Copy: the first AI-powered fashion magazine

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each weekday, we scour the web to aggregate the many stories related to artificial intelligence; we pass along the news, useful resources, tools or services, technical analysis and exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on under the hood in AI.

What is Copy, the first AI-powered fashion magazine, trying to prove?

From Vogue: Copy is a new AI-powered fashion magazine that uses Midjourney to generate ‘highly realistic’ images of models and fashion spreads combined with synthetic commentary from ChatGPT. Creator Carl-Axel Wahlström sees the magazine as an artistic experiment in AI-generated content. “It was love at first sight,” said the creative director. “I saw a lot of red warning flags, but I also felt that I was part of something new and revolutionary.”

More Details:

• Wahlström created the magazine himself as a passion project, without sponsors or advertisers. He sees it as a documentation of ‘where fashion is today’ - based on what AI can currently generate.

• The AI is limited by its training data, which leads to stereotypical and idealized depictions of beauty and fashion. Wahlström acknowledges this but chose not to push boundaries for this first issue.

• The magazine is meant to provide an unsettling view into hyper-perfection and stereotypes in fashion. Wahlström wants to spark discussion through the contrast between the AI-generated glamour and the human-written text.

• Wahlström believes AI could reshape power structures in fashion by removing gatekeepers, though biases shape its output. He hopes AI can point the way forward stylistically.

Takeaways: While AI opens creative possibilities, it also propagates the biases of its training data in dangerous ways - including unrealistic beauty standards. Wahlström knowingly plays with these tensions, crafting a glossy but discomforting world. His conflicting love for the ‘too beautiful’ end result is emblematic. It shows both the impressive beauty and realism AI can achieve and its difficulty imagining beyond human biases and stereotypes that we ourselves often perpetuate. Just as Photoshop warped beauty standards, use of biased/perfectionist AI aesthetics could further entrench damaging stereotypes.

Taking the experiment a step further, Arena privately administered his final exam to ChatGPT alongside his students. The questions in the exam were general enough that attending the class lectures was not a prerequisite for answering them. The results were surprising - we won’t spoil it here.

Disagreements continue over whether AI poses existential risks and how much control over AI is desirable. The outcome of this debate will determine if regulators impose licensing rules and other limits on frontier AI models. The EU's AI Act takes a middle road - setting safety rules for general AI models without licensing. In the US, the open AI coalition will likely resist licensing and push for exemptions. With AI regulation advancing, this divide will be an ongoing source of contention.

• (More on yesterday’s headline) Google’s Duet AI becomes a meeting assistant, doc summarizer and chat companion

• Can we defend against the online anti-science movement?

• GM is using Google’s AI chatbot to handle simple OnStar calls

• USA Today owner pauses AI articles after butchering local sports coverage

• Nvidia’s stock closes at record after Google AI partnership

• Samsung debuts its own 'AI-powered' smart recipe app

Nougat employs transformer neural networks, which have driven recent advances in computer vision and language modeling. In tests on scientific papers, Nougat achieved significantly higher accuracy than previous academic systems like GROBID, especially on mathematical formulas. This research tackles an important challenge - unlocking knowledge trapped in PDFs to make it more accessible. The results in translating scanned images from PDF’s (some of them are over 100 years old) show promise in learning to parse complex document structures end-to-end.

Capella combines structured and unstructured data capabilities into one system, providing more flexibility than traditional databases for modern applications. Capella iQ hopes to take this further by speeding up development significantly. The natural language interface means coding tasks that took hours can now (hopefully) be completed in seconds.

• OpenPipe: convert expensive LLM prompts into fast, cheap fine-tuned models

• More Couchbase in AI: cloud to edge AI with a mobile database platform

• Annual State of AI Report Compute Index