?? Welcome to AI Insights Unleashed! ?? - Vol. 38

Embark on a journey into the dynamic world of artificial intelligence where innovation knows no bounds. This newsletter is your passport to cutting-edge AI insights, thought-provoking discussions, and actionable strategies.

?? What's New This Week ??

Google’s ‘Jarvis’ browser assistant is coming

Google is reportedly?developing?an AI agent codenamed ‘Project Jarvis’ that can control web browsers to complete everyday tasks for users, and a possible preview could come as early as December, according to The Information.

• The system captures and interprets browser screenshots to navigate websites and input information.
• Jarvis will initially focus on consumer tasks like online shopping, research, and travel booking.
• The agent is specifically optimized for web browsers (not full computer use) and reportedly currently operates with a few-second delay between actions.

Google’s ‘Jarvis’ news may have been a bit more exciting if the release (still months away) didn’t come immediately on the heels of Anthropic’s live?computer use?feature. The launch, coming alongside a Gemini upgrade, could give Google some momentum heading into the new year — or disappoint after the longer wait.

Meta builds AI Google Search rival

Meta is reportedly?developing?its own AI-powered search engine, hoping to reduce its dependence on Google and Bing to power real-time information in Meta AI conversations.

• Meta is developing proprietary web crawling tech to power its AI’s real-time knowledge of current events and web info without relying on competitors.
• Meta also recently?partnered?with Reuters for news content, suggesting a broader strategy to control its AI information sources.
• The development comes as Meta AI reaches 185M weekly active users across Facebook, Instagram, and WhatsApp.

The AI race is turning to a new battleground —?search. With Meta’s quest to build a self-sufficient AI ecosystem and tech giants increasingly viewing AI as their core business, the race for search independence could spark new competition in how the top models access and deliver real-time info.

OpenAI launches ChatGPT search

OpenAI just?announced?the integration of new web search capabilities directly into ChatGPT, transforming the chatbot into a real-time information hub that puts the platform in direct competition with Google and traditional search engines.

• The new search feature provides instant answers about topics such as news, sports, stocks, and weather, with direct source attribution and links.
• ChatGPT will automatically trigger searches when needed, though users can manually activate web searches using a globe icon.
• The system runs on a specialized version of GPT-4o, fine-tuned for web info — with outputs including sourcing and clickable links to original content.
• Major publishers, including AP, Reuters, Axel Springer, and others, have signed licensing deals with OpenAI to provide content directly to the platform.

After months of rumors and speculation, OpenAI’s search transformation is finally here. The move could signal a major shift in how we access info online, blending AI's conversational abilities with traditional search functionality, which could reshape both the search engine landscape and user habits.

Apple unveils first wave of Apple Intelligence features

Apple just officially?launched?its highly-anticipated AI system, Apple Intelligence, with the latest iOS 18.1 and macOS Sequoia 15.1 updates — introducing a suite of new features focused on productivity, creativity, and privacy enhancements.

• The initial release brings systemwide writing tools for rewriting, proofreading, and summarizing text, as well as enhanced photo search capabilities.
• A redesigned Siri features new typing support, better context understanding, and upgraded product knowledge to answer questions about Apple devices.
• Only newer devices with the M1 / A17 Pro chips or later can access the AI features, with some users also facing a waitlist system after opting in.
• The next update, expected in December, will include more advanced features like ChatGPT integration, Image Playground, and Genmoji.

The long-awaited Apple Intelligence is finally here… sort of. This first wave of integrations is a far cry from the hyped AI phone revolution expected with the new iPhones. While substantial features should arrive in only another month, this initial release feels much more like a soft launch than a revolutionary upgrade.

GitHub and Microsoft open Copilot to rival AI models

Microsoft's GitHub just?announced?that it is expanding its AI coding assistant to include models from Anthropic and Google, alongside a series of new updates and features to the platform —?breaking from the company’s traditional OpenAI exclusivity.

• The platform will allow developers to switch between assistants, including Claude and Gemini, although OpenAI's models remain the default choice.
• GitHub also?introduced?Spark, a new feature that allows users to build applications with natural language prompts.
• The platform announced features including multi-file editing, Copilot code reviews, new agentic updates to Workspaces, and Apple Xcode support.
• GitHub's decision to embrace multiple AI providers comes as its Copilot service reaches a major milestone with over a million paying subscribers.

While some may point to even more smoke from the Microsoft-OpenAI relationship, this move may simply acknowledge that different models excel at different tasks. If GitHub prioritizes giving users the best developer experience, having model choice is a no-brainer —?even if it ruffles exclusivity feathers.

Meta partners with US Government

During Meta's Q3 earnings meeting, CEO Mark Zuckerberg announced that the company would be working with several US government agencies to investigate how it could integrate its Large Language Model (LLM)—Llama—into public sector applications, to support government initiatives and address societal challenges.

• Meta is already working with the US State Department to see how Llama could help address challenges like “expanding access to clean water and reliable electricity in developing states and supporting small businesses.”
• It’s also been talking with the?Department of Education to investigate how Llama could simplify the complex funding process for students?and is having ongoing discussions with several other departments.
• Meta also confirmed these partnerships aren’t financially driven, instead of payment they want to?position Llama as an invaluable tool for government use, making it the go-to platform for federal initiatives.

This comes as several of Meta’s competitors have taken steps to build closer ties with government bodies and work together to bring advanced AI capabilities into the public sector.

Atlas robot shows off autonomous sorting skills

Boston Dynamics just?unveiled?new footage of its electric Atlas humanoid robot sorting automotive parts, showing off new autonomous and real-time adaption capabilities.

• The all-electric Atlas showcases fully autonomous navigation and manipulation, moving engine covers between bins without human input or pre-programming.
• Atlas uses machine learning and an array of sensors to adapt to the environment in real-time, including detecting bin locations and adjusting to failed insertions.
• The demo arrived shortly after Boston Dynamics' partnership?announcement?with Toyota Research Institute earlier this month.

Robotics and auto manufacturing seem to be a perfect match, with deployments from industry leaders in factories nationwide. Elon Musk?just?predicted that there will be 10B humanoids by 2040 —?and with Boston Dynamics, Figure, Optimus, and others, the race for commercial humanoids is accelerating fast.

More than a quarter of new code at Google is generated by AI

Google's third quarter 2024 earnings call highlighted how important AI is to the company. More than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers. The company reported $88.3 billion in revenue for the quarter. While many people feel Google isn't as reliable as it once was, the company continues to operate a very strong business. Google faces a potentially tough road ahead - a breakup is on the table following a ruling that the company is a monopolist in the search and advertising markets.

?? Key Developments ??

OpenAI plans first custom AI chip

OpenAI is?developing?its first custom semiconductor with Broadcom and TSMC while expanding its chip supplier base to include AMD alongside NVIDIA, according to a new report from Reuters.

• OpenAI has reportedly assembled a team of about 20 engineers, including former Google TPU designers, to develop an AI chip targeted for 2026.
• After initially exploring options to build its own chip factories, OpenAI is instead opting to partner with Broadcom for design and TSMC for manufacturing.
• The company also plans to add AMD's new MI300X processors to its training infrastructure, reducing reliance on Nvidia's GPUs.
• The moves come as OpenAI faces mounting compute costs, with reports suggesting the company could lose $5B this year despite $3.7B in revenue.

The AI boom drives unprecedented demand for specialized chips, forcing industry leaders to rethink their hardware strategies. Scaling compute power is the key to unlocking more powerful AI — and the race may be won or lost on hardware strategy rather than just model capabilities.

Recraft introduces a revolutionary AI model that thinks in design language

Design startup Recraft just?announced?its new V3 AI model, which features precise graphic design skills, text generation, and vector capabilities — and also revealed it was the mysterious ‘Red_Panda’ AI that surged to the top of the image generation leaderboards in testing earlier this week.

• The model achieved a 72% win rate and 1172 ELO score on the Artificial Analysis leaderboard, outperforming established players like Midjourney and FLUX.
• Recraft V3 introduces state-of-the-art text generation abilities, allowing designers to create images with accurate text of any size and length.
• The model also shows improved human anatomy realism, positioning and spacing within a scene, and prompt adherence.
• The platform emphasizes designer control with features like custom brand colors, positioning tools, and collaborative workflows.

This Red Panda reveal is a bit of a shocker, with the company ascending to the top tier of image generators seemingly out of nowhere. But with an emphasis on designer control alongside a top-tier model, Recraft may unlock new creative precision that marks the next leap in AI-assisted design.

Osmo gives computers a sense of smell with AI

Osmo just?announced?the successful achievement of ‘scent teleportation,’ showcasing a demo in which an AI analyzes, digitizes, and reproduces the smell of a plum.

• The company?demonstrated?full scent digitization using a plum as a test subject, creating an exact replica of its smell without human intervention.
• The process combines gas chromatography, mass spectrometry, and AI-driven analysis to create a digital ‘scent fingerprint.’
• Osmo's proprietary AI system uses the world's largest scent database to map and recreate molecular compositions.

While photos, videos, and audio have been digital for decades, smell has never joined in on the fun — until now. Osmo's breakthrough could transform everything from entertainment and communication to remote experiences and illustrates that seemingly anything can be digitized with the right combination of AI and engineering.

New AI model predicts early drug development

Biotech startup Iambic Therapeutics just?revealed?Enchant, an AI platform designed to predict how drug candidates perform in human trials before leaving the lab.

• The multimodal AI system combines extensive laboratory data with limited clinical information to predict a drug's potential success early.
• Enchant sets new accuracy marks for predicting human drug interactions, achieving a 74% correlation compared to the previous 58% SOTA score.
• The technology can begin making reliable predictions after studying five drug molecules, requiring minimal human trial data to generate insights.
• Enchant processes multiple types of research data simultaneously, helping bridge the gap between laboratory findings and clinical outcomes.

One of healthcare's biggest challenges is the massive failure rate of drugs during human trials, despite years of promising lab results. By cutting development timelines and identifying issues earlier, Iambic’s AI can reshape how new medicines reach patients — while reducing the billions spent bringing drugs to market.

UMG, Klay Vision partner on ‘ethical’ AI music model

Music industry giant Universal Music Group is?partnering?with AI startup Klay Vision to develop an ethical AI music generator, which comes as the company still actively fights lawsuits filed against Suno, Udio, and Anthropic.

• The partnership aims to create AI music models that ‘lessen the threat to human creators’ and open ‘new avenues for creativity and future monetization.’
• Klay Vision is actively working on a Large Music Model called KLayMM for commercial use that respects copyright and artist likeness rights.
• Klay Vision is led by former Sony Music and Google DeepMind execs, with the partnership following past AI deals with YouTube's AI Incubator and SoundLabs.

The music industry's largest label is taking a ‘control the narrative’ approach to AI, partnering with those willing to play by its rules and aggressively pursuing those who don't. But Hollywood and media firms have accepted that AI isn’t just a passing fad and are now quickly re-positioning for the new era of creation.

Tiny AI model masters humanoid control

Nvidia just?published?new research showcasing HOVER, a small 1.5M parameter neural network that can control whole-body robotic movement effectively across various modes and input methods.

• Despite being thousands of times smaller than typical AI models, the model achieves superior performance compared to specialized controllers.
• Nvidia trained the system in its ‘Isaac simulator,’ which compresses a year of robot training into just 50 minutes on a single GPU.
• The system works seamlessly with diverse input methods, including VR headsets, motion capture, exoskeletons, and joysticks.
• HOVER also transfers directly from simulation to real robots without requiring additional fine-tuning.

While the AI world races toward larger models, HOVER shows that smaller systems can still outperform. The ability to work across multiple control modes with minimal compute requirements could help democratize advanced robotics control — making building sophisticated humanoid systems more accessible and practical.

AI transcripts create dangerous errors

Several researchers?raised concerns?after finding that OpenAI's Whisper transcription tool suffers from frequent hallucinations and invents text that never appears in recordings despite being deployed extensively in healthcare settings.

• A Michigan researcher found fabricated text in 80% of examined transcriptions, while another reported hallucinations in ‘nearly every’ Whisper output.
• Hallucinations ranged from non-existent medical treatments to racial commentary and violent content.
• Over 30,000 medical professionals use Whisper-based tools despite OpenAI's warnings against high-risk applications, according to the AP report.

Hallucinations are currently an unfortunate fact of life in AI systems — and until they are reliably 100% solved, deployment into high-risk environments could be a recipe for disaster. However, commercial pressures seem to be outpacing proper safeguards, with every industry rushing to take advantage of the AI wave.

Open-source AI must reveal its training data, per new OSI definition

The Open Source Initiative (OSI) has released its official definition of 'open' artificial intelligence. For an AI system to be considered truly open source, it must provide the complete code used to build and run the AI, access to details about the data used to train the AI so others can understand and recreate it, and the settings and weights from the training. The definition directly challenges Meta's Llama, which has restrictions on commercial use and does not provide access to training data. Meta claims that it restricts access to its training data due to safety concerns, but critics say the company is just minimizing its legal liability and safeguarding its competitive advantage.

?? Reflections and Insights ??

AI will use a lot of energy. That's good for the climate

AI data centers are driving demand for clean, 24/7 energy, pushing tech giants to invest heavily in renewable and nuclear power solutions. This surge in demand could accelerate the cost reduction of clean energy technologies due to their learning rates. In the long run, AI's energy needs may spur policy changes and advancements in clean energy infrastructure.

Bill Gates has a good feeling about AI

Bill Gates has spent billions on funding technologies he thinks will shape the future. He has a new Netflix series that covers these technologies, including AI, climate change, inequality, misinformation, and global health. 'What's Next? The Future with Bill Gates' is set to launch on September 18. This article contains an interview with Gates where he discusses why he is so optimistic about these technologies.

Why Mark Zuckerberg thinks AR glasses will replace your phone

This is an interview with Mark Zuckerberg where the CEO discusses new developments in VR, AI, and the fast-growing world of consumer smart glasses. It also includes a bit with Verge's deputy editor Alex Heath that talks about the Orion smart glasses demo he experienced at Meta's headquarters, some of the context around Meta's AR efforts, and how Zuckerberg is approaching his reputation as a leader and the public perception of Meta as a whole. A link to the full podcast is available.

Machines of Loving Grace

In this post, Dario Amodei, the CEO of Anthropic, talks about how AI could transform the world for the better. He sketches out what the world might look like with a powerful AI if everything goes right. The effects of powerful AI are likely to be even more unpredictable than past technological changes. The list of positive applications of powerful AI is extremely long - this post focuses on a small number of areas that have the greatest potential to directly improve the quality of human life.

?? Stay Updated: Receive regular updates delivered straight to your inbox, ensuring you're always in the loop with the latest AI developments. Don't miss out on the opportunity to be at the forefront of innovation!

?? Ready to Unleash the Power of AI? Subscribe Now and Let the Insights Begin! ??