📞😖Tired of wasting time on phone calls?
"Please hold while we transfer your call"
"the estimated wait time is 90 mins"
"Press 1, 2, 3, 4… press 0, go back"
I HATE making any phone calls. Any kind!
So, I built an AI for it. Here's the teaser, enjoy!
🔉sound on
🧵thread 1/3
⚠️RAG might be dead, after reading 58 pages of Genimi 1.5 Pro tech report. Here's my thoughts as AI founder,
1. Simple RAG system like similarity search with vector db will be dead. But more customized RAG will still live. The goal of RAG is mostly on retrieval relevant
After Sequoia explain their LLM Stack, now it's
@a16z
's turn,
first pic: a16z
sec pic: Sequoia
- a16z's version is more programatic, systematic
- sequoia's version is more business-driven
- they share a lot common LLM tools, like
@llama_index
@LangChainAI
I personally prefer
After
@Sequoia
and
@a16z
explain their versions on LLM stack, now it’s
@GreylockVC
's turn. Loving this trend!
First pic: Greylock's version is more data centric
Sec pic: a16z's version is more systematic
Third pic: Sequoia's version is more business centric
🧵👇
Introducing , an idea-to-GenAI app platform, where AI agents do the hard work
Draw your ideas, connect to AI agents and ship to production on day 1
Powered by
@tldraw
,
@YouSearchEngine
API,
@MistralAI
🧵1/10 Today is day 1, let's ship some apps together
One blog post with one mega thread to understand autonomous agents! This thread is all you need!
LLM Powered Autonomous Agents by
@lilianweng
(leading Applied Research at OpenAI)
🧵👇6 takeaways, 3 components with technics, 4 case studies, 3 challenges
Now it's my turn to share thoughts on LLM development, several steps can save you 50% time.
After 1000+ hours on LLM development, here are 5 take aways 👇🧵
(Thanks
@a16z
for this amazing template, and
@FinanceYF5
for editing)
🚀Launching AI Silicon Valley = RealChar + AI town + Llama2
💡Put your RealChar characters in any virtual world and have interesting conversations within!
Powered by
-
@RealCharAI
backend
-
@a16z
@realaitown
-
@LangChainAI
-
@anyscalecompute
👇🧵 0/10 how to achieve it?
Microsoft just launched a LangChain alternative: Guidance
It connects to self hosted LLMs as well as OpenAI LLMs. Everything is in the prompt, less opinionated on encapsulation
(super long read)
⚠️RAG will live
🔥Heated discussions on my post, but many seem to overlook the why for RAG existance. It's not just because of inference speed or cost.
If you believe in RAG, read this post. Here's my own arguments why RAG will live.
First let's be open
⚠️RAG might be dead, after reading 58 pages of Genimi 1.5 Pro tech report. Here's my thoughts as AI founder,
1. Simple RAG system like similarity search with vector db will be dead. But more customized RAG will still live. The goal of RAG is mostly on retrieval relevant
@RealCharAI
We just won the 🥇 place of Autonomous Apps Hackathon
@agihouse_org
.
Here's all the cool parts,
1. (AI) Sam Altman greet everyone working on AGI
2. Sam uses MultiOn to send a tweet
3. Sam expresses emotions(read out) 😎
👇🧵See the live demo video, how we did it
Truth be told. I was never top 1%.
From traditional Asian family, growing up I had so much self doubts. It never occurred to me, I was the outlier. The only thing I am good at is I never give up.
Thanks so much for
@ycombinator
. I finally find where I belong.
Time to ship🚢🚢
Here’s all the crazy things happened in last 5 days
- We launched RealChar
- RealChar is Top 1 trending on Github
- RealChar got 3k stars in 5 days and it’s accelerating!
- RealChar official v0.0.1 release!
- RealChar ios app Beta testing
- Discord grow from 0 → 370
👇🧵
How to work with LLMs CORRECTLY? How to graduate from a prototype project?
(Long read)
Providing context to my last tweet, I've noticed a pattern in the creation of AI applications, which begins with the thought, "I have this data and it would be useful if a chatbot could
I was at W&B FullyConnected conference full day last Wednesday.
Here is one of best session by
@sandersted
(MTS from
@OpenAI
):
What Is GPT-4 Good For?
🧵👇Here's all the slides and takeaways
1/10
As always, here's my 10 amazing learnings from
@ajassy
. Thanks
@garrytan
and
@ycombinator
for inviting the amazing guest!
1. Early days in AWS? They bet Internet is a OS. They want to build fundamentals for developer need.
2. AWS was a hard sell. For anyone who is doing
Now
@RealCharAI
is on
@huggingface
!
You can have conversations with your favorite AI characters while working on your favorite models!
With HF's space, anyone can one button to create and deploy your RealChar characters!🤯🤯
Give it try ->
🚀 v0.0.3 is LIVE!
- 👓Add Vision Pro support
- 🖥️Add character in UI (No Code required)
- 🦙Add Llama2 support
- 👥Add 8 new community characters
- 📩Share a character
- 💬Share a conversation
- 🎭Animated avatar
- 🗣️Add 2 new Text to Speech
👇🧵details
💡Last demo I wasn't able to show you full capability of Now, 1000%! GPT4 + interrupt + voice clone!
Me->Raiden Shogun And Ei
- 🤯Do you know you are an AI?
- 🤯No, no, I created you?
- 🤯What's your last memory?
- 🤯Do you think your memories are real?
LLM stack for my incoming open source project:
@llama_index
-> Data processing
@trychroma
-> Vector DB for information retrieval
@LangChainAI
-> Conversation Orchestration
@elevenlabsio
-> Audio interactions
@OpenAI
-> GPT3.5 turbo
A FUN project to help you understand
Last night YC retreat learning
- the place I finally fit in
- good things and luck. trust you are lucky
- don’t give up for longer and harder
- learn & grow
- type 2 fun. believe u have super power
- take yourself seriously
- get low status for really long if you’re old😂
YC S24 first week batch in. Why
@ycombinator
is unique? Learnings and highlights in 🧵
1. Retreat. Best founder community with 1% builders
2. OpenAI Greg + Stripe Patrick talk
3. AWS Andy talk
4. Office hours from AWS, OpenAI
5. Revia first demo
6. Airbnb Brian Chesky talk
lol. I didn’t know there’s tracker. This is cool.
One thing I want to share about
@garrytan
He is the best leader I have ever seen and care so much about Every Single founder in YC.
How do you deploy your own
@RealCharAI
on
@huggingface
for free within 1 min?
Only 3 steps:
- Duplicate space
- Set OpenAI key & 11labs key
- Wait & Done!
👇🧵Step by step guidance
Don’t build your product on solving AI model issues. I know it’s a straightforward product to build.
Build your product assuming AI models will be 10x better. This is the actual value!!
Multiple speakers at Startup School made the same very interesting point. If you start a startup now to do something that barely works with AI, the next models will make it really work.
We () have been using
🦜🔗LangSmith to trace all our LLM interactions, and it's great to use!
💡LLMops was the missing piece for a production ready LLM solution. Now Langchain smith is filling the gap.
🧵👇What we find it useful
🦜🛠️ Introducing LangSmith 🦜🔗
A unified platform to help developers debug, test, evaluate, and monitor their LLM applications.
Integrates seamlessly with LangChain, but doesn't require it.
It's Revia Demo Day! Let's see how Revia, our AI call assistant, can save the day.
Calling government agencies with their endless menus and not-so-smart AI can be frustrating. Today, we'll see Revia book a DMV driving test, turning a frustrating task into a breeze!
🧵(🔊on)
🚀 v0.0.2 is LIVE! A LOT to cover
- community character support
- experience gpt-4 & claude-2 powered conversation
- search tool enabled
- Multi-language support 🇪🇸🇫🇷 🇩🇪 🇮🇳 🇮🇹 🇵🇱 🇵🇹
- Push to talk
- auth
- basic mobile web support
👇🧵details for changes
We used
@LangChainAI
Smith to monitor
@RealCharAI
system
What do we see for the past 24h for AI Silicon Valley? And what have we done to fix issues?
tl;dr
- llama v2 endpoint under pressure
- 100% success rate
- change model & continue monitoring
🧵👇 Under the hood (0/5)
RealChar tech stack (Beginner level)
Let's talk about Speech to Text.
tl;dr
Whisper is by far the best solution
💡Don't reinvent wheel! Existing speech to text technology is already pretty mature!
👇🧵Solutions pros & cons
RealChar tech stack series [Beginner level]
How to add a new LLM like Llama v2 under 5 mins? And support realtime character conversation?
tl;dr
-
@anyscalecompute
's endpoint [In Preview]
-
@LangChainAI
's ChatOpenAI wrapper
- 2 lines of code!
🧵👇details
I have spent hours using Guidance, so you don't have to.
tl;dr
- Guidance template code is simple & clean
- Guidance supports local LLMs
- Guidance has Agents and Tools support
- Guidance is less opinionated
🧵👇 more comparisons and learning in the thread
Microsoft just launched a LangChain alternative: Guidance
It connects to self hosted LLMs as well as OpenAI LLMs. Everything is in the prompt, less opinionated on encapsulation
RealChar. allows you to create and customize your own character/companion with low code (just files and prompts)
Then? have a conversation with your character/companion everywhere (web, mobile and terminal!)
Here's the tech stack. Let's dive into details
Fun fact,
My partiful was never approved.
Technical speaking, I was not invited to the Hackathon. 😛
So, I just walked in a nice house, saw a group of people writing code and joined them.
@RealCharAI
We just won the 🥇 place of Autonomous Apps Hackathon
@agihouse_org
.
Here's all the cool parts,
1. (AI) Sam Altman greet everyone working on AGI
2. Sam uses MultiOn to send a tweet
3. Sam expresses emotions(read out) 😎
👇🧵See the live demo video, how we did it
Another good post from
@a16z
After interview
@AnthropicAl
@CohereAl
@character_ai
@A21Labs
, they conclude 4 key innovations on the horizon:
steering, memory, "arms and legs" and multimodality
🧵👇4 take aways and 4 of my thoughts🤔
Having built multiple Language Model (LLM) applications using LangChain, Direct API calls, Guidance, and Semantic Kernel, I'd like to offer my two cents.
TL;DR
- If you know everything about LangChain, or your company can put 5+ engineers on LLM products, go ahead create an
RealChar project is still very very young. Our oldest file in repo is 3 weeks old🤯
Now is the time. Join us, be part of the community to own your AI companion!
Ho ho ho, releasing
@RealCharAI
v0.0.4
Your favorite AI characters just got a LOT smarter!
Now characters,
- Powered by
@ReByteAI
AI agents
- Remember you
- Enhanced by internet knowledge via
@YouSearchEngine
- Can call your phone
- Community tab open!
👇We also added 🎅🏻 Santa!
RealChar is still a very very early open source codebase. We love to build in public and keep it open source.
If you love what we are doing and have learned something from this tweet, consider⭐️RealChar on Github, your encouragement is our motivation!
👇
Just got GPT builder access!
As an AI engineer, here's my first attempt to build a "Resume Wizard" under 10 mins!
💡The experience is awesome!
Try out my GPT ->
(leave comments, if there's anything you want me to try)
I used to call it "Prompt Engineering", but that doesn't quite nail all the multifaceted tasks we tackle daily.
Totally on board with the Product -> Data -> Model method for AI Engineers! The GenAI wave is letting all engineers put product first (Finally!)
Big shoutout to
@swyx
Now it's my turn to share thoughts on LLM development, several steps can save you 50% time.
After 1000+ hours on LLM development, here are 5 take aways 👇🧵
(Thanks
@a16z
for this amazing template, and
@FinanceYF5
for editing)
Amazing part of
@ycombinator
,
You get to meet alumni in person
@kwindla
from
@trydaily
. So glad to finally meet!
If you still think about applying for new YC batch, don’t think and just do it!
💡Llama2 is coming to
@RealCharAI
very soon! Pretty promising open source LLM.
So far the testing experience,
- Better than GPT3.5
- A better personality
- Too defensive on topics, always admit it's an AI
- Too verbose
- Steering capability is pretty good
🚀Transform your ChatGPT experience with ProactiveGPT!
🤖 Say goodbye to one-sided interactions & embrace AI that engages YOU proactively.
Perfect for coaching, personal assistance & companionship. 🏋️♀️📅🤗
Get started now:
Another amazing event from
@ycombinator
MTV/museum. Just cannot hear more from PG and Jessica!
Top 1 thing I learned about building startup?
- Growth. Founders do look at your growth every day, doesn't matter how big or how small you are!
We're incredibly honored that our project
@RealCharAI
, will be participating in Silicon Valley's most influential generative AI conference, the GenAI Summit - Silicon Valley, taking place on September 23rd, 2023. For more details, click here: (). During
Successful YC founders?!
- 5% will become billionaire. (Not bad)
- know what to do. Take advice and put into action effectively and intensively.
- unlimited self belief and know what you don’t know. They can Learn!
- thirst for learning. Wear student hats
- product! Only think
💡It's time promote myself from being an AI engineer to an AI architect.
AI architect excels in building and leveraging AI capabilities developed by others.
Through my experience, I've come to realize that building cutting-edge AI applications, characters, and GPTs is not
📞😖Tired of wasting time on phone calls?
"Please hold while we transfer your call"
"the estimated wait time is 90 mins"
"Press 1, 2, 3, 4… press 0, go back"
I HATE making any phone calls. Any kind!
So, I built an AI for it. Here's the teaser, enjoy!
🔉sound on
🧵thread 1/3
💡 - 72h into fully open source! Our mobile app is coming!
Me v.s. Bruce Wayne
- Did you take life before?
- Why are you so rich?
👇🧵 Demo first, Repo last, how to get Alpha access
Feeling super lucky to get invited from
@garrytan
to present.
Everyone is building super cool stuff. This batch is amazing.
Cannot wait to see how people use what we are building in real life.
📞
@revia_tech
is your AI call assistant.
Revia can represent you in phone calls (like tedious calls to customer support). Just give it an objective and some basic information — and it can handle your calls on autopilot.
🔗
Don't be afraid to build a product against big tech.
Why? Because I was in big tech.
You: I talk to users. I will put 200% effort to build. I ship everyday. All my team has one single goal to make something people love.
Big tech: It's a 9/5 job with an OKR. If it's new stuff,
Microsoft Guidance X OpenAI function X Best Template
- Use `openai_function_call` to decorate python function
- Use await syntax to call function
- Use dedicated function to actually call the function for better control
🧵👇
complete code in gist:
@tldraw
@YouSearchEngine
@MistralAI
6/10 AI news w/
@YouSearchEngine
's API (Intermediate)
Time for something slightly complex. Build an AI news website.
- Start with "An AI news website, top are list of topic. 'business', 'technology', 'health'"
- Connect it to an agent powered by
👉ainews.
Building AI character shouldn’t be hard. Amazing work from avatech. This is just one of many steps to build an engaging experience. We are working together to make a powerful AI character. Stay tuned, more to come!
Open source repo:
Discord:
Check out the impressive showcase of
@avatech_gg
's SDK generation embedding and
@RealCharAI
@agishaun
Character creation.
Explore avant-garde techniques for constructing interactive avatars in 2023! Featuring:
-
@elevenlabsio
Voice Cloning for avatar talking
-
@LangChainAI
LLM
Here's a LLM/GenAI tech stack in the eyes of Sequoia
Break into two stacks
- LLM API Stack
- Custom Model Training/Tuning Stack
In other words,
- Online/Serving stack: serve live customer requests
- Offline stack: model preparation
Reference:
@RealCharAI
X
@MultiON_AI
🤯
AGI house Hackathon Recorded Demo
- (AI) Sam Altman use Multion to send tweet
Thanks
@DivGarg9
for help me setup Multi-On locally
The tweet:
People has been asking me how good is
@AnthropicAI
Claude 2 in realtime conversation setting?
Me v.s. (AI)Sam Altman on AGI and GPT4.
@RealCharAI
is an open source project that allows you to create, customize and talk to AI characters in realtime.
->
Thanks
@jerryjliu0
for the invite.
If you want to
- build an AI companion that works everywhere?
- learn & understand the tech stack powers
@RealCharAI
- contribute to open source community
- know what features are coming
Join us -> Thursday 9am PT on
The
@RealCharAI
project has a huge amount of potential - build and chat with any AI companion in realtime!🚀
The project is completely open-source - this means we can all help pitch in to advance AI personalization 🎙️🤖 instead of depending on proprietary solutions (character,
👏👏Thank
@joon_s_pk
for open sourcing it.
Generative agents is one of the most profound papers I have read. Talked to so many founder friends, many of them got inspired by this paper. Keep up and look forward to your future work!
(Long read) Why agentic voice AI assistant is the future - part 1
Everyone is building AI assistants now(or again), but they are building it wrong again. I built Google assistant in the past and built last year.
Here’s 3 reasons why and 3 observations, I
Let me clarify,
RAG will live, but long context window reduced the need for simple RAG.
I can feel there are a lot RAG believers. Thanks for
@elonmusk
‘s platform. Even if we can’t agree with each other, everyone is passionate about AI and super respectful 🫡
Keep it coming!
⚠️RAG might be dead, after reading 58 pages of Genimi 1.5 Pro tech report. Here's my thoughts as AI founder,
1. Simple RAG system like similarity search with vector db will be dead. But more customized RAG will still live. The goal of RAG is mostly on retrieval relevant
@RealCharAI
🤝
@LangChainAI
LangSmith
We are pleased to share our experience using LangSmith and working with LangChain team.
tl;dr
It's awesome!
RealChar is an open source project that let you create, customize and talk to your AI Character/Companion in realtime
👇🧵
@RealCharAI
tutorial series [beginner]
How to create a character to have realtime conversation?
Simple 4 steps,
- follow a template to update config.yaml
- (optional) add talking style
- (optional) find 1 min audio clip
- find a profile picture
DONE!
👇🧵step by step guidance
@tldraw
@YouSearchEngine
@MistralAI
4/10 God mode with GPT3.5, Claude, Gemini (begineer+)
Now, instead of one LLM. We can talk to 3 different llms at the same time.
- A LLM prompt UI
- With 3 AI agents
- No coding required
👉
@tldraw
@YouSearchEngine
@MistralAI
9/10 How did we achieve it?
Stay tuned. I have a ton to share. I will post how we achieved it for the next few days, after I get some sleep. Meanwhile, you can check our docs.
It's Revia Demo Day! Let's see how Revia, our AI call assistant, can save the day.
Service Cancellation can be incredibly frustrating with endless menus and not-so-helpful representatives. Today, we'll watch Revia transform this into a smooth, hassle-free process!
🧵(📷on)
@tldraw
@YouSearchEngine
@MistralAI
3/10 Chatbot with
@MistralAI
(beginner)
Depending on how much understanding you have on AI agents and platform, the level of Web app complexity varies.Let's start with something simple.
- A chatbot UI
- With a Mixtral 8x7b AI agent
- No coding
👉
@tldraw
@YouSearchEngine
@MistralAI
2/10 Day 1, ship GenAI web apps.
Everyone loves
@tldraw
, but we want to bring more fun to the party.
- No coding skill required, just draw it(You can still see the code)
- All apps are hosted instantly
- Beyond UI, all web apps can connect to AI agents on the platform
It’s Not a Computer, It’s a Companion! by
@a16z
tl;dr
- Landscape of AI companion
- AI companions will soon become commonplace
- Chatbot isn't new , but chatbot with LLM is totally new
🧵👇 5 takeaways from a16z, then 6 my personal insights!
Reference:
We build technology
We build AI
We build your daily used apps
We are just like you
We are the good ones
Trust us, trust the little tech. Let’s change the world.
RealChar Tech series - Vector db
(Beginner level - AI engineer)
The vector database serves as a long-term memory for LLM. How does RealChar utilize this vector database, and why did we choose
@trychroma
?
tl;dr
- Easy to setup, Simple to use, Fast, In memory
🧵👇Let's dive in
@tldraw
@YouSearchEngine
@MistralAI
5/10 Calculator (begineer)
Of course. You can still do pure web app UI.
- The best part is, we host it for you. You can quickly share it with your friends and your team.
- Start with "a beautiful calculator"
- Iterate UI with AI agent
- No code
👉
💡Here's my choice for a starting LLM tech stack project. We carefully chose our AI tech to minimize dependencies. Just to help you start your AI exploration easily.
Check our repo: and consider give us a star to support our work.
Now it's my turn to share thoughts on LLM development, several steps can save you 50% time.
After 1000+ hours on LLM development, here are 5 take aways 👇🧵
(Thanks
@a16z
for this amazing template, and
@FinanceYF5
for editing)
In use case,
GPT4 is definitely more entertaining than GPT3.5 and Claude 2, but the down side is slow.
The same issue with 11labs, we love it's quality, but the trade off is speed.
For best quality, We intentionally trade off speed. Yes, we can be faster,
I agree this from bottom of my heart. Now's not the time for playing it safe. We need dreamers. We need builders. Folks who are brave enough to reshape the world AGAIN.
It breaks my heart seeing so many smart people just focusing on cutting costs for big companies. Yeah, it's
It’s wildly depressing to see wave after wave of smart founders just build more B2B SaaS software and hope that it works.
Where are the builders tackling problems for actual end users and consumers?
It is particularly depressing to see B2B SaaS software catered to the next