OpenAI introduces human-like AI
We've got a problem on our hands when it comes to taming those smart AIs and preventing them from going rogue.
Hey there! This is AI Kitchen - the secret ingredient to your AI recipes.
Here’s what we’re serving up today:
Wearable AI projector is coming this year
Midjourney launches Panning
AI translated 5,000-year-old language instantly
Google made sneaky changes in the privacy policy
OpenAI introduces Superaligment, the human-like AI
Let's dive in.
Wearable AI projector is coming this year
Ex-Apple designers' AI startup, Humane, introduces 'Ai Pin' mini projector aiming to revolutionize smartphones.
Highlights:
The device can project an interactive interface on nearby surfaces such as the user's hand or a table.
Equipped with a camera and computer vision software, it can identify objects like food labels.
'Ai Pin' is designed to attach to a breast pocket and utilizes voice commands for email summaries, language translation, and calendar reminders.
We wrote about this device in April. Will it truly replace smartphones? Time will tell.
Midjourney launches Panning
Midjourney, the AI image generator, introduces the new "Panning" feature for extended image creation.
Highlights:
Clicking arrow buttons ⬆️⬇️⬅️➡️ extends the image in that direction.
If you type "/settings" and choose "Remix mode" you can customize your prompt.
Multiple panning can be performed on the same image.
Another remarkable update from Midjourney
AI translated 5,000-yeay-old language instantly
Scientists translated Cuneiform, one of the earliest writing systems in human history, into English using AI.
Highlights:
Data scientists employ NLP techniques to translate Akkadian cuneiform tablets into English.
The Epic of Gilgamesh, one of the oldest surviving literary works, was written in Akkadian using cuneiform script.
The researchers created The Babylonian Engine, a website showcasing their technology with initial translations of Akkadian.
The AI training model for the Babylonian Engine, called Akkademia, is open-source and accessible on the project's GitHub repository.
Why it matters: AI-enabled translations of Cuneiform writings offer valuable insights into our past, unraveling mysteries yet to be discovered.
Google made sneaky changes in the privacy policy
In a recent update, Google revised its privacy policy to allow the collection of publicly shared online information. Google wants to scrape all of the web to train its AI models.
Highlights:
Google is now authorized to utilize publicly shared online data, extending beyond direct user-provided information.
The focus has shifted from improving "language" models to training their AI models and enhancing products like Google Translate, Bard, and Cloud API.
Privacy concerns arise due to the analysis of individuals' online posts using AI systems.
How to protect your data?
Use alternative services prioritizing user privacy such as DuckDuckGo for search, ProtonMail for email, Vimeo for video sharing, and Brave for web browsing. Here are more tools about it.
Utilize incognito or private browsing mode.
OpenAI introduces Superaligment, the human-like AI
We've got a problem on our hands when it comes to taming those smart AIs and preventing them from going rogue
Our current bag of tricks, like using reinforcement learning from human feedback, heavily relies on us humans keeping a watchful eye on AI. But here's the catch: if AI surpasses our intelligence by leaps and bounds, we might find ourselves in a pickle.
So, OpenAI has dedicated to allocate 20% of its computing resources to tackle the alignment problem concerning superintelligent AI within the next four years.
Highlights:
OpenAI is forming a new Superalignment team that aim to create a super smart human-like AI model.
The team will be led by renowned AI researchers Ilya Sutskever and Jan Leike.
This superior AI can evaluate other AI models on a massive scale, tackling tasks that are mind-bogglingly difficult for us humans to assess.
But here is the trick. Let's face it, we don't fully comprehend the inner workings of our current AI models. So how in the world do we build a new one that we can understand and trust to keep all those other "misaligned" AIs in check?
However, OpenAI has set its sights on a few key research priorities for this "superior" model:
Scalable oversight: Making sure those other AI models stick to safety guidelines even in situations where we humans can't directly supervise them.
Automated interpretability: Imagine being able to decipher what the AI is up to and why it's doing what it's doing. That's like reading the AI's mind!
Adversarial testing: This means intentionally putting misaligned models to the test and automatically detecting the worst kind of misalignments.
So, there you have it, the grand plan to solve superintelligence alignment problem.
Top-5 AI tools this week
Alpha by Public - Your investing co-pilot with GPT-4 (link)
PR-agent - Get PR analysis and suggestions inside your GitHub pull request (link)
PMToolsAI - AI co-pilot for product managers (link)
Lemon AI – Open-source Zapier NLA to empower agents (link)
Fliki - AI video creation 10x simpler & faster (link)
Learn
Shopify’s CEO builds a simple webchat interface for the llama.cpp repo (link)
Making a web app generator with open ML models (link)
How to build an ecom chatbot with Botpress and Flowise (link).
Weekly meme
“Influential people as the homeless” post became viral this week (source).
Weekly AI image
Another great series: "ChatGPT, describe a look of the new national superhero. Midjourney, visualize it. Reddit, choose their name and superpowers".
Weekly Thread
Weekly Midjourney Prompt
Painting of lovely teddy bear 3D detail close tromp l'oeil showing the texture of thick oil paint strokes on the rustic canvas, vibrant colors, style Beksiński , basquiat style sharp focus::5 vivid colors --ar 11:16 --s 250
Do you build an AI tool?
If you're using AI in your work or projects, I wanna hear all about it. Hit reply to this email and give me the deets. Who knows, maybe we'll feature your work in our AI newsletter. Keep building, and keep making AI awesome.
That's a wrap for today.
See you next week! If you want more tasty AI treats, be sure to follow our AI Chef on Twitter (link).
If you love this episode and want to support us, spread the word about us by sharing The AI Kitchen with friends. We really appreciate it!
Thank you for reading!
What'd you think of today's edition?
Help me to understand what you think about this episode. Just reply with a number (1, 2, or 3) to this email.
1 - Damn good
2 - Meh, do better
3 - You didn't cook it
Your compadre,
Anton "AI Chef" Cherkasov