Abhishek (key/value)

everyone asks which STT is "fastest" but, fastest at what? bcz "real-time" in speech recognition means two completely different things and most teams are optimizing the wrong one. one word. two problems.

tensorlake team is always cooking best

We haven’t technically launched TLFS - Tensorlake’s distributed file system but apparently we are using it for Rust build caches in CI. 😇

@harshitajain561 how are you doing today harshita? can i have some credits?

:")

@joshmo_dev @herdrdev looks fun xD if you're open to experimenting, i'd love to get your thoughts on some of our models in your setup. happy to send over credits if you'd like to give them a try :)

i'm using cursor to create a voice agent harness that utilises @herdrdev to command a bunch of cursor instances literally playing factorio irl lmao

I think today is the day to be done with riddles and announce I joined @SarvamAI ? I'm here at #sarvamepoch, come say hi if you're around. #epoch

@rivuchakraborty @SarvamAI what a timing!!! huge congratulations 🥳 they are lucky to have you! (treat when?)

I'm based in SF so I gotta try every trick in the book to keep up with the AI boom. Voice AI is especially fascinating because it’d be pretty cool to automate my videos. I work in this field and I'll be talking more about this and about the math of AI. Stay tuned.

@earth_ish @pipecat_ai @smallest_AI no, but the slides are in the post

@StalwartCoder @pipecat_ai @smallest_AI awesome! do you have recordings?

gave a talk last weekend on how voice AI works. the stt→llm→tts pipeline, why you get ~800ms before a conversation feels broken, and how orchestrators like pipecat/livekit run the whole loop live demo ran on @pipecat_ai with @smallest_AI's Pulse + Lightning models (the stack i work on daily) best moment was interrupting my own agent mid-sentence on stage and it just... handled it thanks @PriteshKiri + the @ReactPlayIO crew for putting it together! slides + demo code: github.com/abhishekmishra…

We like demos that talk back!

@howdevyou hahaha very kind dude

Here's my first tweet. Manifesting 'viral' before the first like. @StalwartCoder

👉INTERNSHIP alert! Will be posting about a full-stack engineer internship opportunity today. ₹40-45k/month 3 months On-site in Bengaluru Has very high potential to convert to full time. Good startup (Ankush-approved), with actual prod work. Make sure that you have a hiring-profile set up here - tier3.college/hiring-profile (free). Folks with professional references from folks in the industry here will be preferred first for shortlisting. Happy to answer some questions in replies. Full details will come up later today.

@joeyousss @mehulved

Does anybody have experience of leaving their car at Bangalore airport’s dedicated parking for more than two days? Want honest feedback’s.

I’ve said this before and I’ll say it again: Using LLMs-written content professionally will spoil your career, especially if you’re in early stages. Everyone worth their salt that I know is hugely aversive to reading slop and despite your smart attempts, they can sense signatures of AI writing (even if they can’t articulate what gave it away). When you delegate writing, you delegate thinking and that delegation takes away from you the very thing you need for your future career advancement: careful thinking. I often hear.. “but my writing is bad”. Well, if you keep on delegating, how will it get better? It’s like complaining that you don’t go to gym because you’re unfit. Write the first damn draft yourself and then take feedback from AI. Do not ask it to rewrite, but rather ask it point errors and suggestions for improvement. Writing using LLMs is analogous to binge watching short videos, or eating junk food. Pleasurable dopamine hits in the short term, but complete dependence in the long term.

It's 2030. What did we do right? We changed how interfaces work. Most of the clicking and tapping is gone you just tell the machine what you want, and it listens. Machines interacting with humans the same way humans interact with each other. Not zero typing. Screens still matter. But most of it becomes voice. Full podcast, link in thread.

If you are building for the agentic world, come join us on 27th July in SF for a specially curated session with two founders who are at forefront of deploying agents in production RSVP 👇🏾 luma.com/9smoya03

Smallest AI SF Office - July 27 ! Hosting a handful of founders along with @DeVC_Global and @Razorpay founder Shashank Kumar. Link in description.

You don't need a trillion-parameter model that understands physics to answer a customer support call. But that's not even the main reason we build small models. The real one: LLMs store everything. Literally everything all of it sitting in the weights. Humans don't work that way, and your agent shouldn't either. Full podcast, link in thread.

@tanishasharmax hi Danisha

hello from danisha? 😭

149 new voices are now live on Lightning V3.1 Pro. Expanding your voice library across even more languages: • 67 voices added across 9 Indian languages • 21 voices across Asian & Middle Eastern languages • 61 voices across 11 European languages Available now in the playground. Go try them.