- Everyday AI
- Posts
- Autonomous Agents Arrive
Autonomous Agents Arrive
Chain together high level requests, and the robots will take it from there.
Autonomous Agents Arrive
Chain together high level requests, such as: "Increase net worth, grow twitter account, develop and manage multiple businesses autonomously," and the robots will take it from there.
The Big Stuff
Last week, we saw the start of open source autonomous agents, or applications that can take in high level commands like, "Look for upcoming food events, and then suggest a recipe that suits it.". Projects like Auto-GPT, Baby AGI, and Microsoft Jarvis give us a glimpse into what Andrej Karpathy, founding member of OpenAI's research team, calls "the next frontier" for prompt engineering. So what's the buzz about?
Next frontier of prompt engineering imo: "AutoGPTs" . 1 GPT call is just like 1 instruction on a computer. They can be strung together into programs. Use prompt to define I/O device and tool specs, define the cognitive loop, page data in and out of context window, .run().
— Andrej Karpathy (@karpathy)
6:44 PM • Apr 2, 2023
AutoGPT and BabyAGI are two open source autonomous agent projects that were both released within the last two weeks. Since they've been released, they've exploded in popularity on code repository Github, with AutoGPT getting 16.9k stars, and 4.6k stars for BabyAGI. So what do can projects do, and why are they important?Up until recently, large language models, like ChatGPT, have been limited to relatively small tasks, but lacked the ability to do accomplish high level tasks, such as "Build me a website". Currently, GPT-4 is very good at coding, but it is limited to relatively small tasks. Building an entire app, or a website would not be possible without issuing a lot of prompts. Autonomous agents can make a high-level plan for what to accomplish, break it into tasks, perform each task, then reflect on its work. In other words, very similar to how a human would do it. This is so early, that we're just beginning to understand the capabilities here. Let's look at a few examples (many coding examples since early adopters are engineers):
Solving AI Alignment (click to watch video):
I asked my autonomous AI to solve AI alignment.
Fascinating to watch.
Agents:
- Task execution: text-davinci-003
- Task creation: text-davinci-003
- Task reprioritization: text-davinci-003— Yohei (@yoheinakajima)
2:01 AM • Apr 3, 2023
Coding a website (click Tweet to watch video)
Alright, this is getting too crazy. Soon you won't even need to code anymore.
I setup AutoGPT and it I asked it to build a website for me.
And it succeeded. In under 3 minutes. Using react and tailwindcss. All by itself.
— Sully (@SullyOmarr)
2:08 AM • Apr 7, 2023
Figuring out how to install dependencies (click tweet to watch video):
autogpt was trying to create an app for me, recognized I don't have Node, googled how to install Node, found a stackoverflow article with link, downloaded it, extracted it, and then spawned the server for me.
My contribution? I watched.
— Varun Mayya (@VarunMayya)
9:03 AM • Apr 6, 2023
Searching for destinations
Note: this is RoboGPT, a variant of AutoGPT
Introducing the latest demo of RoboGPT 🤖
Watch how it searches the web for top digital nomad destinations and consolidates the info into a CSV file! 🌍💻
Check out the code on GitHub: github.com/rokstrnisa/Rob…
#GPT4#AutoGPT@SigGravitas@levelsio
— Rok Strniša (@RokStrnisa)
5:45 PM • Apr 4, 2023
Microsoft JARVIS
Microsoft JARVIS is an autonomous agent project that performs autonomous tasks, but accomplishes it by combining multiple artificial intelligence models. For example,
Prompt: Please generate an image where a girl is reading a book, and her pose is the same as the boy in the image example.jpg. Then please describe the new image with your voice.
Input Image (example.jpg)
Output Image
JARVIS would take this prompt as input and then plan out how to accomplish the task. In this case, it would pose control, then post to image, then image class, then text to image. It would then select the appropriate artificial intelligence models to use, execute the tasks, then generate the response.
These autonomous agents have only been out for a couple of weeks. Some people have been referring to them as primitive AGI. We see a lot of potential here, and we're excited to see what's next in this area.
Midjourney "Describe" is Impressive
Sometimes, it's the seemingly small features that can have outsized impacts. We believe Midjourney's "describe" feature is one of these. Instead of starting from scratch with a text prompt, the "/describe" command in Midjourney allows you to upload a photo, and get four (4) text prompts back that let you create new work based on those interpretations. For example, the Starbucks logo:
I asked Midjourney v5 to '/describe' some logos, to see how it would create prompts for them, and to see what it would create in response.
Starbucks
— fofrAI (@fofrAI)
10:45 PM • Apr 4, 2023
Other big news:
Is OpenAI's stranglehold on the competition is loosening? Google plans to roll Chat AI into search and Anthropic is looking to 10X "The best AI of today." OpenAI share their safety approach while Microsoft expands Edge's AI image generator, and Stanford lets us know what's happening with AI. Oh and Italy blocks ChatGPT.
Google CEO Sundar Pichai Says Search to Include Chat AIAnthropic’s $5B, 4-year plan to take on OpenAIMicrosoft’s rolling out Edge’s AI image generator to everyoneOpen AI releases their approach to safetyStanford releases annual AI Index ReportItalian privacy regulator bans ChatGPT
One more thing...
This is awesome 🔥
@WonderDynamics is project to follow.— Smoke-away (@SmokeAwayyy)
5:58 PM • Apr 8, 2023
Smaller But Still Cool Things:
Poe.com introduces custom AI chatbots, which lets you create a custom chatbot based on a prompt. Try the Greg Mushen AI bot, which will take anything you talk about, and somehow turn it into an AI discussion (link)
AI doomsday scenarios are literally impossible (link)
Open source project LangChain raises $10M seed round (link)
Using A.I. To Make Drake Rap About Beans (link)
An LLM running on a calculator? (link)
DoNotPay is launching a new GPT-4 email extension to troll scam and marketing email/text messages by engaging them in an endless A.l. conversation (link)
Attract your first 100 customers with LoopGenius (link)
Going Deeper
LLMs are Better Than Human Annotators (link)
LinkedIn Learning is offering over 200 free courses on AI until June (link)
NASA releases DAGGER, which uses artificial intelligence to predict solar flares (link)
LLMParser makes it easy for anyone to classify and extract structured data from text with large language models (LLMs). (link)
Brainstorming ChatGPT Business Ideas With A Billionaire (link)
Tweets of the Week
Think of ChatGPT with plugins as something akin to WeChat. Instead of a social network, it’s a network of 2 - you and your AI… that’ll do whatever you need. All the regular ChatGPT stuff + advanced math, travel, banking, shopping, reservations, whatever “app” you want/need.
— Adam.GPT (@TheRealAdamG)
3:50 PM • Apr 7, 2023
Companies saying that their goal is to make an LLM 10x more powerful than GPT4 are lost doing uncreative incremental innovation.
IMHO the LLM project is *done* in the same way the SQL project was done in 2000.
The interesting part of the next 10+ years is figuring out what we… twitter.com/i/web/status/1…
— Scott Stevenson (@scottastevenson)
7:02 PM • Apr 7, 2023
@yoheinakajima@OpenAI@pinecone@LangChainAI@bio_bootloader Also, @yoheinakajima is there any chance you would consider putting your project on Github?
— Greg Mushen (@gregmushen)
4:42 AM • Mar 29, 2023
Eye Candy
Guys, this trailer was made entirely by your Twitter replies using text-to-video with Runway #Gen2
Twitter replies. The entire video was made from Twitter replies...Think about how crazy that is.
— Nick St. Pierre (@nickfloats)
1:12 AM • Apr 7, 2023
The new /describe feature in @midjourney is REALLY good. The first image is the original image that I uploaded the other images are the suggested prompts it generated (that are BETTER than the original)
— Matt Wolfe (@mreflow)
9:10 PM • Apr 3, 2023
A lot of 🦋🦋✨ @Mr_AllenT prompt #midjourney5#midjourneyart
— shisele23q (@shisele23q)
3:53 PM • Apr 7, 2023
People as action figures in Midjourney V5:
— 卄卂ㄥ丂 ㄖᗪㄚ丂丂乇ㄚ (@HALSodyssey)
1:22 AM • Apr 8, 2023
This is at least a 3 day VFX shot but it only took me like a minute thanks to @WonderDynamics
Is it perfect? Not yet… BUT I PRESSED ONE BUTTON AND SOMEHOW MADE ALL THIS.
I’m so impressed! 🧵— Wren (@SirWrender)
6:28 PM • Apr 4, 2023
Do you have 30 seconds for a quick survey to help us improve Everday AI?
We'd love your feedback! Click here.
Do you like what you're reading? Share it with a friend.
Keep reading
Everyday AI - "YOU WIN!" Microsoft Prevails
Two Big Announcements. One Clear Winner.
The Simulation Ramps Up. Elon Can't Resist.
Stanford races toward Westworld. Autonomous agent innovation accelerates. Amazon & Elon enter the fray.
AI Frontlines
The Intersection of Artificial Intelligence and Modern Warfare