardware, and beats GPT-4 at a bunch of tasks?
Literally nothing adds up.
Want to thank
@brijbhasin
for giving me a chance to explore VC as a career path. I think we both saw enough the impact AI would have on the ecosystem and decided to just build in the space.
We’re live!
Blinkit customers in Delhi NCR, Mumbai and Bengaluru can now get the all new PlayStation® 5 Slim editions and controllers delivered in 10 minutes!
@RaoSumukh
My IT team had to change 3 macs of mine. They randomly stopped working. No way to fix as well, need to wait till enough macs stop working to send them all to apple service.
My first internship stipend was also 4k/month lol. Realised all these hustle internships are a waste of time without any hard skills. Went back to college, discovered Andrew Ng and
@Sentdex
, and wrote tf code.
Over the past 4 months we have been working on the first module of our SydeLabs () platform - SydeBox.
SydeBox enables you to perform automated red teaming of LLM applications and models. DM for access RT for karma. more 👇
Innovation in FAANG will go down drastically. People will choose to work at the cash-cow business (AWS, ads at google/meta etc), while innovation across the startup ecosystem will go up.( more people starting up, easier to hire good engineers/PM)
Just saw a lady go berserk on a pet owner for tying their dog outside in the heat to punish him. She told him she goes this route on a daily basis and if she saw this again she is going to share the video with AWBI.
We need more of this kind of crazy. 💯
@JustAnkurBagchi
@BlrCityPolice
This happened with me as well 3-4 years ago. I didn’t have money that time so they had to let me go. You should have called 100 then and there though.
As much as I love India & always have, I find upper-middle class Indians, especially elders, to be some of the most entitled & bigoted people that one will ever come across. Took forever and living in a gated community to realise this strongly. There, I said it.
Still reeling from the event, but most of the stuff we showed today is available to try out! Feedback is super critical for us. Happy to support interesting usecases with credits/compute!
I don’t like/watch cricket, but purely from a ROI perspective (time, money, emotions of a billion people), it is such a bad investment. Literally 6-7 serious teams and we still don’t dominate. If India would invest 1/10th the money into football we will be in FIFA WC in no time.
Ola is just Indian Uber.
Flipkart is just Indian Amazon.
Zomato and Swiggy are just Indian Doordash.
These were proven business models.
Just using technology does not make you a startup.
Looking to sell cutouts -
3k+ users. Didn't monetise because anarchy. Lots of chinese users because it is completely private. ML that runs on browser (I did this in 2020).
@ProductHunt
Top 5 of the day -
Single people in 30s with self-sustaining parents & no children have a massive edge over majority of the people in the same age group.
Financially, professionally, personally, emotionally, physical healthwise - every way!!
@_waabi_saabi_
Crazy dude crazy. Swiggy should have a recommendation system for this tho. They have all the data in the world. Crazy they haven’t build something like this
Watching the Bommanahalli metro trial is like seeing a baby take its first step.
I have spent the last 8 years of my work life living and working around the metro construction.
After inhaling much of the pollution during metro construction it feels like my sacrifice has come
Started some initial work on creating Indic-LM, fine-tuning StabilityLM(via LoRA) on indic dataset by AI4Bharat. Tokenised the dataset,uploaded it, trained an initial model, all for less than 5 Dollars on . (this is GPU instance btw).
Neat trick for filtering out high quality data for LLM training. Tried and recommended.(Ofcourse don't do 100k samples if dataset is domain specific) Use xgboost 🔥
One thing I noticed while in India is how little time people in urban cities spend outdoors. Because of how crowded, dirty, and polluted it is, they stay indoors for the most part.
I can imagine this plays a meaningful role in the declining metabolic health of Indians.
Data's equivalent of topofthelyne newsletter is definitely Atlan's metadata weekly. Great insights, and I relate to it as those buzzword have now become a need for us.
Newsletter -
When I first read these returns, I really thought beating nifty etc would be easy. Thankfully realised the mistake early. Renaissance is GOATed.
PS: He also mentions why the world doesn’t need more traders, zero sum game.