The week in AI: Beyond deepfakes

In partnership with

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 and services; we highlight the top research in the field as well as exciting developments in open source. Even if you aren’t a machine learning engineer, we’ll keep you in touch with the most important developments in AI.

Our subject line was chosen this week based on the first two articles covered below. It is rapidly becoming more difficult than ever to distinguish real from fake, truth from lie - images, audio, video, and even the people you’re interacting with. Stay informed and alert.

NEWS & OPINION

“No one’s ready for this.” No weird-looking hands or other telltale signs of AI - The Pixel 9’s Reimagine tool is about to break our trust in photography

-------------------------

The new Pixel 9 lineup’s marketing has been laser-focused on Google’s latest advancements in AI. Multimodal Gemini integration, effortless multitasking, easier note-taking, more immersive multimedia experiences - the list goes on.

But some of these new AI features are raising eyebrows. The Verge’s deep dive into the Pixel 9’s Reimagine tool reveals a concerning reality: while the tool’s AI-driven capabilities can create beautiful, customized ‘reimagined’ images with ease, they also make it disturbingly simple to generate and share highly realistic - yet entirely fabricated - scenes. From car wrecks to disasters to corpses, the power to manipulate reality is now apparently in the hands of… anyone with a Pixel 9. And such powerful tools will only become more mainstream.

Google has implemented some safeguards and policies intended to prevent misuse. However, the guardrails are weak, easily circumvented by creative prompting alone, and the responsibility largely falls on the users to adhere to ethical standards. And, while Google’s own SynthID watermarking system aims to identify purely synthetic images, it was nowhere to be found for the ‘Reimagined’ photos, and Verge staff were able to strip a single line of metadata and pass off manipulated images as authentic.

“We briefly lived in an era in which the photograph was a shortcut to reality, to knowing things, to having a smoking gun. It was an extraordinarily useful tool for navigating the world around us. We are now leaping headfirst into a future in which reality is simply less knowable.” - Sarah Jeong, The Verge

Blade Runner 2024 - human bots, and the quest for authentic digital experiences

-------------------------

It’s not just fake images you have to worry about. Bots on platforms like Twitter became more sophisticated around 2011-2012, with automated accounts used for things from marketing and customer service to spreading misinformation or manipulating public opinion.

Still, you have to see this recent exchange on X/Twitter unmasking a more sophisticated bot to believe it (click the link and scroll to top). In a highly viewed discussion about using AI in college to cheat, one user suspects another of possibly being a bot and replies to it with a prompt for an LLM - which the bot responds to in proper AI fashion. It was a surprising reveal of an AI agent; as one X user succinctly put it: “No god damn way”.

Is that user truly an AI agent, further flooding social media with detritus? Or is it just a troll with a sense of humor, going to ChatGPT to respond and leaving it to our imagination? Whatever the case, the fact that it’s virtually impossible to discern anymore adds just another layer of imperceptibility.

The article (linked in the headline) goes on to examine the emerging role of “AI Hunters” and where and how they are cropping up - eerily reminiscent of blade runners hunting down replicants.

How Arizona State University uses ChatGPT

-------------------------

Back in January, we covered OpenAI’s first partnership with a university - Arizona State University.

ASU has been quite busy since then. The university is integrating the AI assistant into over 200 projects across the campus. ASU is one of the largest public universities in the US - they serve 181,000 students annually and offer over 800 degree options. ASU has been named the most innovative university in America by US News and World Report for 9 straight years.

The university is using ‘ChatGPT Edu’, a version of ChatGPT built for universities to ‘responsibly’ deploy AI to students, faculty, researchers, and campus operations powered by GPT-4o. The university also launched an ‘AI Innovation Challenge’ for faculty and staff, after overwhelming demand within the university to use ChatGPT for teaching, research, and ops.

ASU’s Chief Information Officer remarked, “While we expected that ChatGPT would find a home in engineering and the sciences, we've been surprised by just how diverse adoption has been across almost every department.” ASU is also introducing a number of new AI-focused degrees.

MORE IN AI THIS WEEK

• OpenAI in talks for funding round valuing it above $100 billion

• Google’s custom AI chatbots have arrived

• Nvidia's subdued forecast dampens enthusiasm in AI chip stocks after steady rally

• Perplexity AI plans to start running ads in fourth quarter as AI-assisted search gains popularity

• The California State Assembly approved S.B. 1047, requiring major AI companies to test systems for safety before public release

• Google DeepMind staff call for end to military contracts

• UK scientists to develop AI-powered tool for predicting dementia risk early

• Altman-backed startup to test autonomous vehicle mass transit system in Atlanta

• Mark Zuckerberg and Spotify’s Daniel Ek on why Europe should embrace open-source AI

• Chinese firms continue to find ways to bypass US export controls around AI tech - state-linked entities use cloud services provided by AWS and rivals to access advanced US chips and artificial intelligence capabilities

The Daily Newsletter for Intellectually Curious Readers

• We scour 100+ sources daily

• Read by CEOs, scientists, business owners and more

• 3.5 million subscribers

Sign up today!

TECHNICAL, DEVELOPMENT, RESEARCH & OPEN SOURCE

• Google research on Speculative RAG: enhancing retrieval augmented generation through drafting (2-step RAG)

• Google’s GameNGen: image models generate a completely playable game of Doom

• Cerebras Inference: the fastest AI inference solution in the world (1800 tokens/sec for Llama3.1 8B)

• AI21 releases Jamba 1.5, a state-of-the-art hybrid SSM-transformer (!) architecture 256K context-window model

• Nvidia and global partners launch NIM Agent Blueprints to help enterprises create and deploy custom generative AI applications

• Nvidia and Mistral release Mistral-NeMo-Minitron 8B: with state-of-the-art accuracy that runs efficiently on laptops

• Automating Thought of Search (AutoToS) from IBM and Cornell: completely taking the human out of the loop of solving planning problems with LLMs

• Imagination Technologies (their tech can be found in 13 billion devices worldwide) abandons NPUs in favor of GPUs for the edge/AI era

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