PLEASE NOTE: For Participants of the 18th edition of the Film Spring Open (2023), the course price is: PLN 400.
Over the past two years, there has been an unprecedented acceleration in the development of artificial intelligence systems, mainly generative models capable of processing textual, visual and audio information, which have begun to affect many areas of our lives and communication.
However, the scale of the phenomena we are facing in 2023 has exceeded the fantasies of even the most committed enthusiasts: OpanAI’s introduction of the ChatGPT service, whose market adoption is many times faster than the most dynamic of any product to date in any field, the emergence of generative models within video and 3D, the bringing of generative models within image to full controllability on the one hand and excellent quality on the other, the implementation of AI first in applications for creators, then in search engines and web browsers, then in office tools by Microsoft and Google, and finally the presentation and availability of the GPT-4 model for all these applications, the bimodality of the latter (the ability to recognise and interpret text and images at the same time) and the possibility of linking it to any service – all this has happened in the last few weeks up to the moment of writing these words.
However, even more stunning than the variety of applications of AI is its cognitive capacity: measured by various methods, the intelligence quotient of the GPT-4 system oscillates around 114 points, i.e. above the average human intelligence, but, due to its possession of a vast store of human knowledge and cultural texts, most of the human tests and examinations that GPT-4 can face are solved at the level of the best 10% of humans. In other words, if one assumes that wisdom is, according to a very primitive but common belief, intelligence plus knowledge, then the system possesses an above-average intelligence and general knowledge unavailable to any individual human – in this trivialised sense, GPT-4 is already smarter than any individual human. Other capabilities of the system, such as the ability to create theories of the interlocutor’s mind, the ability to deliberately deceive and the ability to use tools surprise even its designers.
How is creativity possible in this rapidly changing landscape? What can it currently consist of? How do we enter into collaborations with AI systems so that the results of such collaborations are as valuable as possible for creators and audiences? I will seek to answer these and related questions with the course participants.
Programme:
1. The scale and significance of the AI revolution – 3h + 1h discussion
- Where did it all come from and how did we move rapidly from an ‘AI winter’ to an all too hot summer? – Accidental breakthroughs and emergence in artificial intelligence systems.
- Where we are and where we are going: recent discoveries and developments in generative AI models and their societal implications, as well as the predicted directions and expected dynamics of further developments.
- AI ethics, trust and coexistence strategies, AI and copyright law.
2.Prompt engineering and working with text models – 3h + 1h Q&A
General construction and principles of generative models of different kind (no expertise required) – what is the memory of a system, how it is learned, learning phase and inference phase, ways of generating results in different systems, latent space (black box), ideas and meanings in latent space.
- Prompt engineering – what is it and how to develop this skill? General characteristics and examples in the context of textual (GPT-3.5, ChatGPT) and image generative models (MidJourney, Stable Diffusion, Dalle-2)
- Text generation using ChatGPT, OpenAI Playground and complex chained systems.
- Presentation of other specialised models: AIdungeon, Character AI. Comparison of properties of different textual generative systems.
3. Working with visual models – 4h
- Image generation with MidJourney, Stable Diffusion and Adobe FireFly – background information, comparison of working methods, basic applications.
- Advanced image generation methods in Stable Diffusion: inpainting, outpainting, ControlNet, DreamBooth – presentation, advanced working methods and applications.
- From single to moving image: DeForum – animations and animatic.
- First generative video models: Imagen Video, Make-a-Video, RunwayML GEN-1 and GEN-2, Kaiber and others – current stage and pace of development, working methods.
- AI editing: Descript, RunwayML – current stage, pace and developments.
4. Other useful areas of AI development for creators – 3h + 1h discussion
- AI models for working with background sound and music: AudioLM, MusicLM, AIVA and others.
- AI models for working with spoken text: Whisper, ElevenLabs, Bark.
- Other examples of AI applications: 360 image generation, 3D object generation, NeRF.
- Combining models of different types and environments with AI (Hugging Face, Google Colab, nVidia Picasso) 1.5h presentation.
- How to keep your finger on the pulse of AI? Sources of knowledge. Concluding discussion.
We warmly invite you to sing up for the film course online!