William Zeng (@wwzeng1) 's Twitter Profile
William Zeng

@wwzeng1

founder @sweep__ai YC S23 (AI coding assistant for JetBrains)

docs.sweep.dev

ID: 1515423126435221504

linkhttps://sweep.dev calendar_today16-04-2022 20:13:26

183 Tweet

430 Takipçi

536 Takip Edilen

William Zeng (@wwzeng1) 's Twitter Profile Photo

Just released a new tab model! It's competitive with the best Autocomplete models in VSCode, and we're the only plugin that supports this in JetBrains IDEs. 🧹 Our Sweep agent is also getting quite good. More to come soon!

William Zeng (@wwzeng1) 's Twitter Profile Photo

Thanks Josiah Parappally for the feedback on SweepAI! We built Sweep to work in PyCharm, Android Studio, and other JetBrains IDEs. We're also the only JetBrains Plugin that supports next-edit autocomplete. Some of the best developers I know exclusively use JetBrains, and now

Thanks <a href="/jparappally1/">Josiah Parappally</a> for the feedback on <a href="/sweep__ai/">SweepAI</a>!

We built Sweep to work in PyCharm, Android Studio, and other JetBrains IDEs.

We're also the only JetBrains Plugin that supports next-edit autocomplete.

Some of the best developers I know exclusively use JetBrains, and now
William Zeng (@wwzeng1) 's Twitter Profile Photo

Sweep 1.20.3 released! excited to release a much better next-edit autocomplete model and a much better onboarding experience for our jetbrains plugin

William Zeng (@wwzeng1) 's Twitter Profile Photo

Sweep is the best autocomplete for JetBrains IDEs like PyCharm, IntelliJ, and Android Studio. Our end-to-end latency is faster than human typing speed (sub 100ms P50), allowing us to serve suggestions per-keystroke. Here’s a demo of a common refactor you might do. This is not

Kevin Lu (@kevinlu1248) 's Twitter Profile Photo

You’ve never seen a typo in ChatGPT. So why is this a problem in autocomplete? This is because LLMs have been trained on complete sequences of text such as like books or production codebases. This is a problem for code completion as the code is often in a partial state.

You’ve never seen a typo in ChatGPT. So why is this a problem in autocomplete?

This is because LLMs have been trained on complete sequences of text such as like books or production codebases.

This is a problem for code completion as the code is often in a partial state.
joowon (@n0w00j) 's Twitter Profile Photo

i often see ex-jetbrains IDE users migrate to Cursor for bloated / large / monolithic codebases even though it’s not very performant with them. been using Sweep since I was at Ramp and now as a founder who’s more used to a jetbrains ecosystem, and it’s genuinely a lifesaver.

Samo (@samk0_com) 's Twitter Profile Photo

William Zeng sweep is truly great. It allowed me to fully switch back to phpstorm. I would say that it is even better than cursor one bc is not so hyperactive