The GIANT AI Report
The fast-paced AI landscape may seem overwhelming, as we are constantly inundated with new developments and breaking news. The GIANT AI Report is here to break it down into digestible bite-sized pieces that are currently relevant to parenting and education.
In this edition, we look at the latest versions of Stability AI, Claude, and Mistral–as well as ChatGPT’s latest feature that is available in 37 languages! Safety and regulation progresses. What’s up with SORA? Will it spawn a new generation of young filmmakers and content creators? How our educators foresee the use of AI in the classroom and age-appropriate applications of AI round out this report.
Chat GPT Can Now Read Responses Back to You Out Loud
OpenAI has introduced a new "Read Aloud" feature for ChatGPT, allowing users to hear its responses spoken out loud. This feature is available on OpenAI's Android and iOS apps and is also rolling out on the web. It supports ChatGPT 3.5 and ChatGPT 4, offering five different voice options and the ability to speak in 37 languages. While voice capabilities were previously available for ChatGPT, this new feature enables the bot to generate a written response first and then read it back to the user, enhancing its functionality.
People Are Already Saying Stable Diffusion 3 Trounces Midjourney and DALL-E
Stable Diffusion 3, the latest advancement in AI image generation, has introduced a new diffusion transformer architecture and flow matching, indicating significant progress in various aspects. This upgrade addresses previous issues with text recognition, enhancing the model's ability to understand complex text prompts and improving image quality. The comparison with other leading AI generators, such as DALL-E 3 and Midjourney V6, suggests that Stable Diffusion 3 is a major contender in the field, offering enhanced capabilities for realizing specific creative visions. The continued development and competition among these AI generators reflect a fast-paced AI arms race.
The AI Wars Heat Up With Claude 3, Claimed to Have “Near-Human” Abilities
Anthropic recently unveiled Claude 3, a trio of AI language models designed similarly to ChatGPT. These models reportedly set new industry standards in cognitive tasks and are available for use through Anthropic's website, with the most advanced model requiring a subscription. Claude 3, including models like Haiku, Sonnet, and Opus, boasts advanced performance across various tasks, with the Opus model approaching "near-human" levels of comprehension and fluency in complex tasks. Anthropic claims that Opus outperforms GPT-4 on several benchmarks, including undergraduate-level knowledge and coding tasks. The company also emphasizes the models' improved speed, cost-effectiveness, and enhanced vision capabilities compared to previous iterations and other models like GPT-4 Turbo and Google's Gemini.
Mistral AI releases new model to rival GPT-4 and its own chat assistant
Paris-based AI startup Mistral AI is introducing Mistral Large, a large language model designed to compete with top-tier models like GPT-4 and Claude 2 in reasoning capabilities. The company is also launching Le Chat, an alternative to ChatGPT, currently available in beta. Mistral AI, known for raising significant funding, offers Mistral Large through a paid API with usage-based pricing, undercutting GPT-4 Turbo's pricing. Mistral AI's partnership with Microsoft will provide Mistral models to Azure customers, expanding its reach and potentially opening up new collaboration opportunities.
Duolingo stock soars as online learning surge and AI boost forecast
Shares of language learning platform Duolingo surged over 16% as the company forecasted better-than-expected revenue for 2024, driven by increased online learning and AI integration. The company expects to add $1.68 billion to its market value with this forecast. Duolingo's success is attributed to its "freemium" model, with a new subscription tier called Duolingo Max, which includes generative artificial intelligence (GenAI) features, driving demand and higher prices for its offerings. The company reported record total bookings and paid subscribers in the fourth quarter, with impressive growth in daily and monthly active users.
Reddit in AI Content Licensing Deal with Google
Reddit has reportedly signed a $60 million per year deal with Google, allowing the social media platform's content to be used for training the search giant's AI models. This move comes as Reddit prepares for its high-profile stock market launch, aiming to generate new revenue streams amidst stiff competition from platforms like TikTok and Facebook. The deal marks Reddit's first reported agreement with a major AI company, highlighting the growing trend of AI model makers seeking diverse training data sources beyond internet scrapes, despite potential copyright issues. Reddit and Google declined to comment on the deal, which is expected to be detailed in Reddit's upcoming initial public offering filing, set to be made public as early as Thursday.
How Many News Websites block AI Crawlers
Globally, top news websites block AI crawlers, which are used by companies like OpenAI and Google to scrape data for training large language models (LLMs). By the end of 2023, 48% of the most widely used news websites across the ten countries were blocking OpenAI's crawlers, while 24% were blocking Google's AI crawler. Almost every website that blocked Google's AI crawler also blocked OpenAI's crawlers. The article highlights significant differences in blocking AI crawlers by country, with outlets in the Global North more likely to block than those in the Global South. Additionally, legacy print publications were more likely to block AI crawlers compared to broadcasters and digital-born outlets. The findings suggest that newer models are less likely to be trained on news output from certain publishers, which could impact the quality and relevance of AI outputs related to news.
Google Deepmind Forms a New Organization Focused on AI Safety
In response to concerns over Gemini’s potentially deceptive content and prompting issues, Google's AI R&D division, DeepMind, announced the formation of a new organization, AI Safety and Alignment, focusing on AI safety. The organization will include a team dedicated to safety around artificial general intelligence. Despite Google's efforts to address AI safety, skepticism and concerns about the misuse of AI, particularly regarding deepfakes and misinformation, remain high among the public and enterprises. DeepMind's lead, Anca Dragan, emphasizes the importance of addressing present-day concerns while preparing for future AI risks, acknowledging the challenges ahead in ensuring AI safety.
US Justice Dept names first AI officer as new technology challenges law enforcement
The U.S. Justice Department has appointed Jonathan Mayer, a professor at Princeton University, as its first chief science and technology adviser and chief AI officer. Mayer will advise Attorney General Merrick Garland on integrating AI into the department's investigations and prosecutions while addressing concerns about its potential dangers, such as amplifying biases and enabling cybercrime. The department has already used AI for various purposes, including tracing the source of illegal drugs and analyzing evidence related to the Jan. 6 Capitol attack. Mayer will lead a new board of law enforcement and civil rights officials to advise on AI ethics and efficacy and will work to recruit more technological experts to the department.
Tyler Perry Puts $800M Studio Expansion on Hold Because of OpenAI’s Sora
Tyler Perry, the actor and producer known for his Madea films, expressed concern over AI technology like Sora, which can create high-definition videos from text prompts, potentially eliminating the need for physical sets or location shoots. This advancement could significantly reduce film production costs but also jeopardize jobs across various sectors of the entertainment industry. Check out some of the films already created by Sora here.
How Young is Too Young to Teach Students About AI
The survey data from the EdWeek Research Center indicates that the vast majority of educators in the United States believe that students should learn about artificial intelligence at some point in their K-12 education. Nearly 9 in 10 educators feel that AI should be taught before students graduate from high school, although opinions vary on when exactly this should begin. While 65 percent of educators think AI should be introduced in middle or high school, 26 percent of elementary teachers believe students should start learning about AI in grades 3-5. This enthusiasm among elementary teachers may stem from their role in laying foundational skills. However, some teachers feel overwhelmed by the idea of adding AI education to their workload, with 78 percent stating they lack the time and bandwidth due to existing academic challenges, social-emotional learning priorities, and safety issues.
What Is Age-Appropriate Use of AI? 4 Developmental Stages to Know About
According to the teachers and child development experts in this Education Week article, children can be introduced to AI in distinct phases. Upper-elementary students can focus on developing problem-solving skills. While AI can provide answers easily, over-reliance on it can hinder the development of critical thinking. Teachers should model proper use of AI tools and guide students in refining prompts and recognizing AI biases. Thirteen is the age when a person can sign up for a chat GPT account with their parents’ permission. While middle school students may be curious about more adult topics, they still lack impulse control. Educators should provide guardrails and limits on AI use, and capitalize on students' critical and abstract thinking skills. High school students can take a closer look at the limitations of artificial intelligence. Teachers should prompt students to exercise skepticism and critical thinking when using AI-powered tools, helping them understand the potential consequences of misuse.
Will AI Use in Schools Increase Next Year? 56 Percent of Educators Say Yes
A majority of educators anticipate an increase in the use of artificial intelligence tools in their schools or districts over the next year, with 56% expecting AI use to rise. Most foresee a slight increase in AI use, while 6% anticipate a significant increase. Only 1% of respondents expect a decrease in AI use. Districts are increasingly recognizing the importance of generative AI and are focusing on helping teachers and students use tools like ChatGPT effectively rather than considering bans. Some districts are exploring ways AI can save educators time, such as automating tasks like responding to emails or writing speeches.
What Happens When an AI tutor Struggles With Basic Math
In this podcast, Julie Jargon, the Wall Street Journal Family and Tech columnist, explains that while parental controls within social media apps aim to protect children from harmful content and limit their screen time, they can be difficult for parents to navigate. She also highlights the recent congressional hearings and lawsuits against social media companies, which have prompted companies like Meta to implement additional protections for teens. Matt Barnum, the WSJ education reporter, discusses Khan Academy's Khanmigo bot, designed to assist students with schoolwork. While the bot aims to help students understand concepts rather than simply provide answers, it faces challenges in accurately solving math problems. Parents are advised to consult with teachers about the bot's use and accuracy when their children interact with such technology.
Read more
Thank you for riding the AI roller coaster with us! While the news is ever-evolving, The GIANT Room will help you to stay in the (sometimes sideways and upside-down) loop! We are working on ways to safely teach and promote AI literacy. Our Creative Bootcamps apply AI in a fun and frictionless process where kids collaborate with AI and learn age-appropriate prompt engineering while co-designing GIANT Remixes. Have you checked out our GIANT Remixes?