hau @bryanhpchiang Twitter profile

Last Seen Profiles

@stwmaniax

@formidable_eo

@17mydarling

@stw_yuli

@swtlikecandyx

@enfeskanal1

@LaronDelm

@rjgeller

@mnkraft

@hamilton_heavyn

@mezzowl

@MertonSP

@Brejna90

@Samrat_ashok98

@ilovekeyyy

@testadimaglio84

@gyorai_sarapao

@TPackan

@gilmorecardigan

@staciekunihisa

@Jay_Dook

@Pomelo_o17

@chimu_tan_15

@TRAVmoneyofmine

@LoveTheShirt

@bokeplokalmalam

@CitygreenTurf

@crystal_stay

@Bakan_Yuliem

@stw_pdg

@CreatorInst

@FloodForum

@klp6818

@the_upgrader66

@tiovikram

@brentbronze

hau

@bryanhpchiang

1 year

say goodbye to awkward dates and job interviews ☹️ we made rizzGPT -- real-time Charisma as a Service (CaaS) it listens to your conversation and tells you exactly what to say next 😱 built using GPT-4, Whisper and the Monocle AR glasses with @C51Alix @varunshenoy_

290

1K

6K

hau

@bryanhpchiang

2 years

taking a history class as a CS major is a pain with those loong readings ☹️ so i built , your AI knowledge assistant upload any doc and get a buddy that can instantly: - summarize - answer questions quickly - clarify highlighted parts + anything else!

70

111

1K

hau

@bryanhpchiang

1 year

meet lifeOS: an operating system for your entire life 🌐 a personal AI agent delivered directly through AR smart glasses 👓 it uses computer vision to 👁️recognize👁️ your friends’ face then brings up relevant details to talk about based on your texts with them (memory🤯)

84

121

813

hau

@bryanhpchiang

1 year

surprised by how good bard is. - coding quality on par with GPT-4 in my own usage - generation speed is > 2x faster vs ChatGPT i thought the lack of streaming completions was a bug but now i think it's because it's so fast that you wouldn't be able to keep up with it.

41

45

756

hau

@bryanhpchiang

1 year

excited to finally share how we built rizzGPT (and the code!) we imagine a new era of ambient computing enabled by AR + AI, where everyone has their own personal assistant available 24/7. it’s like having God observe your life and tell you exactly what to do next. the deets 👇

hau

@bryanhpchiang

1 year

say goodbye to awkward dates and job interviews ☹️ we made rizzGPT -- real-time Charisma as a Service (CaaS) it listens to your conversation and tells you exactly what to say next 😱 built using GPT-4, Whisper and the Monocle AR glasses with @C51Alix @varunshenoy_

290

1K

6K

32

87

569

hau

@bryanhpchiang

1 year

AI agents in AR glasses = 🤯🤯🤯 my GPT-4 JARVIS can now: - recognize my friends faces - understand what i'm looking at using computer vision - respond aloud via TTS here it analyzes the Buck's menu and tells @DrakosBrown what to get based on his taste prefs + nutrition needs

28

86

548

hau

@bryanhpchiang

2 years

made a fun hack -- it's called ambient it uses AI to instantly convert what's on your screen into a calendar invite just take a screenshot! 📸 here's an example: imessage convo -> cal event

32

44

530

hau

@bryanhpchiang

2 years

midjourney quality is insane but writing prompts can be a struggle :~( wrote an extension to help you get more inspo convert any image you see on Pinterest into the original text prompt to discover new styles / artists 🪄 made w/ Boosts from @browsercompany , code below!

11

53

466

hau

@bryanhpchiang

1 year

@C51Alix @varunshenoy_ forgot to tag @Adriano34554795 , the memelord himself thanks to @brilliantlabsAR for providing the Monocle AR gadget -- super easy to clip onto any pair of glasses + has a camera, microphone, and high-res display this project is perhaps best put by @varunshenoy_ 😎 + 🧐+ 🤖 +

15

49

414

hau

@bryanhpchiang

2 years

this is xtrakt ⛏ instantly get structured data out of any website 📑 just specify the attributes! - open source, github link below - fully parallelized with @modal_labs , so it works on any web page no matter how long BYOK (bring your own key) demo:

21

36

312

hau

@bryanhpchiang

1 year

how much has this founder raised?

34

12

294

hau

@bryanhpchiang

1 year

welcome to the @tensortower 🏰 we've put the most cracked builders from stanford into an ai hacker house for a summer in SF. fully funded & ready for all you ai chads to roll thru...follow and DM @tensortower get the gradients flowing ‼️

14

7

102

hau

@bryanhpchiang

2 years

a fun hack! introducing dashGPT 🏃‍♂️💨 now you can get DoorDash just by talking to ChatGPT 🍔🤯 built with the amazing @twofifteenam @mollycantillon

7

13

119

hau

@bryanhpchiang

1 year

found this on the entrance to my dorm, too based palladium vibes

6

5

105

hau

@bryanhpchiang

1 year

update: here's how we did it + the code

hau

@bryanhpchiang

1 year

excited to finally share how we built rizzGPT (and the code!) we imagine a new era of ambient computing enabled by AR + AI, where everyone has their own personal assistant available 24/7. it’s like having God observe your life and tell you exactly what to do next. the deets 👇

32

87

569

3

26

93

hau

@bryanhpchiang

1 year

@annimaniac @C51Alix @varunshenoy_ don't let ur dreams be dreams.

2

1

74

hau

@bryanhpchiang

1 year

fish tacos were certified bussin: “Haven’t had fish tacos with salmon before. Hit the spot… fresh and filling.” demo built with @OpenAI GPT-4, @Apple Speech transcription, @elevenlabsio TTS for the voice, @PaddlePaddle OCR, and the @brilliantlabsAR Monocle

4

7

64

hau

@bryanhpchiang

2 years

be the first to try it out: this works with any kind of document / text (textbooks, research papers, financial, case studies, legal, reports, etc.) sign in first to use it -- lmk what you think!

12

0

64

hau

@bryanhpchiang

1 year

@ratankaliani @C51Alix @varunshenoy_ only made possible with sbux wifi

0

60

hau

@bryanhpchiang

1 year

@OpenAI rizzGPT is just a simple proof-of-concept of what’s possible. lots more to build here, especially once multimodal GPT4 arrives please DM if there’s anything you want to see happen !! and as promised, here’s the code:

GitHub - acui51/monocle-rizz: rizzGPT

rizzGPT. Contribute to acui51/monocle-rizz development by creating an account on GitHub.

github.com

2

10

54

hau

@bryanhpchiang

1 year

built over the weekend with - @OpenAI GPT-4 - @Apple Speech framework for on-device transcription - the Monocle AR device from @brilliantlabs - custom facial recognition models

hau

@bryanhpchiang

1 year

@C51Alix @varunshenoy_ forgot to tag @Adriano34554795 , the memelord himself thanks to @brilliantlabsAR for providing the Monocle AR gadget -- super easy to clip onto any pair of glasses + has a camera, microphone, and high-res display this project is perhaps best put by @varunshenoy_ 😎 + 🧐+ 🤖 +

15

49

414

2

10

54

hau

@bryanhpchiang

1 year

the interplay of AI and AR will redefine personal computing and help us unlock our full potential. why do you need a physical screen when you have an infinite digital one? why do you need a keyboard/mouse if you can just talk to apps?

2

4

49

hau

@bryanhpchiang

1 year

@chrissyykat @C51Alix @varunshenoy_ ok. when are u free?

0

49

hau

@bryanhpchiang

2 years

behind the scenes: OCR to extract text + GPT-3 to go from unstructured text to the event details here's another example: flyer image -> cal event

3

1

46

hau

@bryanhpchiang

1 year

on generative UI: all these "user describes an app" -> get a custom app interface are cute but missing the point. intelligent apps need to proactively anticipate user intent from existing context and generate the desired interfaces (without explicit instructions).

1

4

46

hau

@bryanhpchiang

1 year

the future of curation: your ai agents that understand your interests scour the web on your behalf, saving the interesting gems (podcasts, articles, tweets, etc.) ur agents exchange recs with ur friends' agents. at the end of each day, you get a digest for what to consume next

Varun Shenoy

@varunshenoy_

1 year

You can read the full essay here.

1

3

22

1

4

45

hau

@bryanhpchiang

1 year

talking to ai should feel like talking to your friend. super excited for wabbit 🐇 also if you're building an API for text to animated talking avatar, please DM me !!!!

2

5

41

hau

@bryanhpchiang

1 year

@mollycantillon

1

38

hau

@bryanhpchiang

1 year

@bilawalsidhu @C51Alix @varunshenoy_ machines just talking to machines all day long

3

1

37

hau

@bryanhpchiang

2 years

DM me if you wanna try it out :~)

7

0

37

hau

@bryanhpchiang

2 years

try it out for yourself:

content.js

GitHub Gist: instantly share code, notes, and snippets.

gist.github.com

1

2

29

hau

@bryanhpchiang

1 year

found FLAN-T5 irl. SF is back @_jasonwei @YiTayML

2

1

29

hau

@bryanhpchiang

2 years

in the future: every app will have a conversational interface, with GUIs rendered on the fly depending on the context. () companies like @Instacart , @Shopfiy already moving in this direction ()

Replit — CLUI: Building a Graphical Command Line

“Command line interfaces. Once that was all we had. Then they disappeared, replaced by what we thought was a great advance: GUIs. GUIs were - and still are - valuable, but they fail to scale to the...

blog.replit.com

4

2

28

hau

@bryanhpchiang

1 year

@OpenAI @Apple @brilliantlabs shoutout to crypto legend @bridge__harris for helping with the incredibly scuffed demo 🙏 stay tuned for more upcoming prototypes 🌞 DM if you want to try this out, reply if you have any other use cases in mind!

3

1

24

hau

@bryanhpchiang

1 year

@DrJimFan ??? this is just GPT4 txtonly hooked up to a bunch of tools. no joint txt-img understanding, which is the actually useful part of GPT4.

0

23

hau

@bryanhpchiang

1 year

@amanrsanger i built this exact thing. the bottleneck is by far speech generation, open-source models are way too slow audio is too high resolution compared to text / no single model can go directly from text to speech right now well solvable, just not in a weekend. DM if youre interested

3

0

23

hau

@bryanhpchiang

1 year

@OpenAI @Apple @brilliantlabs for recognizing faces, i made a database ahead of time: - took a photo for each friend - face detection model to crop out the face - created feature embeddings for each face when lifeOS is running, it detects faces and computes embeddings on the fly, finding the nearest

3

1

21

hau

@bryanhpchiang

2 years

@twofifteenam @mollycantillon behind the scenes: browser automation to reverse engineer a @DoorDash API + hooked up to ChatGPT all of this happens in the background! the user never sees it.

2

1

22

hau

@bryanhpchiang

1 year

@OpenAI we found that transcription speed depends on wifi speed since its being done in the cloud, so we had to leave the hackathon and go to a local starbucks to get faster wifi for filming the demo 😭 in the future, we’d do transcription locally on host (whisper.cpp) to avoid this

1

3

21

hau

@bryanhpchiang

1 year

appreciate all the replies. my current mental model: you & AI iterate on some shared, dynamic canvas (context) you can gesture to parts of the canvas, but you can't directly change it. instead you dictate your feedback to the AI. the AI generates, you discriminate.

1

4

22

hau

@bryanhpchiang

1 year

generative AI makes this future possible. first, multimodal perception capabilities (audio, text, images) help AI understand what’s going in your life. this context is key for the AI to provide hyperpersonalized support. in rizzGPT, context = ongoing conversation.

1

2

20

hau

@bryanhpchiang

1 year

@SilasAlberti @bfspector @punwaiw @lktong_ @leithnyang @julianhquevedo @varunshenoy_ @andrewparkk @usygoosy @siddrrsh (IN SPIRIT) + LIZ, WILL + MORE SPECIAL GUESTS

2

3

18

hau

@bryanhpchiang

1 year

second, LLMs let us directly talk to our devices, apps, and the Internet (Language User Interfaces). instead of tapping on GUI buttons and controls yourself, you’ll simply tell your assistant what you want done. this makes hands-free AR actually viable.

1

3

19

hau

@bryanhpchiang

1 year

@sjwhitmore @jasonyuandesign when the co website is straight up tome >>>

1

0

18

hau

@bryanhpchiang

1 year

for rizzGPT: there’s a webapp on the host device (phone) that communicates with the Monocle via bluetooth. raw conversation audio (from host mic) is converted to text in realtime ( @OpenAI Whisper in cloud). GPT uses the raw transcript text to generate what the user should say.

3

2

16

hau

@bryanhpchiang

1 year

what if you had the AI roleplay an inquisitive student and have it grade the human """teacher""" afterwards ?

Chris Piech

@chrispiech

1 year

I have been working in scaling high quality education for over 10 years, especially using AI. However, there is another idea for scaling high quality education that I am more excited about. 🚨There are so many people who want to teach! Because tutors learn. 🧵 1/n

3

16

174

2

0

17

hau

@bryanhpchiang

1 year

@JonathanZWhite mit media lab walked so that dumb cs students with APIs could run

0

16

hau

@bryanhpchiang

2 years

@modal_labs github link: GPT can do unstructured -> structured so well but instead of asking for JSON, you can also ask for CSV to save tokens!

2

0

15

hau

@bryanhpchiang

1 year

@OpenAI the host webapp sends micropython directly to the monocle to display the optimal rizz response. all of this happens while the user still looks engaged + attentive in the conversation! there’s zero context switching.

1

2

13

hau

@bryanhpchiang

1 year

@BingBongBrent @C51Alix @varunshenoy_ great questions. will do a more detailed write up soon!

0

1

14

hau

@bryanhpchiang

1 year

DM if you have any other fun use cases in mind! and mega thanks to my roommate isandro (product design chad) for designing the cardboard monocle phone holder

1

0

13

hau

@bryanhpchiang

1 year

@BenjaminDEKR you may not like it but you will have it its time to accelerate

1

0

10

hau

@bryanhpchiang

1 year

@PascalPixel this is so cool. so how does one get started modding a browser? bls don't reply with "fork chromium"

2

0

11

hau

@bryanhpchiang

2 years

building a new keeb 🎹 live at

1

0

10

hau

@bryanhpchiang

1 year

@alexgraveley this misses the point he doesn't want anybody to access (steal) public twitter data

2

0

11

hau

@bryanhpchiang

1 year

@Jenstine u rlly just

1

2

11

hau

@bryanhpchiang

1 year

@theappletucker @C51Alix @varunshenoy_ we were inspired by ur rizzness

0

1

9

hau

@bryanhpchiang

1 year

@KevinAFischer @C51Alix @varunshenoy_ if only gpt4 had more personality

2

0

10

hau

@bryanhpchiang

2 years

@shwin_m shhhhhh

0

9

hau

@bryanhpchiang

1 year

@geoffreylitt the entire point of a conversational interface is so that you don't have to manually tweak the UI !!!!!! the spreadsheet should be to the side (persistent) with a chat panel to keep iterating. copilot 365 is doing this right

1

0

8

hau

@bryanhpchiang

1 year

we have to redesign the underlying physical and digital infrastructure to unlock the full potential of agents. an interface for every service / process crypto is this for value

david

@davidtsong

1 year

A short sci-fi future about AI agents in 2025 🔮 Substack link in bio 🫡

0

8

0

1

8

hau

@bryanhpchiang

2 years

@sophfuji @modal_labs i just ask ChatGPT to do it for me

1

0

8

hau

@bryanhpchiang

1 year

@KaseyKlimes Maybe we should just make Howl's Moving Castle a real thing -- portable cities.

0

6

hau

@bryanhpchiang

1 year

@_mattneary the internet isn't personalized though. there's still a lot of value in curating things that appeal to you and _you_ only. although perhaps that can be automated if we can build digital versions of ourselves that understand our preferences.

2

0

7

hau

@bryanhpchiang

1 year

@nikkiccccc @OpenAI @LKGGlass @DBtodomundo so sick! how did you animate the character?

1

7

hau

@bryanhpchiang

1 year

@sjwhitmore @C51Alix @varunshenoy_ thanks sam. maybe we can put ur agent on the Monocle one day 😎

0

6

hau

@bryanhpchiang

1 year

@k3ntosan @C51Alix @varunshenoy_ @Adriano34554795 @brilliantlabsAR here!

hau

@bryanhpchiang

1 year

excited to finally share how we built rizzGPT (and the code!) we imagine a new era of ambient computing enabled by AR + AI, where everyone has their own personal assistant available 24/7. it’s like having God observe your life and tell you exactly what to do next. the deets 👇

32

87

569

0

2

6

hau

@bryanhpchiang

2 years

@jan_ruettinger try it! you can just ask if there are any unfavorable clauses :)

0

6

hau

@bryanhpchiang

2 years

@imjaredz hahhahah math symbols r a bit tuff to grok

1

0

6

hau

@bryanhpchiang

1 year

@rahulgs any way to support openai?

3

0

6

hau

@bryanhpchiang

1 year

@nategrebelsky if we had LLMs when google glass was launched, itd prob do hella well ngl

1

0

4

hau

@bryanhpchiang

1 year

@W4ngatang @alexandr_wang @alexjkwang

1

0

5

hau

@bryanhpchiang

1 year

@nonmayorpete feels like a bad-faith figure to me v clear that the authors are just trying to shill their own (not so relevant) work lol

1

0

5

hau

@bryanhpchiang

9 months

@AviSchiffmann wrong. hardware is a necessary step, but it will not be the first. the bulk of our lives is digital; there's more than enough context for our AI friends to really get to know us. simply adding more vectors for capturing context won't get you very far

1

0

5

hau

@bryanhpchiang

1 year

@_Cybershota you will love it trust

2

0

5

hau

@bryanhpchiang

1 year

@concept_central take me back to the good ol days

0

3

hau

@bryanhpchiang

2 years

@varunshenoy_ gots to see it thru

0

3

hau

@bryanhpchiang

1 year

for visual applications (creative 2D/3D work), maybe permit some low fidelity (ie sketches) input from user to canvas. voice is powerful because it does not require user to context switch (can talk & use hands for two separate things at the same time).

1

0

3

hau

@bryanhpchiang

1 year

@usygoosy ask @punwaiw

0

4

hau

@bryanhpchiang

1 year

@thatdaniperson yo chill

0

4

hau

@bryanhpchiang

1 year

@Aizkmusic example?

1

0

4