- ThePrompt
- Posts
- 10 AI trends by Stanford ๐ฅ
10 AI trends by Stanford ๐ฅ
PLUS: 300 million jobs will be lost to automation
Hi folks!๐๐ป This is The Prompt! Iโm currently in Italy, and although ChatGPT is banned โ the API itโs not. Bing, Perplexity, the OpenAI playground & all other apps are still available.
Like why did they even bother to ban ChatGPT only? ๐ค๐ผ
Governments are always lagging behind when it comes to techโฆ so I guess this shouldnโt come as a surprise.
Anyways, let's see whatโs new๐๐ป
FEATURED
Stanford published their AI Trends report for 2022
Stanford released a massive AI trends report for 2022 yesterday.
The report is 300+ pages long, but weโve extracted the highlights for you:
Industry over Academia: In 2022 there are 32 significant models created by the industry, compared to just 3 by academia. Reason: money and data. Also, most AI PhDs go to work for the industry, and not academia.
Large language models (LLMs) are getting bigger and more expensive. GPT-2 (Released in 2019) had 1.5B parameters and cost an estimated $50K USD to train. PaLM (Released in 2022) had 540b parameters and cost an estimated $8M USD.
AI systems become more flexible. Some language models are capable of navigating multiple tasks (for example, vision, language).
Misuse of AI is rapidly raising. The misuse of AI has increased 36 times since 2012, as people use this tech to post deep fakes and other misleading content.
Employers in the United States are increasingly looking for workers with AI-related skills. Across every sector.
USA leads with the most AI investments. The Focus area is medical and healthcare, data management, processing, cloud, and fintech.
Businesses are embedding AI in many areas. The most common areas are to automate service operations & customer analytics.
AI tools like Github Copilot are making developers 94% more productive.
More students specialize in AI & most AI PhDs are male.
Governments are sprinting to adapt to AI. The US specifically had 35% of all state-level AI bills passed into law.
The report is for 2022.
Now itโs April 2023, and it feels like most of these numbers are already outdated.
PROMPT HUNT
MidJourney: New Describe command to extract prompts
Midjourney released a new command called /describe. You upload an image and get 4 prompts that try to explain the image you uploaded.
Then you can generate images using those prompts.
I played around with it, and it can be useful to find out what kind of prompts work with Midjourney.
Someone already tried to /descibe popular logos and recreate them - the recreated Starbucks logo looks ๐ฅ
WHAT ELSE IS GOING ON
๐ง๐ปโ๐ป Goldman Sachs predicts that 300 million jobs will be lost to automation. They also released an AI report, that focuses on economic growth.
๐ LangChain announces 10M seed round. LangChain is a framework for building apps powered by language models, currently offered as Python and Typescript packages. They revolutionized the creation of AI-powered apps with their tech - bringing data connectors & interactivity to LLMs.
๐๐ปโโ๏ธ AWS launched a generative AI accelerator. Accepted startups will receive up to $300,000 in AWS credits to help build their companies.
RESOURCES
The best resources we came across lately that will help you become better at writing prompts & building AI apps.
๐ Reid Hoffman & Sam Altman discuss learnings from GPT-4 [ Podcast ]
๐๐ป Build your own agents powered by LLMs [ Like Langchain but simpler ]
๐ฅ Build the Stable Diffusion algorithm from scratch [ Technical tutorial ]
๐๐ผ Reversed prompt engineering tutorial [ Twitter Thread ]
TOOLBOX
The latest AI tools to use or get inspiration from.
SiteGPT - ChatGPT for every website
ToolBuilder - No code AI tool building platform
Useful LoremIpsum - AI UX writer for designers and developers
Video2Recipe - Convert cooking video to recipe
e2b - open-sourced AI-powered IDE
YCinterview - YC interview prep
Nimble - Fully autonomous fulfillment
PROMPT OF THE DAY
TOOL
Midjourney
PROMPT
a comic drawing of a 1930 motorbike in front of an old Tokyo shop, ivy plants and flowers, japanese landscape, saturated plain colors, american scene matte painting, matte drawing, detailed, by artgerm and skottie young --v 4 --ar 3:2
RESULT
LATEST PAPERS
MonoAvatar: Learning Personalized High Quality Volumetric Head Avatars from Monocular RGB Videos
LLM-Adapters: An Adapter Family for Parameter-Efficient Fine-Tuning of Large Language Models
Summary of ChatGPT/GPT-4 Research and Perspective Towards the Future of Large Language Models
PODIA-3D: Domain Adaptation of 3D Generative Model Across Large Domain Gap Using Pose-Preserved Text-to-Image Diffusion
Subject-driven Text-to-Image Generation via Apprenticeship Learning
What'd you think of today's edition? |