Matthias Bastian

Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.

Matthias Bastian

Jun 6, 2023

AI in practice

Clipdrop presents Reimagine XL, an enhanced image reimagination tool that removes direct pixel sourcing from the original image for "bold and unique" output. Instead of a text prompt, it uses existing images as "inspiration".

The model excels at handling multiple aspect ratios and requires no technical expertise. With advanced image encoder technology, Reimagine XL delivers a variety of outputs for applications ranging from fashion sketches to landscapes. Read our article about the original Reimagine here, and watch the video for some impressions on XL.

[Reimagine XL]

Comment Source: Stability.ai

Matthias Bastian

Jun 6, 2023

AI in practice

The RedPajama project has released RedPajama-INCITE-7B-Instruct, the highest scoring open model on the HELM benchmarks, outperforming other state-of-the-art open models such as LLaMA-7B, Falcon-7B, and MPT-7B. The instruct-tuned model is optimized for a wide range of tasks and excels in few-shot performance.

RedPajama-INCITE-7B-Chat is available in OpenChatKit and is optimized for chat applications, while RedPajama-INCITE-7B-Base targets full reproducibility with ten checkpoints and a completely open training process. The whole RedPajama-INCITE family is clean for commercial use.

[Hugging Face - Chat Demo]

Comment Source: Together.xyz

Matthias Bastian

Jun 6, 2023

AI research

ReWOO: New method makes augmented language models more efficient

Language models with access to tools, so-called extended language models, potentially have many more capabilities than native language models. The ReWoo method could make them much more efficient.

Matthias Bastian

Jun 6, 2023

AI in practice

Apple doesn't talk about AI, but puts it in everything anyway

Apple doesn't talk about AI, but packs it in everywhere anyway

Matthias Bastian

Jun 6, 2023

AI in practice

OpenAI is launching a $1M Cybersecurity Grant Program to develop AI-powered defense capabilities, quantify the effectiveness of AI models in cybersecurity, and foster discussions at the intersection of AI and cybersecurity. The program aims to benefit defenders by leveraging coordination toward a common goal of security. Project ideas include automating incident triage, detecting social engineering tactics, and streamlining patch management processes. OpenAI will accept applications for funding or other support on a rolling basis, but will not consider offensive security projects.

[Apply here]

Comment Source: OpenAI