Lightnews — Scholar-powered news

Dylski @2ne1.com · 2d

Experience the future of art. Don't miss "Evolution and Foundation."

🗓️ Public Viewing: Oct 17–26, 2025, 11am-6pm.
📍 Oxo Gallery, @oxotowerwharf, London.
Sponsored by @GoogleDeepMind.

#DigitalArt #LondonExhibition #ArtandAI #GoogleGemini #HumanAICollaboration #AIArt

An intricate AI-generated abstract image on a light gray background, depicting complex, symmetrical biological or molecular structures. The structures are composed of numerous small, irregularly shaped forms and strands in colors like deep purple, olive green, earthy brown, light teal, and golden yellow. The central part of the image appears to split, revealing internal complexity. White text is overlaid at the bottom: "Evolution + Foundation" in large font, followed by "William Latham Stephen Todd Dylan Banarse," "Google DeepMind," and "AI Art" in smaller text.

1

Dylski @2ne1.com · 2d

Artist William Latham on the results: "We were fascinated to see what this new approach of ‘Targeted Evolution’ would produce and the outcomes have far surpassed our expectations."

It’s a glimpse into a new frontier of co-creation.
#HumanAICollaboration #AIArt

A 4x4 grid displaying 16 different 3D digital forms from the "Evolution and Foundation" exhibition, generated by an AI in response to the "prompt: chameleon". Each form is a unique, abstract creature with organic textures, combining brown, bumpy shell-like parts with green, vine-like structures and blue accents. Some have long, curly tails. The grid showcases the diverse outcomes of the evolutionary art process, all attempting to interpret the same creative goal.

1 1

Dylski @2ne1.com · 2d

This is a true human-AI collaboration, bringing together artist William Latham, mathematician Stephen Todd, and @GoogleDeepMind Research Engineer Dylan Banarse.

The work is curated by Prof Frederic Fol Leymarie of @GoldsmithsUoL.
#AIart #GenerativeArt

A highly detailed, hand-drawn evolutionary tree titled 'The Empire of Form' by William Latham (1985–86). The vast black-and-white line drawing showcases the FormSynth process, mapping the evolution of forms from simple geometric primitives at the top, through hundreds of generations, into a complex and diverse array of surreal, organic, and mechanical-looking structures at the bottom. This artwork visualizes the rule-based, manual creation process that was a precursor to the computer-generated evolutionary art featured in the 'Evolution and Foundation' exhibition.

1 2

Dylski @2ne1.com · 2d

Inside the gallery, you won't just see the final art. You'll explore giant "evolutionary trees" showing the work's journey over hundreds of generations.

You can even read the AI's "thoughts" on its creative choices and listen to an AI-generated podcast about the project's genesis.

An evolutionary tree diagram from the "Evolution and Foundation" exhibition, visualizing the generative process of evolving a 3D digital form, labeled 'a diatom', over 200 generations. The complex branching tree begins with simple shapes at the top and diversifies into a wide array of colorful, intricate forms, including spiky starbursts and textured spheres. A prominent red line traces the primary lineage from an early ancestor to the final, large and complex diatom form at the bottom.

1 1

Dylski @2ne1.com · 2d

More than AI art, this is AI as the artist.

Our 'Foundational Evolution' process marries 35yrs of genetic grammar from artist William Latham & mathematician Stephen Todd with Gemini. The AI interprets abstract visuals like an inkblot test, selecting which forms 'survive' & evolve.

A two-part diagram from the "Evolution and Foundation" exhibition illustrating a genetic algorithm for generative art.
The top flowchart shows an AI-driven process: A "Tournament objective prompt" directs the Gemini AI, which evaluates pairs of images from the "Mutator" system. Gemini selects "Winners," whose "Genes" are mutated to create new forms in a continuous loop.
The bottom diagram visualizes a "binary tournament" in action: a "Current generation" of abstract forms is randomly paired, Gemini selects winners (green checkmark) and rejects losers (red X), and a "New generation" is formed from the winners and their mutated offspring.

1 1

Dylski @2ne1.com · 2d

What happens when a pioneering artist hands creative control to an AI?

Announcing "Evolution and Foundation," a new London exhibition where Google's Gemini acts as the artist, directing the evolution of complex 3D art.

🗓️ Oct 17-26, 2025
📍 Oxo Gallery, London (@oxotowerwharf)

1 3 14

Dylski @2ne1.com · 10d

Announcement: Our "Evolution and Foundation" AI Art exhibition which opens in mid October in London. Sponsored by Google DeepMind. Details to follow.

#AIart #GeminiAI #HumanAICollaboration #GenerativeArt #AIcreativity #GoogleDeepMind #WilliamLatham #goldsmithsuniversity

1

Dylski @2ne1.com · Jun 17

Why does the concept of 'red' not feel like the experience of red? Our paper argues this disconnect exists because our deep-seated need to be understood will always exceed the limits of the language we invent to achieve it.
arxiv.org/pdf/2506.12086

#Consciousness #Neuroscience #Philosophy #AI

arxiv.org

3 6

Dylski @2ne1.com · May 28

Image Breeder - Evolve Imagen prompts using Gemini.
Start with a prompt, then mutate, crossbreed and/or guide with further prompting.
Works with free-tier but free credits run out pretty fast!
aistudio.google.com/app/prompts?...
AI Studio is pretty amazing!

1 2

Dylski @2ne1.com · May 11

Rediscovering my Softpop SP2, NTS-1 and Minirig 4 portable system 🙂

#bastlinstruments #minirigs #korguk

Dylski @2ne1.com · Apr 24

Version with digital glitching added to camcorder feedback. #glitchart #videosynthesis #modularsynth

Dylski @2ne1.com · Apr 23

What can you do with one?

1 1

Dylski @2ne1.com · Apr 21

Video feedback with modular synth.

#glitchart #videosynthesis #modularsynth

1 4

Dylski @2ne1.com · Apr 11

Could these principles inform more engaging, collaborative, and perhaps even understandable AI? By giving agents a drive not just to learn about us, but to achieve mutual awareness with us? Explore the possibilities in our paper! 👇 #FutureofAI #AgentAI #GoogleDeepMind

2

Dylski @2ne1.com · Apr 11

We propose this intrinsic motivation could have evolved via the Baldwin effect, potentially facilitating social coordination during the Paleolithic and even laying the groundwork for communication and culture. #Influence #Impressionability #Reciprocity #SocialCognition

1 2

Dylski @2ne1.com · Apr 11

Such agents motivated by reciprocal understanding strongly prefer interacting with each other over inanimate objects or predictable 'shadows'. This intrinsic reward even fosters cooperation where only one agent receives an external reward. #Cooperation #Altruism #GameTheory

1 1

Dylski @2ne1.com · Apr 11

We model the "Wanting to Be Understood" drive in RL agents via intrinsic rewards for mutual awareness. Unlike a simpler 'curiosity' reward (a drive to understand), this involves reciprocity – influencing AND being influenced by another agent. #ReinforcementLearning #SocialAI

1 1

Dylski @2ne1.com · Apr 11

We used simulations with RL agents in the "Perceptual Crossing Paradigm" (PCP) – virtual avatars interacting on a circle. We tested various internal reward functions, including artificial curiosity and drives for imitation, influence, and anticipation. #ReinforcementLearning

1

Dylski @2ne1.com · Apr 11

“Wanting to be Understood” - Could a deep human need to be understood be the crucial evolutionary 'gadget' bootstrapping cooperation, culture, and language? We explore this idea using AI simulations in our new paper: arxiv.org/abs/2504.06611 🧠
#Evolution #Cognition #AI #GoogleDeepMind

1 3 15

Dylski @2ne1.com · Apr 11

Such agents motivated by reciprocal understanding strongly prefer interacting with each other over inanimate objects or predictable 'shadows'. This intrinsic reward even fosters cooperation where only one agent receives an external reward. #Cooperation #Altruism #GameTheory

1

Dylski @2ne1.com · Apr 11

We model the "Wanting to Be Understood" drive in RL agents via intrinsic rewards for mutual awareness. Unlike a simpler 'curiosity' reward (a drive to understand), this involves reciprocity – influencing AND being influenced by another agent. #ReinforcementLearning #SocialAI

1

Dylski @2ne1.com · Apr 11

We used simulations with RL agents in the "Perceptual Crossing Paradigm" (PCP) – virtual avatars interacting on a circle. We tested various internal reward functions, including artificial curiosity and drives for imitation, influence, and anticipation. #ReinforcementLearning

1

Dylski @2ne1.com · Apr 9

#Promptbreeder uses direct and indirect ways to prompt LLMs to create and refine diverse task-prompts using thinking styles and mutation-prompts.

We use the #LLM to generate new mutation-prompts to improve the way Promptbreeder improves. Self-referential and self-improving!

Dylski @2ne1.com · Apr 9

Here #Promptbreeder is evolving task-prompts to solve a task where it has to state whether a given word has the same meaning in two different sentences.

Prompts accumulate subtle but effective mutations over time leading to the highest performing prompts.
#PromptEngineering #LLM

1

Dylski @2ne1.com · Apr 9

The LLM proposes task-prompts, refines them and most excitingly proposes new ways of refining them in a evolutionary loop, achieving state-of-the-art prompt strategies on commonly used arithmetic and common sense reasoning benchmarks.
#PromptEngineering #LLM #AI #Promptbreeder

1