Div Garg
@DivGarg_
Followers
20K
Following
2K
Media
240
Statuses
1K
Building the future @the_agi_company | Founder @MultiOn_AI | Adjunct Faculty @Stanford CS | prev @Apple, @GoogleAI, @Nvidia, @Uber
Stanford, CA
Joined April 2016
We just solved the long-horizon planning & execution issue with Agents 🤯!. Excited to announce that @MultiON_AI can now take actions well over 500+ steps without loosing context & cross-operate on 10+ websites as part of a single task 🤩!!. Our latest upgrade cracks the goal
78
155
1K
Ever thought ChatGPT could order you dinner? Turns out it can, and even do lots more 🤯. Introducing MULTI·ON: .Your AI Web Co-Pilot powered by ChatGPT . Sign up 👉: And here's me ordering & eating the *world's first burger ordered by ChatGPT!*
36
190
1K
Unveiling the first Super App in the @OpenAI plugin App Store: the MULTI·ON Browser. Well, what can it do? Literally Everything!. MULTI·ON gives ChatGPT superpowers to interact with anything on the internet on your behalf 🤯. The Web Browser just got reinvented in the age of AI🚀
LIFTOFF!.
58
136
1K
Excited to announce that the lectures for CS 25: Transformers United V2 ( are now publicly available 🤩!. Our intro lecture featuring @karpathy is out and available to watch here 👉:. Spread the word 🤗!.#AI #Transformers.
7
161
718
Announcing the public release of our lectures from the first-ever Transformers seminar: CS25 Transformers United ( held at @Stanford. Our intro video is out and available to watch here 👉: Spread the word 🤗!.#AI #Transformers.
13
149
714
Compute Secured. We are building Foundation AI Agents. The time to join is NOW 👇.
26
28
571
Although my profile still says on leave of absence from CS PhD @Stanford, . I am officially dropped out without a current option of going back. You can don't do REAL great things without taking real risks
39
5
567
We won the #1 place 🎉 at the #langchain AGI Hackathon this weekend building Life Copilot 🔥🤩. Life Copilot is an AI assistant that can understand text + images & control a browser to surf the internet using @MultiON_AI to do things for you! . See the demo below 👇: (1/N)
18
71
503
Introducing MULTION VISION - a robust Vision to Actions Agent that can reliably interact with 95% of the Web using just pixels 🚀. Here's a demo of it autonomously ordering a burger on Doordash 🤩:
31
66
491
Super excited to announce we will be publicly releasing the lectures of CS 25: Transformers United V2 ( held @Stanford over the next week 😀. See the course preview below 👇:. Release schedule to follow 🤗.#AI #Transformers
6
114
460
I was wrong. Agents will take your job. We are sorry for this casualty @MultiOn_AI . this is 20 pages on one of the best bot-protected websites on the web in a 48hr challenge.
12
53
445
Very excited to introduce our concept of Internet of Agents! 🤩. We gave @MultiON_AI the ability to launch Manager agents & spin up as many worker agents as needed on demand that can do tasks in parallel & communicate to do work 🤯!!. Now anyone can spin up a swarm of 1000 -
26
54
413
We had our first CS25: Transformers United class with a special surprise guest lecturer: @karpathy 😊!!. A great talk on Self-attention & Transformers taught in the unique Karpathy style. Our amazing students showed up despite the rain and we also got a post-class lunch 😀
10
21
407
Thrilled to share a monumental achievement 🤩: . Our Web AI Agent, @MultiON_AI, just made history passing the online California Driving Test, becoming the first virtual AI to earn a driving permit in CA! 🤯🚀. This marks a groundbreaking moment: the first fully autonomous
31
40
380
One simple way @OpenAI is looking to acquire superhuman data is to have models be evaluated with human + AI instead of pure human annotators. Theoretically this scales linearly with AI progress rather being upper bound by human skills. Here's slides from @janleike - co-lead of
Proposed Theorem: it’s impossible to acquire super humanity skills when relying purely on data created by humanity. Why do I say super humanity and not super human? Because for example a translation algorithm is already better than any single human in terms of how many.
16
57
382
Excited to announce that one of the most popular CS class at @Stanford - CS 25: Transformers United - is back this Fall. Now in V3! . discussing and breaking the latest breakthroughs with Transformers: 👇 (1/N). (w/ @stevenyfeng, @karanps, @chrmanning).
5
46
368
We came out of stealth today 🚀!. Very excited to today officially announce ourselves @MultiON_AI, our vision for AI & Agents and thank all our many backers and investors! . Read our first official blog: Building a Brighter Future for Humanity with AI Agents.
Today we are excited to officially announce and share our story and vision for the future of AI and Agents in everyday life. We are excited to partner with @generalcatalyst, @Amazon Alexa Fund, @mavenventures, @SamsungNext, individuals from @OpenAI, early backers of.
39
26
367
Excited to finally share my blog post on Software 3.0 🔥. The post was originally written before GPT-4 release, and the predictions seem to have held surprisingly well. Knowledge cutoff date 28 Feb 2023. . Please read and share!!.
19
66
318
How it started exactly one year ago.
We came out of stealth today 🚀!. Very excited to today officially announce ourselves @MultiON_AI, our vision for AI & Agents and thank all our many backers and investors! . Read our first official blog: Building a Brighter Future for Humanity with AI Agents.
9
8
296
We are excited to announce the release of our Agent API in public beta along with our Agent Playground @MultiOn_AI 🚀🚀. Our Agent API is the first-ever autonomous API that allows going from natural language to web actions using our Language Action Models (LAMs) all running in
Introducing Agent API: bringing actions to your devices and applications.
20
31
296
Agent 🍓.
Announcing our latest research breakthrough: . Agent Q - bringing next-generation AI agents with planning and AI self-healing capabilities, with a 340% improvement over LLama 3's baseline zero-shot performance!
18
20
297
We achieved something this morning with has never been done before in history🔥. We see it as a major milestone for real-world AI. Very excited to announce publicly soon ✨.
22
10
280
Excited to announce that this winter, we are bringing back CS25: Transformers United V2! at @Stanford -discussing and breaking down the latest breakthroughs with Transformers: 👇 (1/N). (w/ @stevenyfeng, @RylanSchaeffer, @chrmanning) . #AI #Transformers.
6
51
279
Now you can effortlessly retrieve structured data from any website using just a natural language prompt 👀. Introducing the Natural Language API for Universal Web Understanding 🌐. OR. Automated Scraping On Steroids 🔥. Imagine 8 billion people able to build a scraper on the fly.
Introducing Retrieve API: the best-in-class autonomous web information retrieval API. Developers love our Agent API ❤️. Since its launch, we have consistently received feedback that many use cases rely on intelligently leveraging the Agent API to retrieve information from the
12
27
272
When you win a @StabilityAI hackathon, you get their CEO & Founder to come to your class 😃. We had the pleasure to host @EMostaque for our CS 25: Transformers United final mixer. Thanks to everyone for making the class a huge success!. Very excited for @StabilityAI's plans &
5
17
260
Had a blast reconnecting with @goodfellow_ian since I interned with him @Apple. fun chat on agents & nuclear fusion
9
2
257
Announcing the AI Agents 2.0 hackathon hosted by @MultiOn_AI x @AgentOpsAI 🔥. We are excited to launch the era of Agents 2.0 - reliable, observable & safe agents that actually work 🤯. 👉 Come build the best Agentic Applications & hardware integrations
7
43
237
We unlocked a big achievement in AI 🚀. Very excited to announce that our Web AI Agent @MultiON_AI in default mode now runs faster than the human speed of clicking and typing 🤯🤯!!. The implications are potentially huge - this makes it irrelevant for you to manually interact.
19
27
226
Welcome to the era of Generative Browsing 💥🔥. We have reimagined how you interact with the internet and are excited for you to try it 🎉. Intro video: Full thread below 👇.
Unveiling the first Super App in the @OpenAI plugin App Store: the MULTI·ON Browser. Well, what can it do? Literally Everything!. MULTI·ON gives ChatGPT superpowers to interact with anything on the internet on your behalf 🤯. The Web Browser just got reinvented in the age of AI🚀
9
53
217
The Agent Hackathon was an extraordinary event! Proud to have initiated and organize it with @agihouse_org. It was an honor to be amongst an all-star opening panel including @karpathy, @DrJimFan, @hwchase17, @swyx, @amasad, @adamdangelo 😀🌟 . Takeaways & Winner announcements 👇
7
20
219
Chapter II. Update: I will be moving on from @MultiOn_AI to explore new things! . I am enthusiastic about the road ahead and remain as an advisor to the team. Excited for what's NeXT.
Today we want to share a few year-end MultiOn updates. We're doubling down on our renewed consumer vision for AI and have fantastic new teammates to bring it to life. And to cofounder Div Garg, who recently departed the company, we thank him for his contributions and look.
22
3
218
Week eight of CS25 - Transformers United: We had .@scychan_brains & @ToniCreswell from @DeepMind give fascinating talks on In-Context Learning & Faithful Reasoning in LLMs 😃 . Stephanie discussed how data distributional properties can drive emergent in-context learning in
3
31
197
Announcing @MultiOn_AI x @modal_labs! 🚀. Now you schedule long-running AI agents to do things on your behalf autonomously in the background everyday 🔥. Try out our Twitter AI News Agent that finds the latest AI news everyday & tweets it out 👇:.
7
21
190
Very thrilled that @MultiON_AI got a news feature from Stanford HAI for turning ChatGPT into Web-Browsing Co-Pilot 😃. Please read and share!!. @StanfordHAI, @StanfordAILab.
11
49
182
I am very excited to announce we (w/ @ChetanyaRastogi, @advaypal, @chrmanning) will be organizing the first Transformers-only seminar: CS25 Transformers United! at @Stanford focusing on the latest breakthroughs with Transformers in diverse fields: (1/N).
7
32
167
Sharing my condolences on the loss of Mr. Ratan Tata, a visionary leader and extraordinary human being. At 16, I was fortunate to receive a full scholarship to study in the US, thanks to his generosity—a gift that profoundly shaped my life's journey. Here is the letter of thanks
7
5
171
Announcing the release of one last special lecture for CS25 Transformers United ( given by the AI Godfather @geoffreyhinton 🤩!!. See our watchlist here 👉: Deep thanks @chrmanning, @StanfordAILab, @StanfordOnline.#AI #Transformers.
2
36
160
Here's the first flight ever to be fully booked by an AI ✈️🔥😎. Welcome to the next generation of Artificial Intelligence 🚀. You don't wanna miss this event (@MultiON_AI) .👇(unmute)
Going to the most amazing hackathon ever held 🤩🔥. We will be building autonomous AI agents that can take Actions and are able to do anything & everything over the internet using @MultiON_AI 💥🚀. Come thru! Maybe will see sparks of AGI :). (capacity.
16
22
158
Awesome catching up with @JeffDean at Google i/o after 4 yrs since we got lunch during my time @Google!. Was really fun to chat about his early Google days & our similar paths with starting @MultiOn_AI 😀. He was blown by some new things we are cooking 👀
5
7
160
Got to give an instructor lecture at CS 25: Transformers United @Stanford with @stevenyfeng & @_KaranPS_.on Going beyond LLMs: AI Agents, BabyLMs & Intermediate-Guided Reasoning. Made me fully realize that that General-Purpose Agents are indeed going to be the next big thing.
7
15
161
Developing Agent Q (🍓) at @MultiOn_AI has been an extraordinary journey with incredible folks! It feels like @rm_rafailov was practically living in our office!. We’re currently looking for exceptional talent to join us in our journey to push the frontier of AI Agents and.
Super excited to announce what we have been working on in the last six months - Agent Q is out now! This is a framework for self-supervised agent reasoning and search that can self-correct and autonomously improve by self-play and RL on real tasks on the real internet! 👇
13
13
157
It's a Christmas miracle ✨. Now official in Delaware. Fun times ahead 🚀 . ho ho ho
Chapter II. Update: I will be moving on from @MultiOn_AI to explore new things! . I am enthusiastic about the road ahead and remain as an advisor to the team. Excited for what's NeXT.
18
6
156
And all Vector DB startups are UI wrappers around KNN applied at scale - a naïve implementation of which takes an hour for a freshman 😂.
5
14
143
The last time I attended @NeurIPSConf as a participant in '22, my wish was to have my own AI startup & have our logo up as a official sponsor of one of the best AI conference on Earth! 🔥. Wishes do to come true and if all goes according to the plan, we might as well be throwing.
7
7
140
Prediction:. By early 2025, the number of AI agents active at a moment will exceed the population of humans on Earth.
10
13
138
Exciting progress applying novel RL-inspired loss functions to enhance LLM training, building on my original IQLearn work! (. We’re breaking out of the GPT-style next token prediction training, discovering new algorithms to advance large model.
Imitation is the foundation of #LLM training. And it is a #ReinforcementLearning problem!. Compared to supervised learning, RL -here inverse RL- better exploits sequential structure, online data and further extracts rewards. Beyond thrilled for our @GoogleDeepMind paper!. A
4
15
121
Excited to announce that going forward the MultiOn name and brand will be part of AGI, Inc. (@the_agi_company). 🚀🚀.
20
2
122
✨ Early sparks of a working proto JARVIS 🤩. Daily Uber ride: (now reaching ~99% accuracy on this task with @MultiON_AI). Watch in audio 👇
7
19
113
Almost a year ago, I pitched to a couple of VCs that training & running large AI models are going to become the hottest & most urgent problem in the Software space. Looking at the exponential scaling trends in model size & compute, it was obvious. I was pretty much laughed at as.
10
4
108
Excited to announce that our Web AI Agent.@MultiON_AI is now 10X faster than human finger speeds 🚀🚀. Here's our very first SloMo video shot in 0.2x 🔥:.Watch it in audio 🔊 to see the difference! . You can choose for yourself between human finger typing or our AI agent which
7
9
107
Very excited to announce the official launch of the MultiOn API 🚀. Now anyone can build on top of MultiOn to take Actions on the web 🔥🤩. Check out some fun Autonomous App examples we have already released:
.@MultiON_AI is bringing the power of agents to the web!. They're one of the first real world, production agent applications we know of. Their integration with LangChain Toolkits makes it easy to personalize and automate everyday tasks. more on the blog.
9
17
104
We had our first CS25: Transformers United V3 class at this week with @sharan0909 - the Llama2 team lead at @Meta!!. A great talk on the details of Llama2, beating the Chinchilla Optimal, and why data quality matters more than the number of params of an LLM. Anticipating Llama3
0
8
106
This is like using a rocket to crush an ant, but what tech luxurie looks like in 2024 ✨. (Request for beta access on our Discord 😎). 🔊
16
12
104
Now you can auto spin ec2 instances using Agents!! 🚀. #MultiOn #Agents. Up next : we will let @MultiON_AI spin up its own gpu training jobs 🤩. 🔊
7
12
92
Introducing the MultiOn Task Scheduler - an Autonomous Desktop App we built on top of the @MultiON_AI API & @LangChainAI 🚀. With @MultiON_AI, my computer comes alive every morning and goes and wishes happy birthday to my friends on my behalf 🤯. Now SKY IS JUST THE LOWER LIMIT.
12
15
101
Agents still doesn't stop amazing me even with everyday use ✨!. This task is actually hard with hidden drop downs & an un-intuitive UI which can be confusing even for a human & requires the AI to be capable of predicting "what's next" without hallucinating. but @MultiON_AI fully
3
10
100
classic Palo Alto hacker garage. (no the bed is not mine, but the robot is)
5
3
98
Introducing a new work that holds a very special place for me and hope may prove monumental to the Statistics & RL communities. Project: (w/ @JoeyHejna, Matthieu Geist, @StefanoErmon).@StanfordAILab @GoogleAI . I will try to best explain it below 🧵 (1/N).
1
10
99
will add 🦄 slayer to resume.
Today, we’re announcing some updates to our strategy and some changes to our leadership and team. More details are in our blog:
9
1
96
Our agents are becoming too good!! . Although, maybe you don't want your agents to do everything (yet!) 😂. ⚠️ DO NOT attempt this at your home or office!
10
8
91
Excited to chat on the future of the Internet!. We at @MultiOn_AI are building the Motor Cortex layer for AI to enable seamless natural language interaction with the Web, with millions of concurrent AI Agents running in the cloud to interact with any web service . This is the.
We’re excited to be hosting a webinar on the future of web agents with none other than @DivGarg9 from @MultiOn_AI 🌐🤖. MultiOn is an advanced platform enabling you to build personalized web agents that perform tasks online from start to finish (e.g. scheduling meetings to
9
15
92
This is really WILD 🤯🤯. Turns out we supported Mobile all along with @MultiON_AI, but just didn't know!! . One of our early users had our AI Agent do Walmart shopping on his behalf to make Spaghetti by automating his phone!! 🤩🔥. Now a personal & private AI agent that can do.
Here's a glimpse of mobile AGI and how it will change the way we use our devices forever (I had to stop it before it went through and actually placed the pickup order!). Running on my Android phone (Moto G Power 2021) via Kiwi Browser and @MultiON_AI's browser extension. 👏
9
18
90
I am likely living a year ahead of most people when it comes to seeing Autonomous AI becoming widely deployed & large in the society . If we tread the right paths,.the near future is going to be the golden era of humanity 🚀.
6
5
84
Introducing our new SOTA imitation learning method "IQ-Learning" that outperforms prior works by up to 7x on Atari. To be presented in @NeurIPSConf 2021 with a spotlight!. Arxiv: Github: (1/N)
3
15
86
Going to the most amazing hackathon ever held 🤩🔥. We will be building autonomous AI agents that can take Actions and are able to do anything & everything over the internet using @MultiON_AI 💥🚀. Come thru! Maybe will see sparks of AGI :). (capacity.
14
12
84
I woke up in the morning and found this dude sleeping in my garage 😲. The Llama functions API (@llamaAPI) was built overnight by @edreisMD in our Palo Alto garage & @MultiON_AI is proud to be the sponsor 🤩🔥. Try it at: #built_in_garage #multion_house
🔧🦙Function calling with Llama2?. In our first addition to 🦜🧪langchain_experimental, we're excited to integrate with llamaapi. LlamaAPI is a hosted version of llama2 that adds in support for `functions` in the same way @OpenAI does. Thanks to @edreisMD, links in 🧵
4
11
85
Had a fascinating chat with @AndrewYNg on the future of web AI Agents . & how they are poised to revolutionize how we interact with the digital world and get things done 🚀. Big things ahead! 🌐 @MultiOn_AI
5
7
78
AI models now are a true race to the bottom. This is going to become one of the biggest commodities in 2024. Pls go work on something different if you are in AI. There is still much more left to be done than pure supervised training & architecture optimization. AI ≠ model.
8
10
75
Thanks @Jason for giving us a rare A+ for @MultiOn_AI as the 100th AI demo on TWiST pod & lurking in our discord 🤩. We are excited to define the next generation of useful Artificial Intelligence 🚀
The 100th AI demo on the show!. MultiOn: --- AI that handles tasks in the background of any digital experience.--- Chrome Browser extension. @Jason: A+.@Sundeep: A+
7
7
77
Week seven of CS25 - Transformers United: We had .@vivnat from @GoogleAI, @GoogleHealth give the first public talk on Med-PaLM: the current SOTA Transformer model for Medical Q/A & on building Foundation Models for Medical Science and Healthcare . A very interesting talk
5
3
72
Hot take: Coding in an early stage startup is like diffusing a high-stakes hostage situation. Every day, a hostage dies by default. You have to take as many headshots as fast as possible (release features, fix bugs) & eliminate the bad guys (technical debt, blockers) while.
5
6
72
Time to bring out the big guns ⚔️!. This beast can run 100,000 concurrent agents locally 🚀🚀. We gave a new meaning to residential IPs 😏
6
6
67
Just as the TCP/IP protocol was the foundation of the Internet, we believe in the need for a robust Agent Protocol to empower agents to inter-communicate and work with each other - eventually which may culminate in the foundation of a new Internet of Agents. Join us in the.
We're hosting a small Twitter space on Monday 11am PT and will discuss Agent Communication Protocol and how we as a community can build this in public together. We'd love for everyone to join!.
4
13
67
Ever wondered how can we make robots 🤖 understand natural language to solve complex long-range tasks? Introducing LISA: Learning Interpretable Skill Abstractions, which learns a middle human-interpretable language protocol to condition agents. 👇 (1/N)
2
10
64
@MultiON_AI The idea is to simply leave it running it everyday and have it to do tasks for you in the background (more scheduling features soon!). We have set up a X handle where @MultiON_AI will go find & tweet about the latest AI news everyday 🤩. (no need for AI news influencers anymore.
3
3
66
Well this blew up 🤯. We will be having the amazing @mckaywrigley and @sxwy joining as co-organizers along with @agihouse_org for the Autonomous Agent hackathon. @karpathy will also be joining to give an intro talk. To give more time to prep, the hackathon will now be happening.
Going to the most amazing hackathon ever held 🤩🔥. We will be building autonomous AI agents that can take Actions and are able to do anything & everything over the internet using @MultiON_AI 💥🚀. Come thru! Maybe will see sparks of AGI :). (capacity.
6
7
63
Announcing our new revamped developer docs for our Agent API .@MultiOn_AI and release of 1.0 Python/TS SDKs 🔥! . The new destination for all your action needs: .` . Checkout 👇.Docs: .Github examples:
We are excited to announce the release of our Agent API in public beta along with our Agent Playground @MultiOn_AI 🚀🚀. Our Agent API is the first-ever autonomous API that allows going from natural language to web actions using our Language Action Models (LAMs) all running in
7
7
63
Week three of CS25 - Transformers United: We hosted @_jasonwei from @GoogleAI - the first author of very interesting works like Emergent Abilities & Chain-of-thought reasoning in LLMs. A great talk on improving reasoning & the phase-change like Emergence property of LLMs!
3
4
63
Plan to onboard 1000 folks next week to @MultiON_AI 🤩, who wants access?. Our waitlist is now reaching 30k, and folks with the most referral signups & interesting use cases will be prioritized. You can also reply on this thread below to get included in the first batch 👇.
65
7
61
It seems clear that the first thing made irrelevant by AI & @OpenAI releases will be programming itself. Software productivity is going to become 100x & get democratized to every person regardless of age and degree. Here are the levels of Autonomous Programming I see happening:
8
5
61