- The AI Synthesizer
- Posts
- Skyward Synapses: AI Insights #3
Skyward Synapses: AI Insights #3
Midjourney just got better at generating hands!
Luke Skyward
May 15, 2023
Hello, AI Enthusiasts, and Welcome to this week's Skyward Synapses: AI Insights!
In today’s email, we will talk about:
Google trying to chase OpenAI with their updated LLM.
Midjourney 5.1 is significantly better at generating hands!
Do you consider buying ChatGPT Plus? Check out poe.com instead!
Google MusicLM is great at generating music!
Let’s get started!
🎬 Video of the week
This week Google presented their Keynote presentation, where they talked about their recent upgrades to their AI Chat system - Bard, and introduced their updated LLM, called PaLM 2.
The whole Keynote is 2 hours long, but the part regarding LLMs lasts around 10 minutes. You can check it out below:
I’m very excited to see some competition in the space of LLM systems. OpenAI definitely has the upper hand for now, but it would be great to see Google try to bridge the gap!
I summarized the new capabilities of Bard in a Medium article.
💡 Idea of the week
The idea of the week is music generation with AI!
Google gives now access to an early version of MusicLM - their latest music generation model.
MusicLM is poised to take center stage in the realm of music generation. Building on Google’s earlier research, AudioLM, this model synthesizes high-fidelity audio from text prompts. From modifying existing audio based on text prompts to generating diverse music compositions, MusicLM is set to transform how we create and experience music.
In this Twitter thread, you can see what the model is capable of:
I'm testing MusicLM, the music generator from Google.
Gonna start with some of their sample prompts and then try some of my own.
— Pete (@nonmayorpete)
6:14 PM • May 11, 2023
🧩 Prompt of the week
This week I played with Midjourney 5.1. I wrote a summary of my findings in this thread:
We laughed that AI generates creepy hands.
It's not the case anymore!
I did a deep dive into the new version of Midjourney.
Here's what I found!
— Luke Skyward (@Olearningcurve)
10:02 AM • May 14, 2023
The most significant thing I found out is that the new version of Midjourney is significantly better at generating hands! It still sometimes generates anomalies but it happens less frequently. Look at these two results I was able to get:
A cinematic photography of romantic couple holding their hands, artistic image, realistic, romantic, hands close up --ar 3:2 --v 5.1 --style raw
A cinematic photography of romantic couple holding their hands, sitting at a cafe in Paris, artistic image, realistic, romantic --ar 3:2 --v 5.1 --style raw
Two people learning how to play guitar, each holding their own guitars, artistic image, whole body shot, realistic --ar 3:2 --v 5.1 --style raw
In my opinion, the results are mindblowing!
📚 Resources
poe.com might be an interesting alternative to ChatGPT Plus. At the same price level, it has access Claude-instant-100k - a model with a HUGE context window (100k tokens). In comparison, GPT-4 has a context of only 8k tokens!
Context of 100k tokens will let you summarize whole books and articles (it is equal to around 100 pages of text).
You can check out more details in this post:
Everyone talks about GPT-4/ChatGPT, but Anthropic's Claude model is my favorite now.
Just noticed their new 100k context window version of Claude instant is on poe . com.
(btw this tweet is not sponsored, it's a genuine attempt to get you onto the best tools out there.)
This… twitter.com/i/web/status/1…
— Rob Lennon 🗯 | AI Whisperer (@thatroblennon)
1:59 PM • May 14, 2023
To infinity and beyond,
Luke Skyward