- AI Logs
- Posts
- Tech giants form coalitions, US-China tensions and the chip industry, Meta AI multimodels stopped in EU
Tech giants form coalitions, US-China tensions and the chip industry, Meta AI multimodels stopped in EU
Plus: Brainstorming AI workers could take over workplaces in 2025
|
Has it just been a week since our last newsletter? Trump gets shot and then Biden drops out of the presidential race? It’s been an interesting couple of weeks, for sure.
As the political landscape shifts dramatically, AI regulations are emerging as a significant battleground.
With Kamala Harris at the helm of the Democratic party and political scores and polls tilting in favor of Trump, there has been clamor around a new AI executive order that the Trump administration (if elected) will use to replace Biden’s. Meanwhile, Harris is expected to uphold Biden's AI executive order, focusing on human rights and algorithmic fairness.
Promoted by a group that includes Trump allies, the new AI plan that would launch a series of “Manhattan Projects” will create industry-led agencies to evaluate AI and protect against foreign threats, aiming to make America a leader in AI, reported the Washington Post. In contrast to Biden's safety-focused AI order, the effort involves the America First Policy Institute and is gaining support from some Silicon Valley investors.
Tech giants form coalition, won’t rely on government intervention
Meanwhile, leading tech giants in the US have united to form the Coalition for Secure AI (CoSAI) to address concerns around AI safety.
The coalition includes major players such as Google, OpenAI, Microsoft, Amazon, Nvidia, and Intel, as well as IBM, PayPal, Cisco, and Anthropic.
Operating under the nonprofit OASIS, CoSAI aims to establish best practices, tackle AI challenges, and enhance the security of AI applications.
The initiative responds to issues like potential data leaks and automated discrimination, reflecting a strong industry commitment to advancing AI security standards more swiftly than relying solely on government measures. And speaking of government measures…
…US-China tensions and AI demand strain chip industry
Meanwhile, whoever the next president of the United States will have a China AI task at hand. Geopolitical tensions between the US and China are impacting the chip sector as AI demand soars.
Former President Trump criticized Taiwan for monopolizing US chip business and using America for defense. Meanwhile, the Biden administration is considering stricter trade restrictions to prevent China from acquiring advanced chipmaking tools. The US is also trying to persuade allies to restrict their semiconductor technology exports to China.
Meta suspends AI multi-model launch in EU and Brazil
Meta will not offer its AI multimodels in the EU due to the region’s (and the world’s most) strict data protection law–General Data Protection Regulation (GDPR).
This comes after Apple's decision to withhold new features in Europe. Meta will also suspend its generative AI tools in Brazil after a privacy dispute. Apple objects to the Digital Markets Act, which demands greater product interoperability and may compromise user privacy.
Whereas, Meta’s issue is with GDPR’s lack of clarity on data use for AI training. Despite these regulatory challenges, both companies are unlikely to abandon the European market as generative AI evolves.
Meta’s most powerful AI model yet
Back in April, Meta announced it was developing an open-source AI model to rival the best private models like GPT-4 and Claude 3.5. And yesterday, the company released Llama 3.1, its largest open-source AI model yet.
Mark Zuckerberg predicts that the model will become the most widely used by year's end, hinting that it will overtake ChatGPT as the most used assistant.
Meta is expanding its AI assistant and adding a feature to generate images based on specific likenesses. Llama 3.1, with 405 billion parameters, was trained using over 16,000 Nvidia H100 GPUs, likely costing hundreds of millions of dollars.
More on this in the ‘AI image of the week’ section below.
Olympics and AI
That’s a crossover we thought we’d never see. But it’s 2024, and AI is everywhere, so why the heck not? Google is getting ready to make a splash at the 2024 Olympics. As the official AI sponsor for Team USA, Google’s presence will be hard to miss when the games kick off on July 26th, they said in a blog.
Thanks to their partnership with NBCUniversal, you’ll see Google’s AI-powered features in action throughout the broadcast. Expect to see Google Maps’ 3D views of venues like Versailles, Stade Roland Garros, and the Aquatics Centre.
Announcers and commentators will weave Google Search AI Overviews into the broadcast and use the insights from it to answer questions about the Olympics and Paralympics during the segments.
AI companies under fire for using YouTube to train models
AI companies, including major names like Anthropic, Nvidia, Apple, and Salesforce, have used subtitles from thousands of YouTube videos to train their models, despite YouTube’s prohibition against such unauthorized use.
This includes transcripts from educational channels and popular media shows. Many creators, like David Pakman and Dave Wiskus, are upset about their work being used without consent, arguing it’s a form of theft and exploitation.
The dataset, known as YouTube Subtitles, is part of a larger compilation called The Pile, which has also included data from various other sources. Despite the backlash and ongoing legal challenges, AI companies continue to rely on such datasets for training their models.
YouTube now lets users ‘Hum to Search’ using AI
Hunting for tunes on YouTube Music just got a whole lot easier. The platform has rolled out a quirky new AI feature that lets you find songs by humming or singing a snippet.
Dubbed ‘sound search’ or ‘hum to search,’it is as simple as tapping the search bar, hitting the waveform button, and letting your inner diva shine. You can hum a few bars or beatbox your heart out, and watch the app work its magic. Say goodbye to the frustration of having a tune stuck in your head for hours!
Tinder’s AI will automatically pick the best photos for your dating profile
Tinder has officially launched its “Photo Selector” feature, which is now available to all US users and will be available in international markets later this summer.
By analyzing a selfie’s facial geometry, the AI curates 10 optimized photos from the user’s camera roll, focusing on elements like lighting and composition.
Ending with some OpenAI news…
Two months after launching GPT-4o, OpenAI has introduced GPT-4o Mini, a lighter, cheaper model aimed at developers. Costing just 15 cents per million input tokens and 60 cents per million output tokens, GPT-4o Mini is significantly cheaper than previous models and outperforms GPT-3.5.
OpenAI promises it will broaden AI application possibilities due to its affordability, and currently only supports text and vision in the API, “with support for text, image, video and audio inputs and outputs coming in the future,” said the company in a blog.
IE+ SUPPORT INTERESTING ENGINEERING
Invest In Science And Engineering
Enjoy exclusive access to the forefront of AI content, highlighting trends and news that shape the future. Join a community passionate about AI, delve into the latest AI breakthroughs, and be informed with our AI-focused weekly premium newsletters. With IE+, AI reporting goes beyond the ordinary - and it is Ad-Free.
NEWS
NeuralGCM: Google leverages AI for fast, accurate weather forecasts
Google claims Neural GCM generates 2–15 day weather forecasts that are more accurate than traditional atmospheric models.Automated Armageddon: Can AI avoid becoming the next military blunder?
AI shows much promise for military planners around the world, but could misplaced trust in it be a ticking time bomb?Tokyo Drift 2.0: AI masters art of drifting two GR Supra in tandem
Researchers have developed a foundation for creating safer and more responsive vehicles by mastering the art of autonomous drifting.56-gram drone gets ant-inspired AI eyes to navigate autonomously
It combines visual homing (using visual cues for orientation) with odometry (measuring distance in a specific direction) for navigation.
MUST-READ
Capgemini, a global leader in technology services, announced to CNBC that large-scale businesses are expected to implement “multiple-agent AI systems” as early as next year.
In a research report just released, “Harnessing the Value of Generative AI,” Capgemini quantified that 96% of organizations worldwide have Generative AI currently on their boardroom agendas.
Only 39% are taking a “wait-and-watch” approach, which demonstrates the reality of its potential to enhance productivity and innovation and disrupt industries.
AI PICTURE OF THE WEEK
After releasing Llama 3.1, Meta released statistics of how the model is performing on par with leading models such as GPT-4, GPT-4o, and Claude 3.5 Sonnet across various tasks.
The company said that Llama 3.1 is better balanced in terms of helpfulness and harmlessness than its predecessor.
Here’s a comparison of the performance of fine-tuned Llama 3 models on key benchmark evaluations. The table compares the performance of the 8B, 70B, and 405B versions of Llama 3 with that of competing models.
Source: Llama 3.1 Paper
In an open letter, Meta CEO Mark Zuckerberg compared Llama to the open-source Linux, noting that even though the operating system attracted criticism early in the day, it is now widely used in cloud computing and as the core of Android, hoping that the same would be true someday for Llama 3.1 in LLMs.
AI TOOL OF THE WEEK
Google Food Mood is a fun and interactive tool that uses AI to match your mood with delicious food options. You can combine two cuisines into a single recipe. Nevertheless, it's a great way to discover new cuisines or find the perfect meal based on how you're feeling.
Currently, Google Food Mood is primarily accessible through the Google Arts & Culture platform. You can find it here, and the following steps detail how to use it.
1. Access the tool
Go to the link mentioned above and click on ‘Launch Experiment.’
2. Select your mood
You'll be presented with a variety of mood options. Choose the one that best reflects your current state of mind. You can select whether you want a main course, a starter, a soup or a dessert. You can tell for how many people you’d be cooking. And of course, the cuisines you’d like to mix it up with.
For example: I want a starter for 2 people + mixing influences from Qatar and Slovakia.
3. Dietary restrictions
There is also the ‘diet options’ button, which lets you choose whether you want vegan, vegetarian, or gluten-free options.
4. Explore food matches
Once you've selected your mood, Google Food Mood will generate a list of food options that are typically associated with that mood.
5. The results
Google will use AI to not only give you a full recipe to create your desired dish but will also generate an AI picture of how your food will look once you finish making it.
Source: Google Food Mood
Tips for a Better Experience
Be honest about your mood. The more accurate your mood selection, the better the food recommendations.
Experiment with different combinations. Don't be afraid to mix and match cuisines. You might discover a new favorite!
Use it as inspiration. Food Mood can be a great starting point for meal planning or ordering takeout.
Additional Reads
🚨 The Blueprint: IE's daily engineering, science & tech bulletin.
⚙️ Mechanical:Explore the wonders of mechanical engineering.
🛩️ Aerospace: The latest on propulsion, satellites, aeronautics, and more.
🧑🔧 Engineer Pros: For expert advice on engineering careers intelligence.
🎬 IE Originals:Weekly round-up of our best science, tech & engineering videos.
🟩 Sustainability: Uncover green innovations and the latest trends shaping a sustainable future for the tech industry.
⚡Electrical: From AI to smart grids, our newsletter energizes you on emerging tech.
🎓 IE Academy: Master your field and take your career to the next level with IE Academy
Want to share your feedback? contact@interestingengineering.com