sf
Joined September 2021
We tested our dubbing AI on a 3 mins clip of @AliAbdaal. Here is how it performed 👇 🌟 Imagine the performance when the model is provided with 100 minutes of his content. 💫 Now imagine 1000. 🗣️The stories we tell don't need to be confined to a geographical boundary anymore. India would hear Ali in Hindi, Pakistan in Urdu, South Africa in Afrikaans, and more. 🚀Ideally, there is no stopping your content anymore. To the World 🌍 CC: @apmparker @jamiewhiffen @GeorgeBlackman_ @kliszczakjakub #YouTube #contentcreation #creators #podcast
14
7
4
59
25,100
every month i give a fresh chance to siri to help me most basic tasks of scheduling etc it disappoints every time
2
4
160
He is probably the only person I have seen at this scale who pursues his hobbies so passionately.
1
178
its so cool to see that people still use bing and duckduckgo
1
746
adarsh retweeted
Can someone explain why I need to bring my phone back every 2 years and get a new one at $0 with the same plan 💀 What is this business model even
44
2
2
326
83,070
this was so long coming finally some semantic search on people data
2
75
❤️❤️❤️
Figma in India: our Bengaluru office is officially open! Design and Make things with us (We’re hiring)
249
Bose <--< Amar Bose how did i not know this
80
is it just me or claude code is getting harder and harder to use
2
100
ahhhhhhhhhhhhhhhhhhhhh
1
85
every time i am stuck at terminal command, i use @warpdotdev but then i look how complicated that is and i switch back
81
today is a wild day!
🚨 Airbnb just turned customer support into an AI self-learning machine 🤯 Their new research, Agent-in-the-Loop (AITL), is wild it transforms live customer support into a continuous training pipeline. Instead of waiting months for offline fine-tuning, AITL loops human feedback directly into daily operations. Every few weeks, the system gets smarter on its own. Here’s the trick 👇 Each human agent gives 4 live signals when reviewing AI responses: • Which output they preferred • Why they chose it • Whether the info was relevant • What knowledge was missing Those signals constantly retrain the retrieval, ranking, and generation models turning customer support into a feedback flywheel. The payoff is real: +11.7% retrieval recall +14.8% precision +8.4% helpfulness +4.5% agent adoption No more static models. No more 3-month retraining cycles. AITL creates a system that learns while it works humans → agents → infinite improvement. Full paper: arxiv. org/abs/2510.06674
2
219
learn how zeropath works. i can promise it will be worth your time
1
211
today seems to be the day of launches - Claude Haiku 4.5 - Amp Code - new Mac
1
2
203
i wanna consume data like an llm at this point
94
switched to @claudeai for daily tasks but its been such a bad experience with tool calls and context so far going back to gpt
71
this is so so cool. been there, faced that problem kinda solution
I always wanted a pen that tracked everything I ever wrote down So we built one
1
143
no shit sherlock you had one task
76
who all applying? looking for a team
Announcing the @MetorialAi Hackathon hosted by @ycombinator. > Winner gets guaranteed YC interview! > Nov 1st-2nd, 24 hours overnight at the YC Office > YC Partner Andrew Miklas will judge > There might or might not be a rave after 👀 Limited to 200 spots. Apply now: ychackathon.com
121
how was sleeping on @posthog mcp for so long The Best Thing
2
1
124
yk sometimes you hear people speak and at the end you think about what you learnt and its null like absolutely nothing that's how that conversation was wasted 10 mins or whatever of my life
Robby Stein from Google Search team live now!
1
172