If Something is Free - You are the Product

Good morning.

“If something is free - it means YOU are the product”. 

You have probably heard this before.

The most notorious example of this is Facebook. You don’t have to pay to use the service, but by using Facebook, you are profiling yourself by putting in information , befriending certain people, or liking certain pages. That information can be used to serve you very personalized ads that you are almost guaranteed to engage with.

So why is ChatGPT offered for free?

Knowing that is costs millions upon millions of dollars to operate it ?

Simple.

They are using the questions you are asking it to further train and finetune the model. Essentially OpenAI has crowdsourced the model training to the entire world.

ChatGPT has roughly 180 Million Users. Can you imagine the treasure trove of information you can gather from having that many people asking questions , uploading documents and confiding their secrets.

Why do you think Elon Musk has bought Twitter ? Because Twitter is a goldmine of information. The entire scale of human emotions, good and bad, and information, true or false, is comprised in that enormous dataset.

But back to ChatGPT: can you imagine the amount of confidential data that is being leaked from companies because users are using ChatGPT on the job?

We’re in a weird phase where most companies are not taking a stance on the use of these tools. It’s a real dilemma. If all employees are using it, the company’s productivity would skyrocket. But at what cost?

If nobody is allowed to use it, you’ll get beaten by your competition that is embracing these tools.

Microsoft Copilot is positioning itself as a “safe” AI tool - only trained on your internal data. But its results are disappointing in relation to ChatGPT.

It’s a giant puzzle we’re going to have to solve in the coming years. Leave your thoughts in the comments !

Welcome to the Blacklynx Brief.

AI News
From Brainwaves to Image

• Researchers at Radboud University have developed an AI system that can accurately reconstruct images based on brain activity. Using fMRI scans from humans and electrode recordings from a monkey, the AI learned to focus on specific brain regions, improving image reconstruction accuracy. This breakthrough could lead to significant applications, such as aiding stroke victims and recreating dreams.

• A viral video shows a Phoenix police officer pulling over a Waymo robotaxi driving on the wrong side of the road, only to find no driver in the vehicle. The robotaxi rolled down its window and connected the officer to a Waymo support representative. This incident, caused by confusing construction signs, highlights the challenges and gaps in traffic laws as self-driving technology advances..

• OpenAI experienced a hack in early 2023 that exposed internal discussions about AI designs, but core systems and customer data remained safe. Despite informing employees and the board in April, OpenAI did not report the breach to the public or authorities. In response, OpenAI created a Safety and Security Committee to address future risks, including hiring former NSA head Paul Nakasone.

• The World Artificial Intelligence Conference in Shanghai showcased major advancements by Chinese companies in AI, despite U.S. sanctions on advanced chips. SenseTime introduced SenseNova 5.5, claiming it outperforms GPT-4 in most metrics, and Alibaba’s Tongyi Qianwen models saw a significant increase in downloads. Despite restrictions, China’s AI firms continue to compete at the forefront of AI innovation.

• Google DeepMind researchers introduced JEST (Joint Example Selection Technique), a new method that accelerates AI model training and reduces computing needs. JEST uses two AI models to select the most valuable data, making training up to 13 times faster and 10 times more efficient than current methods. This technique significantly cuts energy consumption and speeds up the release of advanced AI models.

• The OpenAI Startup Fund and Thrive Global have launched Thrive AI Health, an AI-powered health coach designed to drive personal behavior change. It will focus on sleep, nutrition, fitness, stress management, and social connection, offering personalized recommendations based on scientific research, biometric data, and individual preferences. Led by former Google product leader DeCarlos Love, this initiative aims to leverage AI to address chronic disease and healthcare costs, potentially revolutionizing personal health management.

• Odyssey, a new startup founded by self-driving car industry veterans, has unveiled a 'Hollywood-grade' AI video generation platform to transform visual effects and storytelling. The platform includes four specialized AI models focusing on geometry, materials, lighting, and motion, and the team comprises experts from tech giants and blockbuster films. Odyssey, which is building its own 3D data for training, announced a $9M funding round with notable investors. This development could significantly advance AI's role in Hollywood, video games, and other media, despite facing strong competition.

• Japan's Defense Ministry has released its first basic policy on using AI in military applications to address recruitment challenges and keep pace with global defense technology. The policy prioritizes seven areas, including target detection and unmanned systems, and emphasizes human control over AI, excluding fully autonomous lethal weapons. By ruling out autonomous lethal weapons, Japan aims to set a responsible model for AI use in the military, potentially influencing global practices.

• Venture capital firm Andreessen Horowitz (a16z) is acquiring a large number of GPUs, including Nvidia H100s, to support AI startups through an initiative called 'Oxygen'. The firm plans to expand to over 20,000 GPUs and rent them to portfolio companies at below-market rates in exchange for equity. This strategy aims to provide startups with essential GPU resources, giving a16z a competitive edge in attracting AI ventures and reshaping the AI VC landscape.

• Anthropic has upgraded its 'Artifacts' feature, allowing users to publish, share, and remix creations, and introduced new prompt engineering tools in Claude’s developer Console. These updates include advanced testing, side-by-side output comparisons, and prompt generation assistance. The shareable Artifacts enhance collaborative learning and AI-assisted content creation, potentially revolutionizing online education, knowledge sharing, and collaborative work.

• Oura has introduced an AI-powered health coach called ‘Oura Advisor’ within its app, designed to help users interpret and act on data from their wearable rings. The Advisor provides personalized insights and advice on sleep, activity, and recovery, with customizable communication styles and a ‘Memory’ function for personalized health context. This launch comes as Samsung prepares to unveil its competing Galaxy Ring, highlighting the growing competition in AI-driven health technology.

• Microsoft has stepped down from its observer role on OpenAI's Board of Directors, and Apple has canceled plans to join as an observer, amid increasing regulatory scrutiny in the AI sector. Microsoft cited confidence in OpenAI's new direction for its departure, while Apple made no specific comment on its decision. These moves occur as U.S. and EU antitrust authorities investigate partnerships between big tech companies and AI startups.

AI Tip of the Week

There’s more than OpenAI’s ChatGPT. A few weeks ago Anthropic released its latest model Claude 3.5 Sonnet.

On the LMSYS Chatbot Arena Leaderboard you can follow along how models are doing relative to each other.

GTP-4o is still leading the race but closely followed by Claude and then 3 of Google’s Gemini models.

Most people reading this newsletter I suspect are only working with ChatGPT. But Claude has been really surprising me with what it can do.

This week’s tip is quite simple. Fire up the free version of Claude and put in some of the questions you ask ChatGPT and see what Claude comes up with.

Good luck with it !

Quickfire News

• WhatsApp beta revealed an upcoming 'Imagine' AI avatar generator feature, which will allow users to create personalized avatars of themselves in various settings.

• Noland Arbaugh, Neuralink's first human patient, hinted at the potential to control the Tesla Optimus humanoid robot within the next year using the brain implant.

• OpenAI co-founder Andrej Karpathy spoke at UC Berkeley’s AI hackathon, predicting that rather than a single AI, there will be multiple AIs populating a digital space, collaborating on tasks and having their own Slack threads.

• Mantee Robotics showcased a new video of its humanoid robot, Menteebot, navigating new environments by following humans, demonstrating potential for various real-world applications.

• An international research team introduced EMOKINE software, an open-sourced AI tool that can decode emotional expression through motion capture and whole-body movements.

• Midjourney released its ‘Year One‘ book, celebrating one year of Midjourney Magazine and showcasing inspiring prompts, milestones, and profiles of popular community members.

• Apple’s Siri upgrades expected in Spring 2025, not part of Apple Intelligence rollout this Fall.

• Google considered limiting the ‘AI Overviews’ feature to its own app, excluding Safari users.

• Stability AI’s new licensing model: SD-3 free for commercial use up to $1M in annual revenue, quality concerns addressed.

• YouTube's new AI-powered eraser tool: removes copyrighted music while preserving other audio.

• OpenAI updated ChatGPT macOS app: encrypts locally stored conversations after a security flaw discovery.

• Kuaishou’s KLING video model now available as a web app for up to 10-second generations, requires a Chinese phone number.

• EU chief Margrethe Vesteger warned about Nvidia’s AI chip supply dominance, calling it a ‘huge bottleneck’.

• Poe launched ‘Previews’: allows users to generate and interact with web apps within chats, using LLMs like Claude 3.5 Sonnet for coding.

• Luma Labs' chief scientist Jiaming Song: real-time AI video generation is less than a year away, with Dream Machine model showing reasoning and predictive capabilities.

• Magnific AI introduced a Photoshop plugin: allows AI upscaling and enhancing tool use directly in Adobe’s editing platform.

• Nvidia launched a competition: aims to create an open-source code dataset for training LLMs on hardware design to automate future GPU development.

• Taiwan Semiconductor Manufacturing Co.'s valuation briefly surpassed $1T after Morgan Stanley increased its price targets.

• AI startup Hebbia secured $130M in funding: boosts company’s valuation to around $700M for its complex data analysis software.

• Study on ChatGPT’s coding abilities found major limitations using GPT-3.5, but research criticized for not using newer models.

• Elon Musk revealed Grok 2 will likely be ready next month; xAI training on a 100k H100 system, the most powerful training cluster.

• OpenAI blocked API traffic from restricted areas like China, though access remains via Microsoft Azure.

• AI video startup Captions raised $60M in Series C funding led by Index Ventures and Jared Leto, valuing the company at $500M.

• Elon Musk confirmed xAI ended discussions with Oracle on a $10B AI server deal; cluster to be built internally in Memphis, TN.

• Roblox published research on using machine learning to combat voice abuse, significantly decreasing severe reports and violations.

• AI gaming startup Volley announced a $55M series C funding round to develop voice-controlled, AI-powered games for Alexa, TVs, and mobile devices.

• Republican National Committee released the GOP platform for the 2024 election, intending to repeal President Biden’s AI Executive Order and support AI focused on free speech and human flourishing.

• Neuralink hosted a livestream with updates on its brain-computer interface (BCI), discussing telepathic integrations, limb control with Tesla's Optimus humanoid robots, and new patient testing.

• AMD is acquiring Silo AI, a Finnish startup specializing in end-to-end AI solutions, for $665 million to enhance its chip development and compete with Nvidia.

• The Washington Post introduced 'Climate Answers', an AI chatbot that provides answers to user questions based on the outlet's reporting on climate issues.

• Scale is partnering with AWS to promote enterprise and government adoption of generative AI, offering tools for customization and secure deployment.

• Free event: Dan Shipper will reveal his top 10 AI use cases based on hundreds of executive interviews. RSVP available.

• OpenAI CTO Mira Murati stated in an interview that the Sora video generation model still lacks a public release date due to ongoing safety testing and usefulness evaluations.

• OpenAI announced a partnership with Los Alamos National Laboratory to ensure the safe use of AI models like GPT-4o in bioscience labs and assess their potential to enhance research and reduce risks.

Closing Thoughts

That’s it for us this week.

If you want a version of this newsletter with pictures etc. Be sure to subscribe below:

https://www.blacklynxbrief.com/subscribe