Sky Gupta أُعيد تغريده
Sky Gupta
1.2K posts
Sky Gupta
@skyGuptaCS
Experimenting with AI | Decoding the algorithms | div centerer
Earth انضم Mayıs 2025
147 يتبع208 المتابعون
Sky Gupta أُعيد تغريده
We just announced Lightning V3: a TTS model designed for Conversational Voice Agents.
It is available in two versions + state-of-the-art in overall naturalness and other voice metrics.
In this article we discuss:
+ Why conversation is the hardest test for a voice model
+ How Lightning v3 sets a new bar on open TTS benchmarks
+ Why naturalness and expressiveness are not the same thing
+ Persona specific evals are the next frontier..and more.
Try now!
Sudarshan Kamath@kamath_sutra
English
Sky Gupta أُعيد تغريده
i wonder if you could run your vitest test suite in parallel using dynamic workers - 1 test per worker and massively parallelize testing
English
Sky Gupta أُعيد تغريده
Sky Gupta أُعيد تغريده
Building projects using AI is getting easier and faster.
Would be cool to see agents doing reverse engineering at the same pace. Decompiling binaries, tracing API calls, finding exploits, etc.
Sundar Pichai@sundarpichai
A few updates from across Google this week🧵 A new full-stack vibe coding experience is now in @GoogleAIStudio with the @Antigravity coding agent and a built-in @Firebase integration. Turn your prompts into amazing production-ready apps.
English
Told my friend at Google to create an account on our platform, now we are “loved” by users at Google.
English
Sky Gupta أُعيد تغريده
Re-introducing...Bond, your AI Chief of Staff.
You worked 10 hours yesterday. Maybe 2 of them actually mattered.
The rest? Standups, Slack threads, status updates, triaging emails.
Executives don't need to work harder. They need Bond
👉 bondapp.io
English
Sky Gupta أُعيد تغريده
Google is the default search engine on Comet iOS (unlike on Comet desktop): Most mobile browser searches are around navigating to restaurant or local shops, checking scores, shopping, hotels. Google does a much better job here than anyone else in the world, including Perplexity.
English
Sky Gupta أُعيد تغريده
Physical residencies: Rarest skill in this industry
- work with best founders in the industry
- jam with top tier VCs
- travel to 5 countries a year
- get paid to do this
tag anyone you know who can do this. tag yourself. no agents only humans. cyborgs okay
Pareen@pareen
do you want to learn how to manage physical founder residencies? i have an opportunity for you. it will be hard, you will have to learn a lot quickly, travel to different places, live with founders, and be on top of your game. but it will be worth it coz only a handful of people in this industry can do this currently. reply if you are interested. QT/RT/tag 🙏
English
Sky Gupta أُعيد تغريده
do you want to learn how to manage physical founder residencies?
i have an opportunity for you.
it will be hard, you will have to learn a lot quickly, travel to different places, live with founders, and be on top of your game. but it will be worth it coz only a handful of people in this industry can do this currently.
reply if you are interested. QT/RT/tag 🙏
English
Sky Gupta أُعيد تغريده
Sky Gupta أُعيد تغريده
The same agent you know and love!
Now smarter and cheaper :)
Try at orchids [dot] app
Orchids@orchidsapp
Introducing Orchids Agent 1: 4x smarter, 3x cheaper and GPT 5.4 Agent 1 is already in place for all users - nothing you need to do on your end. GPT 5.4 is also available to all users. Try at orchids [dot] app.
English
Sky Gupta أُعيد تغريده
New benchmark suite for AI coding agents: SWE-Atlas!
the effort is to measure something different from classic SWE-bench patch-fixing basically deep codebase understanding (runtime analysis + multi-file reasoning).
in Codebase QnA, it's pretty hard with top models were only around ~30% strict pass rate.
English
Sky Gupta أُعيد تغريده
Early days but a big milestone for us! This model is still in preview and we expect to tune behavior a lot over the coming weeks - but wanted to get folks a snapshot as soon as we could.
Cognition@cognition
We are sharing an early preview of our ongoing SWE-1.6 training run. It significantly improves upon SWE-1.5 while being post-trained on the same pre-trained model - and it runs equally as fast at 950 tok/s. On SWE-Bench Pro it exceeds top open-source models. The preview model still exhibits some undesirable behaviors like overthinking and excessive self-verification, which we aim to improve. We are rolling out early access to a small subset of users in Windsurf.
English