The line up this week:

• Hit me with the Latest 👊🏻

• The Imitation Game Part 2

• AI Art just not Human Enough!

• Autonomous Agents, First Look

• Have you heard of @huggingface?

Latest

• Stability.ai announced a new AI to add to their suite of offerings, this time an audio and music model Stable Audio. It’s an exciting development as Stability Ai’s Stable Diffusion has been a popular tool for Art and we’d expect the same here. Looking forward to seeing how it’s received and what creations come of it.

• Google’s unreleased Gemini Model is five times more potent than Chat GPT apparently. According to the article Google have expectedly ramped up their investment in AI since the release of One AI’s ChatGPT 3.5. It also speculated that Google might be cautious about when it releases the Gemini model as it may threaten their current Search business model.

• A closed door Senate forum with a selection of leaders in the industry was held this week in Washington. It featured a variety of guests with varied views albeit the big players who would have a vested interest in what kind of regulation is sort. This is seen as an important step in the US Governments understanding of the issue AI might pose for it’s citizenry. Scepticism and debate all around, but it was good to see an Open Source representative in Hugging Faces’s Clément Delangue. In a show of commendable transparency he posted his opening statement to the forum here.

• Apple announcements from the ‘Wanderlust’ event saw the release of the iPhone 15 and Apple Watch Series 9.

  Love this tweet from Tim Urban

  Also, if you haven’t seen the new Apple commercial that features the CEO Peter Cook and well let’s just say a special guest 🍃🤔 It’s worth a watch just for the cringy giggles 🤭

• Peter Gabriel’s latest video is fully AI generated.

  It was created by Lamson as part of the Peter Gabriel / Stability AI #DiffuseTogether competition that launched in April 2023. You can also see this great video by Junie Lau who used a range of AI tools.

  Looking forward to a re-mastered AI version of the eclectic Sledgehammer film clip from 1986 😁

• This Week in Startups podcast featured Nikesh Arora

  Talking specifically about the state of AI API’s and content providers online. A battle is brewing. Will Chatbots replace all product UI’s and essentially be our online interface?

• Another great episode of TWIST featured the Chris Later. Chris, originally the co-founder of LLVM. He’s basically the guy who invests ways of taking complex systems at the back end and adding a layer of simplification for front end products. His work with Apple and LLVM was instrumental in the development of the iPhone platform. He is now the founder and CEO of Modular.ai. With Modular he hopes to do the same with AI products. He identifies the not widely known difficult problem of deployment for AI systems and plans to solve it!

• HeyGen have released an amazing video AI that does language translation not only for the audio, but seamlessly to the speakers mouth movements as well. Video’s posted this week of people using the tech have been pretty impressive. You can also see this example on the HeyGen Youtube Channel.

The Imitation Game Part 2 :🙈 👀 :

Continuing on from last weeks deep dive into the AI Art Revolution.

Memory and it’s Limit

The Library of Inspiration

There are a few types of memory we could talk about with respect to informing Visual Art.

1. Real World Sensory

2. Internal Reconstructed

3. Reference Image Bank

Real World Sensory memory is collected from our eyes throughout a persons life. A constant recording and stored in the brain.

Internal Reconstructed memory is how the brain stores it. The wetware inevitably can have a filter effect on the memory of an image and also distort, warp and colour ones experience of it. The memory of images may also be reconstructed in dreams which further enhance this effect.

The Reference Image Bank is the memory of the constructed works of others.

For humans, we have this vast repository upon which to inform our art from all of these forms of memory.

For an AI like a Diffusion Model they have their training data. This is purely an Reference Image Bank of other’s creative work.

Therefore, the finite library of images in its training data limits its range.

However, debatably this does not imply that only imitations are in the output’s of Diffusion Models, but simply a reconstructive process or remixing. This kind of process has always existed in the field of Art. It results in derivative work, but may still be considered original as it is not a 1-1 copy.

Interpretation

It is interesting to see how a Diffusion Model interprets what it is prompted. In lies the art of Prompt Engineering.

Prompt Engineering is the skilled process of effectively directing and tweaking the model for a desired result with text. It involves everything from, commands, prescriptive ordering and iterative remixing processes. Each model can have a specific ‘recipe’ upon which to best utilise it’s interpretation of your text.

The Human Input with all it’s stored memory of visual inspiration combined with the Diffusion Models interpretive labelling from text and pattern recognition allow the process of making Art both personal and universal.

Each Model will interpret a prompt differently to one another based on it’s creators engineering.

Thoughts on the Creative Results of AI ART

The Cutting Edge of Art & The Originality Curve

The highest forms of art are those that are unique and original. They are a reflection of the collective conscience, define fashion and inevitably become part of the zeitgeist.

All the ingredients utilised to their greatest potential are integral to achieving this.

Limiting the creation of imagery to the interaction of prompting a Diffusion Model however, is unlikely to achieve it at this stage.

The data set of utilising only a Reference Image Bank in the models training limit its imagination.

Will the next [insert great visionary artist] emerge from this process?

Who knows, like remixing has done for music, a ‘'pop will eat itself’ meta style may result, but I suspect we have already been doing this at scale.

As with all AI it’s about the Human and AI collaboration that yields the best results. It’s when artists train models by integrating work from their own hand, combining many models and constant iteration that will yield the next wave of Artistic Genius!

Derivative Artwork

Given that a large volume of work created commercially or just for fun are of a derivative nature, i.e. in the style of, genre or subculture icons. Artwork from these models will likely permeate in greater and greater numbers throughout the internet.

Next week we will wrap up our look into Human and AI creativity with our final summary of the AI Art Revolution!

AI Art and the Copyright Dilemma

Wired Magazine recently reported on an award winning AI artwork by Matthew Allen that recently failed to gain copyright status for its creator. The Copyright Office has ruled that art created by a Diffusion Model, in the case MidJourney is too machine made and not enough human.

The artwork was made with 624 tweaks and iterations by Allen with some final fixing and upsizing in Photoshop. The Copyright Office even went so far as to agreeing that the Photoshopped work was original, but the other parts generated by the model could not be copyrighted.

Allen is appealing the decision, but given such a precedent it could take some of the lure away from AI usage for artists and creatives. Images created by AI will have to be altered enough by the author to gain ownership over it.

It is clear there will be some stark debate over the coming months and years. Hopefully with more understanding about the process of Diffusion Models and a resolution on the usage of Artist’s work in training these models will see a thriving emersion into Art creation using AI.

Autonomous Agents: First Look

AI Agents are just one way Chat Bots or LLM’s can be used as virtual assistants. They can be assigned tasks and set off to do the necessary research to complete them.

It’s development has a long way to go, but it does offer up great possibilities. Having tried the AI Agent this week as ‘Mighty Jo Detective’ (you can give your agent a name) to do some research for the newsletter, it was a bit like getting a 75 year old to assist you, but with lightning speed. The results were mixed in regards to its effectiveness at the tasks, but I was able to get maybe one or two leads.

Now, disclaimer: there are most likely more enhanced capabilities with the upgraded services since Mighty Jo was only using Chat GPT 3.5 and from first go there is a learning curve. So looking forward to trying them out some more and we’ll keep you posted as to the development of these Autonomous Agents in the future.

Here are a couple of other ‘non technical’ AI Agents to try

GPT Researcher

Agent GPT

Introducing @huggingface 🤗

[That’s not the real logo, it’s a cheap emoji knock-off 🙄]

What did you think about this weeks Newsletter?

Did we nail it? Can we do better? Want to know more? Hit reply and let us know!

If you like our content and think it might be worth a shout out, please do!

Right click the below button and share the link.