• WeeklyDispatch.AI
  • Posts
  • The week in AI: US State Department issues warning on EU's AI Act

The week in AI: US State Department issues warning on EU's AI Act

Plus: Introducing LLaVA, the open source alternative to GPT-V

Presented by

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

Last week, we highlighted a major generative AI release in DALL-E 3 and a big upgrade to Canva’s platform. Not to be outdone, Adobe has unveiled major upgrades to its suite of AI-powered creative tools through Firefly v2. Firefly is a text-to-image generator, and the upgraded version with better outputs is now available in Photoshop, Illustrator, and as a standalone web app.

Adobe also introduced the first generative AI model for vector graphics in Illustrator, allowing users to create editable vectors from text prompts. The new ‘Firefly Design’ model in Adobe Express can generate social media templates and designs for users as well.

The US State Department has issued a warning to the EU about some of the shortcomings of the sweeping EU AI Act. Brussels has gained a reputation for creating excessive red tape in the technology sector, and the US is concerned that under the proposed Act innovation could be stifled even further. The EU is already lagging behind major AI players like the US, China, India and the UK in terms of AI investment, development, adoption and infrastructure.

The final version of the EU AI Act is still being agreed upon, but is expected to go into effect by 2025.

The demand for AI chips is so much greater than current supply levels that both Microsoft and OpenAI have joined the growing list of tech giants looking into (or in some cases already) developing their own. While Nvidia’s graphics processing units are currently the dominant AI chip, GPU’s are not particularly efficient at handling AI processes; subsequently, a slew of xPU’s are being developed by tech companies to increase the efficiency and availability of chips to power artificial intelligence:

To name just a few. And developing the chips is only a part of the equation - almost all of the advanced AI chips are manufactured by just a few companies: TSMC (Taiwan), Samsung (South Korea), and GlobalFoundries (US) account for 80-90% of total advanced chip production. Currently, a chip packaging shortage is the main supply bottleneck for the most advanced AI chips.

Despite numerous upgrades and updates, text-to-image service Midjourney is still struggling with racial and cultural bias. In experiments by postdoctoral researcher Arsenii Alenichev, Midjourney consistently depicted white people as doctors and Black people as patients, even when prompted with text specifying the opposite. It also produced absurd images reinforcing stereotypes about Africa.

This isn’t a new issue - CNN had a feature on bias/stereotypes in AI image generators (including Midjourney) well over a year ago. The fact that this offensive issue hasn’t been resolved yet is beyond the pale. Midjourney is currently the most popular text-to-image service, with millions of users worldwide and extensive resources. It’s well past time for these services to take responsibility for their image outputs rather than passing them off as neutral computer-generated content.

The massive surprise attack launched by Hamas militants in Gaza on Israel over the weekend represents a major intelligence failure, as Israel's sophisticated surveillance and AI systems failed to provide advanced warning. Questions abound regarding Hamas’ concealed preparations and why warning signs were missed - and Israel had just been touting the effectiveness of their advanced systems to NATO’s military chair less than two weeks ago. It’s a sobering reality check for a country whose Defense Ministry has touted itself as an emerging AI superpower.

More in AI this week:

Harness the Power of AI

AI’s been here. Now learn where it’s going. The AI Boot Camp from Columbia University serves as an introduction for those just entering tech, and builds on that foundation with specialized AI skills. 

This learning experience is perfect for those looking to upskill and set themselves apart. Upon completion, you will be prepared to lead AI conversations — and initiatives — to bring about key results for organizations. 

No previous programming skills are required to apply.

Why choose the AI Boot Camp?

  • Lead the way in all things AI: Learn how to leverage AI and machine learning to automate, solve problems, and drive results. No previous programming experience required.

  • Showcase AI skills to employers: Build a portfolio through challenges and team-based projects. Gain access to a network of 250+ employers

  • Explore funding options: Find payment plans and other financial resources to meet your needs.

Trending AI Tools & Services:

  • Adobe Firefly v2: Adobe’s upgraded text-to-image service

  • ElevenLabs: text-to-speech service, now with AI dubbing and translation

  • Pix: get personalized picks for TV shows, movies, books and podcasts

  • Relume: access the world’s largest library of Figma & Webflow components and build better websites in hours, not days

  • Cube from CSM: turn any input into game-engine ready 3D assets

  • LLM Boxing: see how two of the top open source LLM’s stack up against each other in the ring

  • Hotshot-XL: an AI text-to-GIF model

  • Hyperficient: science-backed workout/gym AI

  • Summify: summarize videos, podcasts, talk shows, interviews, documentaries, and more with a click of a button.

Guides/useful/lists:

Social media/videos/podcasts:

  • (Discussion) OpenAI's justification for why training data is fair use, not infringement [Y Combinator]

  • Joe Rogan podcast with OpenAI CEO Sam Altman [Podcast]

  • ChatGPT-4V Multimodal decodes a Redacted government document on a UFO sighting released by NSA [X]

  • (Discussion) AI could finally let us talk with animals [Reddit]

  • ChatGPT makes a game for $1 [YouTube]

  • "Godfather of AI" Geoffrey Hinton: The 60 Minutes Interview [YouTube]

  • GPT-4 Vision access in ChatGPT - full tour & impressive results! [YouTube]

Open source & technical:

That’s it for AI this week. We’ll see you next Thursday!