The week in AI: Massive Google layoffs, Chrome's AI updates

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, and highlight the top research in the field as well as exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on in AI.

Our sponsor this week is GrowthSchool - unleashing potential and transforming lives:

Become an AI & ChatGPT genius for free (Holiday Sale) - Join a 3-hour ChatGPT & AI workshop (worth $99) at a 100% discount! 

Register Here: (FREE for the First 100 people) 

NEWS & OPINION

Google: massive layoffs, new AI features for Chrome, and a stunning text-to-video model

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

Google has done their best to dominate the headlines in AI this week, for better or worse. CEO Sundar Pichai sent a memo to employees over the weekend informing them that approximately 12,000 jobs would be cut as the company continues its pivot towards an AI-first organization. Over 1,000 jobs had been cut this year even before the announcement, including sales team members and FitBit (acquired in 2022) employees. Google’s South Korea workers are not taking the layoff news well.

Additionally, Chrome has received 3 new generative AI features: a tab organizer, a theme customizer, and a browser-embedded ‘help me write’ tool for when you need a quick AI-assist. You’ll be able to try out these new features in Chrome on Macs and Windows PCs over the next few days, starting in the U.S.

Finally, the company showcased Lumiere, a state of the art image and text-to-video diffusion model. Unlike previous SotA models such as Runway’s Gen-2, Lumiere uses a new architecture that generates the entire video in one go, rather than creating distant keyframes first and then filling in the gaps. By generating the whole video at once, Lumiere can more effectively maintain temporal consistency – making the motion in the video look much more natural and coherent than any AI-generated video model before it!

Meta’s Zuckerberg wants to open source AGI - sort of

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

In a sweeping interview with The Verge, Meta CEO Mark Zuckerberg outlined his views on the future of AI while working to achieve artificial general intelligence (AGI) - the current holy grail of the industry. While a debate has emerged over whether Meta’s advanced models like Llama-2 are truly open source, they are at least free to download, modify, and deploy - a stark contrast from the likes other cutting-edge AI systems. Here are a few important notes from the interview:

• Meta might have more compute and scaling capacity for AI than any company on the planet. They possess over 340,000 Nvidia H100 GPU’s, the (for now) industry standard for intensive AI workloads; they shipped in 3 times more H100 GPU’s in 2023 than any other company in the world besides Microsoft. They also have a massive stockpile of the A100 and other advanced AI chips.

• Meta is training Llama-3 now, although there is not yet a timeline for release. Zuckerberg admits that Llama-2 was not a state of the art model on release, but with Llama-3 and beyond their ambition and expectations are much higher.

• While making advanced AI openly accessible is a dangerous proposition, Zuckerberg criticized other industry leaders who initially promoted open-source AI but have since become more secretive, suggesting a strategic shift driven by the realization of AI's immense value. For him, the threat of further inequality and lack-of-access driven by closed AI is greater than that of bad actors using powerful systems.

ChatGPT Enterprise goes to university

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

OpenAI has announced its first official partnership with a university. Starting in February, Arizona State University will become a ChatGPT Enterprise customer - the first higher education institution to join that platform. With ChatGPT Enterprise, ASU will gain access to OpenAI’s top models with no usage caps and increased fine-tuning capabilities to build personalized AI tutors for students, and more.

Time flies in AI: around this time just a year ago, it might have been difficult to imagine such a development. Many schools had banned the use of ChatGPT outright, citing an academic integrity crisis with concerns over the potential for students to use the tool for plagiarism or to bypass learning processes.

Amazon research highlights depth of AI-generated trash content online

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

The internet is increasingly becoming inundated with low-quality, AI-generated content. That may not be news to you, but a recent study by the Amazon Web Services AI Lab reveals just how deeply this trend has already infiltrated online content. The yet-to-be-peer-reviewed study found a staggering 57.1 percent of internet sentences have been translated into multiple languages, often poorly - suggesting the widespread use of language models like ChatGPT for both content creation and translation. This development is drastically impacting the effectiveness of Google Search.

The issue is also particularly dire for "lower-resource languages," where less data is available for effective AI training, leading to a proliferation of subpar, AI-translated materials that degrade with each translation cycle. The AWS researchers suggest this is a strategy to generate ad revenue through clickbait, resulting in vast swaths of the web being swamped with AI-generated gibberish. To make matters worse, an abundance of poor-quality data makes training a capable model in these languages that doesn’t incorporate this gibberish into its outputs very difficult.

The study suggests improving language model training by better detection methods and then filtering out low-quality, machine-translated content, especially in less commonly used languages. That could be a tall order under the circumstances.

• New MIT CSAIL study suggests that AI won’t steal as many jobs as expected

• New theory suggests chatbots can understand text

• AI today and trends for an AI future

• TikTok can generate AI songs, but it probably shouldn’t

• Trump shares AI picture of himself praying, with six fingers

• OpenAI CEO Sam Altman is fundraising for a network of AI chips factories because he sees a shortage now and well into the future

• Lawyers are rapidly embracing AI: here’s how to avoid an ethical disaster

• Figure's humanoid robots are about to enter the workforce at BMW