The Penis Paradox

Good morning.

Wow , what a week again. Meta , Apple and Mistral have released new models that are giving OpenAI’s ChatGPT a run for its money.

But I want to open with something else entirely.

Technology. The Internet. AI.

The last few decades have been nothing but incredible. The pace of evolution has been relentless.

Humanity has been given tools so powerful that we might go down in history as a new race of demi-gods.

Technology has giving us abilities that were unthinkable a few decades ago.

You can go on YouTube right now and witness one of the greatest commencement speeches of all time (and no it is NOT Steve Jobs).

David Foster Wallace is no longer with us - but his words have the potential to transform your life and can be heard time and time again.

You can listen to lectures from esteemed professors at MIT and Harvard, learn astronomy and physics from Carl Sagan himself,you can watch someone fly around the world without leaving your desk, watch concerts, .. you can do ANYTHING.

But there’s a dark side: you can use this technology to fill your days with crap and stupidity. Like convincing yourself the world is flat. Or listening to a remix of ‘They’re taking the hobbits to Isengard” for 10 hours (1,4M views on YouTube)

Although - granted - the YouTube comments section of these types of videos are pretty great.

Or a video with “10 hours of nothing”.

Followed by potentially the greatest YouTube comment in history.

Anyway, I teased you a penis and that is what you’re getting.

My point is this: we have been granted immense power and yet - because we are flawed human beings - we choose to do stupid stuff with our new found powers.

The founders of AI startup company HeHealth found themselves brainstorming an idea for an AI application. We’re already letting AI scan for anomalies on our skin or our teeth, so why not extrapolate ?

So they developed an app where men can take pictures of their penis and upload it to a central database. The AI will then tell you if your penis is diseas- free or kind of suspect.

The FDA was not happy with the app and finally this week the startup declared they were shutting it down.

It’s a great illustration of something that we perhaps all can feel (and this is no longer about the penis). That we’re monkeys playing with fire or things we can’t understand. Limited by our own imagination.

Let’s call it ‘The Penis Paradox’

Something to ponder on this weekend.

Welcome to the Blacklynx Brief

AI News
Battle of the Chatbots

• OpenAI has launched GPT-4o mini, a more affordable and compact version of its GPT-4o model, aimed at making AI more accessible for developers and businesses. Priced significantly lower than GPT-3.5 Turbo, it performs better on benchmarks and is now available to ChatGPT Free, Plus, and Team users. The model supports extensive text and vision inputs, with future plans for audio and video capabilities, marking a significant step forward in AI integration.

• Mistral AI and Nvidia introduced Mistral NeMo, a new open-source language model with 12 billion parameters that outperforms similar models on key benchmarks and features a large 128K token context window. It excels in reasoning, world knowledge, and coding, and can run on standard GPUs, making it highly accessible for businesses. Additionally, Mistral launched 'Tekken', a tokenizer that enhances text representation across 100+ languages.

• Groq has released two new open-source AI models, Llama 3 Groq Tool Use 8B and 70B, which surpass top models like GPT-4 Turbo in function calling benchmarks. The 70B model achieved the highest accuracy on the BFCL Leaderboard, while the 8B model ranked third. These models, trained on synthetic data, are available through Groq's API and Hugging Face, offering advanced tool use and near real-time speeds.

• OpenAI is in talks with chip designers like Broadcom to develop its own AI chip by 2026, aiming to reduce reliance on Nvidia GPUs. They have hired former Google employees and are exploring various chip packaging and memory components, considering outside investment for infrastructure.

• Coatue Management's report on AI humanoids and robotics highlights that significant advancements will be gradual due to physical and cost limitations. Despite a surge in new robotics companies, the report emphasizes that AI and quality training data will drive progress, predicting a shift from warehouse automation to complex tasks like surgery and firefighting.

• Apple has released a new 7 billion parameter open-source AI model, DCLM-7B, which outperforms Mistral 7B and approaches Llama 3 and Gemma in performance. Apple also provided the complete pretraining dataset and training code, making the model and its development process accessible to researchers and developers on Hugging Face.

• Elon Musk's xAI announced the "Memphis Supercluster," the world's most powerful AI training cluster, using 100,000 Nvidia H100 GPUs. Musk also revealed that Grok 2.0 is completed and Grok 3.0, expected to be the world's most powerful AI, will be released in December 2024. Additionally, Tesla plans to produce humanoid robots for internal use next year.

• Google researchers have developed NeuralGCM, an AI-powered weather and climate model that combines machine learning with traditional physics-based methods. It is more accurate for short-term forecasts and 100,000 times more efficient than other models, capable of running quickly on a laptop.

• Researchers from MIT and ETH Zurich have created an AI model to identify different stages of ductal carcinoma in situ (DCIS), a preinvasive breast tumor, using tissue images. The model analyzes cell composition and spatial arrangement, detecting invasive cancer-related cell states in normal tissue, potentially revolutionizing breast cancer diagnostics.

• Meta has released Llama 3.1, including a groundbreaking 405 billion parameter model, the largest open-source AI model to date. This model rivals leading closed AI systems and is expected to be significantly more cost-effective. Zuckerberg sees this as a pivotal moment for open-source AI, democratizing access and enabling startups, enterprises, and governments to develop customized AI solutions.

• In an open letter, Zuckerberg emphasized the societal benefits of open-source AI, highlighting its potential to drive productivity, creativity, and innovation globally. He believes making AI accessible to a broader audience will spur economic growth and equalize technological progress, making advancements safer and more inclusive.

• Zuckerberg also criticized closed ecosystems, using Meta's past experiences with Apple as an example of how closed systems can stifle innovation. He advocates for an open AI ecosystem to prevent monopolies and foster industry-wide progress, aiming to lead a new wave of global innovation by making state-of-the-art AI accessible to all.

• Five U.S. Senators have written to OpenAI CEO Sam Altman, demanding information on AI safety efforts following reports of rushed safety testing for GPT-4 Omni. They request that OpenAI's next model be made available for government testing and ask if OpenAI will commit 20% of its computing resources to AI safety research, as previously promised.

• Adobe has released major AI-powered updates to Illustrator and Photoshop, using its Firefly AI model to enhance creative workflows. New features include Generative Shape Fill and Text to Pattern in Illustrator, and an AI-powered Selection Brush Tool and improved Generative Fill in Photoshop, aiming to significantly boost designers' productivity by integrating AI into everyday tools.

• Mistral released Large 2, an AI model with 123 billion parameters that outperforms larger models like Meta's Llama 3.1 in code generation and math. It offers a 128,000 token context window, supports multiple languages, and minimizes hallucinations, available for trial on Le Chat and cloud platforms with a paid license for commercial use.

• Kuaishou Technology's Kling AI video model is now available globally, generating up to two-minute videos in China and five-second videos elsewhere. It uses advanced 3D reconstruction for natural movements, allowing detailed prompts and offering 66 free credits daily for high-quality video generation.

• Microsoft's Bing is testing a generative search experience that combines traditional results with AI-generated content, including summaries and related sections. The feature aims to enhance user interaction while maintaining website clicks, though its impact on SEO for creators and businesses is yet to be fully understood.

Quickfire News

• OpenAI is reportedly in talks with Broadcom to develop a new AI chip to reduce reliance on Nvidia and enhance global semiconductor infrastructure.

• AI pioneer Fei-Fei Li launched World Labs in April, a spatial intelligence startup valued at over $1 billion, focusing on AI for understanding and navigating 3D environments.

• Meta and Apple plan to exclude EU customers from upcoming AI releases due to regulatory uncertainty with GDPR data protection laws.

• Tech giants Google, OpenAI, Microsoft, and Amazon formed the Coalition for Secure AI (CoSAI) to create best practices and tackle AI security challenges.

• OpenAI introduced new compliance and administrative tools for ChatGPT Enterprise to support large-scale AI deployments in regulated industries.

• Sam Altman announced on X that the alpha release of OpenAI’s Voice Mode is expected later this month, with general availability following shortly after.

• Ukraine is reportedly developing AI-enabled war drones to overcome Russian signal jamming and enable swarm operations on the battlefield.

• Electronic Arts used AI to create digital clones of 11,000 college football players for their upcoming video game in 3 months using athletes' headshot photographs.

• Perplexity rolled out Voice Mode to its AI assistant on its iOS app, allowing Pro users to chat and ask questions to the AI search engine through various voice modes.

• Google is reportedly developing a ‘Prompts Gallery’ for its Gemini chatbots, allowing users to curate and share favorite prompts.

• AMD claims its new laptop chips can outperform Apple's M3, offering better performance in multitasking, image processing, and gaming.

• California state legislators are pushing for a bill that would require big tech companies to test AI for "catastrophic" risks before public release.

• Anthropic is reportedly working on a new screenshot tool for Claude, potentially allowing users to seamlessly take screenshots from other screens or tabs.

• NVIDIA is reportedly working on a version of its new flagship AI chips for the Chinese market to comply with current U.S. export controls.

• The Monetary Authority of Singapore committed $74.3M to support financial institutions in building capabilities in quantum computing and artificial intelligence.

• Cohere announced a new funding round of $500M from investors including Cisco, AMD, and Fujitsu, raising its valuation to $5.5B.

• Disney Research published a new paper on its bipedal robots, detailing a system that combines expressive motions and dynamic mobility for lifelike performance.

• Elon Musk tweeted that Tesla will have humanoid robots working internally next year, with high production for other companies expected by 2026.

• Scaled Foundations announced GRID Beta, a free cloud-based IDE to develop AI for robots, aiming to build General Robot Intelligence.

• The FTC launched an investigation into eight companies offering AI-powered "surveillance service pricing" to understand how consumer data is used to personalize pricing.

• Meta AI unveiled a new 'Imagine me' selfie feature, enabling users to create stylized AI-generated selfies based on prompts, and expanded language support with integration into Meta's Llama 3.1 405B model.

• Harvey, an AI platform for legal services, secured a $100M Series C funding round led by Google Ventures, valuing the company at $1.5B, with plans to expand its AI capabilities in the professional services sector.

• Elon Musk proposed a $5 billion Tesla investment into xAI, subject to shareholder and board approval, potentially enhancing Tesla's AI leadership and FSD capabilities.

• OpenAI announced GPT-4o fine-tuning, offering 2 million free daily training tokens until September 23 to help developers customize the model for specific applications.

• Cohere launched Rerank 3 Nimble, a new foundation model for enterprise search and RAG systems, offering 3x faster performance while maintaining accuracy and supporting over 100 languages.

• Stability AI unveiled Stable Video 4D, a new AI model that turns single object videos into multiple videos from eight different angles, with applications in game development, video editing, and virtual reality.

• Microsoft announced collaborations with Mass General Brigham and University of Wisconsin to advance AI foundation models for medical imaging, aiming to enhance radiologists' efficiency and improve patient outcomes.

• Google announced updates to the Play Store, including AI-powered app reviews, curated spaces, and personalized recommendations.

• OpenAI could lose up to $5 billion this year due to high operational costs, including $7 billion for AI training and inference and $1.5 billion for staffing, despite rapid revenue growth, according to an analysis by The Information.

• Microsoft partnered with Lumen Technologies to expand its AI workload capacity, addressing infrastructure shortages in its data centers, while Lumen will adopt Azure cloud services to improve its cash flow.

• Udio released version 1.5 of its AI music generator, featuring improved audio quality, key control, enhanced language results, and new platform features like stem downloads and audio-to-audio remixing capabilities.

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter : make sure to send the link below to your colleagues and friends :

https://www.blacklynxbrief.com/subscribe