AI application

Deepmind’s new AI game should be a game-changer in the real world too

Summary Deepmind’s AI system “DeepNash” masters the complex board game Stratego. For the Deepmind research team, DeepNash is a potential springboard for AI capable of mastering complex everyday situations. The turn-based conquest game Stratego is extremely complex, far more complex than the Chess and Go board games already mastered by Deepmind, as it requires significantly …

Deepmind’s new AI game should be a game-changer in the real world too Read More »

Meta shows generative AI model with database access

Summary Generative AI models like DALL-E 2 or Stable Diffusion store their knowledge in their parameters. Meta shows a model that can access an external database instead. Current generative AI models like DALL-E 2 generate impressive images, but store their knowledge in model parameters. Improvements require ever larger models and ever more training data. Researchers …

Meta shows generative AI model with database access Read More »

Stable Diffusion API Now Available, Stability AI Partners with AWS

Newsletter Thanks to the official API, Stable Diffusion can now be integrated more easily into existing and new applications. This could lead to the creation of an application ecosystem around Stable Diffusion. Stability AI, the startup behind Stable Diffusion, released an early version of the REST API as a developer preview. The API gives developers …

Stable Diffusion API Now Available, Stability AI Partners with AWS Read More »

Hidden capabilities of major language models: is emergence the norm?

Summary Emerging capabilities in large language models are a good argument for further scaling. Is emergence the norm? When large language models evolve, AI researchers observe unpredictable and sudden jumps in performance on certain tasks. These emergent abilities can be defined as abilities that occur in models above a certain size and do not occur …

Hidden capabilities of major language models: is emergence the norm? Read More »

ChatGPT is a GPT-3 chatbot from OpenAI that you can test now

Summary OpenAI launches a chatbot for the first time in the test phase. It’s called ChatGPT and is intended to help OpenAI develop better AI systems through user feedback. ChatGPT is a new dialog-optimized AI model from OpenAI. As the latest language model for GPT-3, ChatGPT was trained with human feedback. So-called Reinforcement Learning from …

ChatGPT is a GPT-3 chatbot from OpenAI that you can test now Read More »

RenderDiffusion generates a 3D scene from a single 2D image

Summary The leap from 2D to 3D poses challenges to existing delivery methods. RenderDiffusion looks promising because it can render a 3D scene from a single 2D image. For 2D images, diffusion methods have made great progress in recent months. Gradually, researchers on this path are also seeing success for 3D objects. Google, for example, …

RenderDiffusion generates a 3D scene from a single 2D image Read More »

OpenAI’s latest GPT-3 model generates better and longer texts

Summary OpenAI releases a new language model for GPT-3 trained with human feedback. It brings many improvements, according to OpenAI. The new “text-davinci-003” GPT-3 model is based on the InstructGPT models introduced by OpenAI earlier this year, which are optimized with human feedback. These models have already shown that AI models trained with RLHF (Reinforcement …

OpenAI’s latest GPT-3 model generates better and longer texts Read More »

Stableboost generates up to 200 frames at once with Stable Diffusion

Newsletter Even with a perfect prompt, there are virtually an infinite number of ways an AI-generated image can turn out. The Stableboost website shows up to 200 after a single run. DALL-E 2 and Midjourney generate four frames per run, Dreamstudio, StabilityAI’s official web app for stable streaming, even up to nine – but even …

Stableboost generates up to 200 frames at once with Stable Diffusion Read More »