Lewis latebraking into the corners compared to George gave him such an advantage even if he didn't *exactly* have the same pace as George. Just phenomenal drive. This side by side is so insane to watch
no really bcoz no one understands. teammates whine over p9-10s and refuse to help eo but carlos really kept lando (his opponent) in his drs while LEADING THE Fucking RACE. no one else could ever dare to do that. that's why singapore 23 is for the history.
@madhavchanchani India should just be post training deep seek etc for a while. Better to focus on energy, manufacturing, EV stack, older node chip stack for EVs etc, and make some money first. Frontier model building is a rich country game.
Meet Vedant Shrivastava, Nisarga Adhikary, and Sarthak Sidhant ,they exposed CBSE in every possible way :
17 years old Vedant Shrivastava :
> Applied for the CBSE re-evaluation process
> Got a different Physics answer sheet
> Posted it on X
> Got labelled "Pakistani" by the BJP IT cell
> Brought CBSE to its knees and proved them wrong
19 years old Nisarga Adhikary :
> Bro hacked the CBSE website
> Reported the vulnerability to them
> But they didn't take any action
> Then bro posted it on X
> Showed everyone that it could be hacked
18 years old Sarthak Sidhant :
> Bro exposed a CBSE tender
> Took it to X
> Wrote a detailed thread
> Explained how the CBSE OSM tender conditions allegedly favoured COEMPT
> And today came on the media
> Exposed them with facts
These three boys are doing what whole media failed to do, all of them took x to expose CBSE .
Pi’s so good. Claude code has become
Insufferable rn the rate limits are crazy for the pro plan. Switched to openai pro plan and started using Pi.
Minimal, opinionated, everything feels so quick and crisp. Moreover, it’s a beautiful piece of malleable software that you can extend to your own needs.
@pragdua@ankitkr0 Awesome. I’ve been thinking about building something like this. Do you have support for branches, as in if I branch from conversation to another , Is there a way I can visualise it here ?
You should watch this.
It just shows how disconnected we are from the small group of people making decisions that will impact our future heavily.
These people have so much ai psychosis. If you listen to how she speaks, everything is personified, it is undoubtable she believes this is a living computational organism.
Just like how a model can hype up an individual into psychosis through reinforcement, a small group of people are giving themselves psychosis through reinforcement.
Wild times we live in
I'm going through MIT opencourseware on distributed systems and here's a thought:
All novel tech has always been supported and grown through Academia. The whole cloud and distributed computing boom and now all this AI stuff. Academia has always been part of it all.
I strongly feel you shouldn't decide where to live purely based on costs. I find cities to be beyond concrete, they're so rhythmic & we absorb the culture around us. It's so important to live somewhere that injects hope, ambition & love into your veins.
Having your computer run overnight on a task and then checking on it in the morning is one of the best feelings in tech.
When I was a kid it was doing a multi part download or a slow running task.
Now it is to run a harness / agent loop / something GPU bound overnight.
If you're a backend developer who wants to learn about databases and distributed systems, the reading list and lecture videos for the MIT distributed systems course are terrific places to start.
You can check all the material here:
https://t.co/Wk2cKGtKmV
Managed to build a harness for myself and built an ios app to test it out. Only thing I did was PR reviews -- everything else automated.
Harness : https://t.co/BLedeafW1A -- built it as a CLI, think create-react-app but gives you an opinonated harness which your agent can go yolo on.
ios-app : https://t.co/vKK7o8nAGQ -- Piper helps you bridge auth-walled X articles into any read-it-later app of your choice.
Mistakes happen. As a team, the important thing is to recognize it’s never an individuals’s fault — it’s the process, the culture, or the infra.
In this case, there was a manual deploy step that should have been better automated. Our team has made a few improvements to the automation for next time, a couple more on the way.
Managed to get like ~2 hrs of voice notes from my whatsapp. I setup autoresearch repo of my own and ran the loops. I was using Sonnet-4.6 and man did it burn through a lot of my credits. I ran it for a total of like 4hrs max. This is not enough and yet it managed to halve the WER of the vanilla whisper-small.
Plan is to use https://t.co/GajgYYAQmd as harness and use Kimi-2.5 . This setup should be wayy cheaper and should help me run the loop for long.
Find the current state here : https://t.co/RvOWc809Fz
Check the results.tsv to see the experiments the agent conducted.
Managed to get like ~2 hrs of voice notes from my whatsapp. I setup autoresearch repo of my own and ran the loops. I was using Sonnet-4.6 and man did it burn through a lot of my credits. I ran it for a total of like 4hrs max. This is not enough and yet it managed to halve the WER of the vanilla whisper-small.
Plan is to use https://t.co/GajgYYAQmd as harness and use Kimi-2.5 . This setup should be wayy cheaper and should help me run the loop for long.
Find the current state here : https://t.co/RvOWc809Fz
Check the results.tsv to see the experiments the agent conducted.
Without any prior understanding of ML research, I'm trying to setup Karpathy's autoresearch to finetune an STT model custom-tailored for me. The idea / my assumption is that companies like Whisprflow do personalization at the tooling layer (hotword lists, post-processing) etc. Now what if we can get this personalization at the model layer.
Wish me luck lol
I can see some improvements, now I’m onto getting some 5-10 hours of clean and transcribed audio data of my own to run Autoresearch lol. And since I switch between my native language ( മലയാളം ) and English when I talk, it’s interesting to see if it can pick that up. And for getting the training data, I’m employing pseudo -labelling with a Teacher model from Sarvam