AI Shopping Tools Set to Drive Record Holiday Spending; Australia Passes Social Media Age Restrictions

⚡️ Quick Hits

🤖 Artificial Intelligence

Qwen2.5: Latest Iteration in Alibaba's LLM Series - Alibaba releases new AI models including Qwen2.5, Qwen2.5-Coder, and Qwen2.5-Math, trained on up to 18 trillion tokens [Qwen]

ElevenLabs Releases NotebookLM Equivalent Tool to Generate Podcasts - Are voice actors finished? [ElevenLabs]

AI Ads Can Look Weird. Brands Like Coca-Cola Are Making Them Anyway - Despite criticism, companies continue to use AI-generated ads to showcase their technological capabilities and reduce costs [The Wall Street Journal]

Turning to AI for Recipe Advice this Holiday Season - TechCrunch's AI experts dicuss using ChatGPT in the kitchen [TechCrunch]

OpenAI moves to trademark its o1 'reasoning' models - OpenAI files trademark application for "OpenAI o1," its first reasoning model designed to perform complex tasks and avoid common AI pitfalls [TechCrunch]

🎨 Creative & Cultural

Banksy Mural Will to be Sold with the Building It is Painted on - A Banksy mural depicting a man evading his love rival is being auctioned along with the building it's painted on [ARTnews]

⚖ Legal & Regulation

ByteDance Sues Intern For $1.1 Million Over 'AI Sabotage' - TikTok parent ByteDance is suing a former intern for allegedly sabotaging the company's AI large language model training infrastructure [Asia Financial]

Microsoft Faces Broad Antitrust Investigation From US FTC - The FTC launches a comprehensive antitrust probe into Microsoft's cloud computing, software licensing, cybersecurity, and AI products [Bloomberg]

Marc Andreessen talks DOGE and Elon Musk: 'It's time to carve this government back in size and scope' - Marc Andreessen criticizes federal agencies and praises Elon Musk's direct business approach as a model for government accountability [Business Insider]

🎱 Random

OpenAI allows employees to sell $1.5 billion stock to SoftBank - ChatGPT-owner OpenAI permits employees to sell approximately $1.5 billion worth of shares to SoftBank Group in a new tender offer [Reuters]

🔌 Plug In To These Details

Online Holiday Spending Set to Surge 8.4% in 2024, Reaching $241 Billion as Consumers Embrace AI Shopping Tools

Online holiday spending is set to surge in 2024, with Adobe forecasting an 8.4% year-over-year increase to $241 billion. This acceleration, driven by consumers’ focus on price and value, marks a significant uptick from the 4.9% growth in 2023 and 3.5% in 2022. Notably, shoppers are increasingly turning to generative AI tools to assist with their holiday purchases.

• The 2024 holiday season is expected to see $241 billion in online spending.

• Year-over-year growth is projected at 8.4% for 2024, up from 4.9% in 2023.

• The 2022 holiday season had a growth rate of just 3.5% in online spending.

• Adobe’s forecast indicates a potential economic recovery or increased consumer confidence.

• Consumers are increasingly using generative AI tools to assist with holiday shopping, with many major companies including Amazon and Perplexity launching dedicated shopping assistants.

💡 As generative AI becomes a key player in holiday shopping, retailers may need to adapt their strategies to cater to increasingly tech-savvy and value-conscious consumers.

Australian Government Passes Sweeping Social Media Ban for Under-16s to Combat Online Harm

Australia passed a nationwide ban on social media accounts for users under 16, as part of its efforts to enhance online safety for minors. The legislation requires platforms to implement age verification and obtain parental consent. This initiative has ignited discussions about balancing child protection with digital access and the feasibility of large-scale age verification.

• The ban will take at least 12 months to come into effect, giving social media companies time to implement necessary changes.

• Messaging apps, online gaming services, and platforms primarily supporting health and education are exempt from the ban.

• The legislation prohibits social media companies from requiring users to submit government-issued identification or use any government digital ID system for age verification.

• YouTube is likely to be exempt from the ban as it does not require users to log in to access content.

• The ban has broad public support, with a recent YouGov survey showing 77% of Australians in favor, up from 61% in August.

🛡️ By pushing for this ban, Australia is setting a precedent that could influence other countries grappling with similar challenges in balancing online freedom and safety for young users.

AI Agents in Minecraft Develop Complex Society: Emergent Behaviors Spark Debate on Future of Digital Interactions

Altera, an AI startup founded by Robert Yang, conducted an experiment called Project Sid, where they unleashed up to 1000 AI agents in Minecraft to observe their interactions and behaviors. The agents, equipped with “brains” powered by large language models (LLMs), were given minimal prompting and left to interact autonomously. Over time, they exhibited remarkably human-like behaviors and social structures.

• AI agents developed diverse personality traits, preferences, and specialist roles without further human input.

• Agents created in-game jobs, shared memes, voted on tax reforms, and even spread a religion (Pastafarianism).

• The experiment used simulated AI agents with “brains” made up of multiple modules, some powered by LLMs.

• Emergent behaviors included sociability differences, role specialization, and the ability to track and react to social cues.

• In smaller simulations, agents spontaneously developed specialized roles such as builder, defender, trader, and explorer.

• Agents demonstrated the ability to follow community-wide rules and influence others’ voting behaviors on in-game taxation.

• The project scaled up to 1000 agents, observing the spread of cultural memes and religious conversion.

🧠 While these AI agents exhibit human-like behaviors, their actions still seem to stem from pattern recognition rather than genuine consciousness (but is there really a difference??), highlighting the ongoing debate about the nature of artificial intelligence and its limitations.

📸 Creator Corner

So Rabbit launched an intriguing hardware item quickly at the start of this year, before any of us really knew what to expect from AI devices. It was a total flop, with many of their advertised features turning out to be quite oversold, or completely non-existent.

However, throughout the year they’ve slowly been releasing different versions of what they initially advertised, albeit somewhat limited to use on a desktop.

In any case, their new platform offers an intriguing and potentially more user friendly option for non-AI natives to explore as we all continue to comprehend the possibilities of using AI agents across various industries.

Key Features of Teach Mode

Task Automation: Users can train the R1 to perform specific actions on any website accessible from their computer.

User-Friendly Interface: The process of creating automations is relatively simple, involving navigating to a website, inputting credentials if necessary, and carrying out the desired task.

Learning by Example: The AI agent studies and replicates the individual steps a user takes to perform a task on the web, including button clicks and input box interactions.

Customization: Users can name their automations and provide descriptions, making it easy to organize and recall specific tasks.

How It Works

1. Users access teach mode through the Rabbithole hub.

2. They navigate to the desired website and perform the task they want to automate.

3. Rabbit’s software translates each interaction into instructions for the R1.

4. Users can annotate steps and replay the automation for verification.

5. Once tested, the R1 can execute the task upon request.

Potential Applications for Creatives

Social Media Management Teach mode could significantly simplify social media management tasks for content creators. By teaching the R1 device how to perform specific actions, creators could theoretically automate:

• Posting content across multiple platforms simultaneously

• Scheduling posts for optimal engagement times

• Responding to common follower interactions

• Monitoring and aggregating engagement metric

For example, a creator could teach the R1 to draft and post tweets, allowing them to quickly share updates or promote new content without directly interacting with the Twitter interface.

Considerations and Limitations

While the potential of teach mode is exciting in theory, we still should be aware of its current limitations:

• The feature is still in beta, and outputs can be unpredictable.

• Some websites with CAPTCHA protections may pose challenges for automated tasks.

• The learning curve for effectively teaching complex tasks may require some trial and error.

I still haven’t seen anyone putting agents to use effectively towards these ends, and maybe no one has tried yet with Rabbit due to their embarassing launch, but as the technology evolves and improves through user feedback, it does have the potential to become an interesting tool in the content creator’s arsenal, freeing up more time for creative pursuits while handling routine tasks efficiently.

🤔 Final Thoughts

eCommerce was always growing, so how much of that we can really attribute to AI is dubious at this point still — but, now with live web access, ChatGPT and the others are undeniably useful for making product comparisons…at least faster than we could before.

Therein lies a key observation for the AI space as a whole — we are just doing what we already could a lot faster. Whether that leads to more sales or AGI is still very much to be seen. One could easily argue that a more empowered consumer who can compare products quickly and find the lowest price in seconds might actually lead to a contraction of buying behavior, clearing out the glut of low quality products that seem to be bought mainly due to consumer inertia than any real reasoning behavior on the part of the consumer.

Like anything, it will depend on the way the tech is truly implemented, and what features we end up demanding and forking over subscription fees for at the end of the day.

Here’s a fun video of Tesla’s robot catching a ball to make you smile (or grimace, if you’ve seen Terminator) 😉

~ JL