Get Out Of The AI Closet Now

Good morning,

Ugh. So much to talk about this week and so little time !

Last night, Tesla presented the “Cybercab” to the world. Full self-driving. No more steering wheel, no pedals. You just lay back and enjoy the ride.

The vision is no more accidents, no more deaths. We also saw the unexpected entrance of the “Robovan” - although that looks a bit like it was slapped together last minute.

I used to be very skeptical about the concept of driverless cars and “Full Self-driving”.

But when I was in San Francisco in March I saw my first Waymo and it was almost a religious experience. I started to study how it works and now I actually believe we’re going to get there (LiDar people LiDar). But .. that is for another edition of the newsletter. Certainly in the US I can see that working but here the government has some work to do.

I drive around in a Tesla and if I didn’t step on the brake a few times my car would have gotten me killed already. So little bit skeptical there.

Another notable event this week is that the “godfather of AI” Geoffrey Hinton and his colleague John Hopfield won the Nobel Prize in Physics for discoveries that paved the road to artificial intelligence. In an interview afterwards, Hinton warned again about the dangers of AI - that it’s on its way of becoming smarter than us and that humanity is not ready for that.

Quite significant that one of the smartest people in the field uses every opportunity he has to warn us that AI essentially is like bringing an alien life form to life, and he also said “these LLMs really understand what we’re saying”.

Either way, this is not good. Or he’s delusional or he’s right. The implications of both options are scary. Also good material for another newsletter.

BUT .. I had planned to talk about something else.

I wanted to talk about something nobody is talking about. Or DARES to talk about.

According to one study, more than 60% of employees in big corporations are using AI. Yet, ask their managers, and they’ll swear it’s less than 10%.

Which means people are hiding it.

We’re living in a strange limbo where everyone—from your accountant to your neighbor’s cat—is using ChatGPT, but no one’s willing to say it out loud.

Few weeks ago I found myself at a party surrounded by “management consultants”. So I asked them about AI.

The results of this non-scientific survey? They were all heavily using AI in their assignments. I basically had to drag the answer out of them and thank the Lord for alcohol or I would have gotten zero answers.

Meanwhile, Mr. Manager sits in his corner office, blissfully unaware, humming the company anthem of “AI is bad” and “self-crafted work is pure.” At the same time, that consultant you just hired is billing you for hours of manual labor while the real heavy lifting gets done in five minutes by a machine.

And it’s your own fault.

In many companies AI is treated like that family in Harry Potter treated him. They shove it under the stairs and shout at it.

This secretive AI usage is a ticking time bomb. Do you really think there’s not a massive data leak happening right now, with your company’s sensitive information being siphoned off to tech giants every time someone plugs a task into the AI pipeline?

Companies need to wake the hell up and set some rules.

In what sectors - other than consulting is AI prevalent ? According to an anonymous study in Denmark, the vast majority of marketing professionals, journalists, and software developers are riding the AI wave. No shockers there. But wait—teachers clock in at almost 40%. Teachers !

We need to out ourselves as AI users to the companies we’re working for. You’re not “cheating” when you use it, you’re being more efficient. It has a bit of a bad rep, I feel.

We won’t tackle the risks of AI when nobody is willing to admit they’re using it.

Want to tell me how your company uses AI or you use AI at your company? Let me know ! (Anonmously) (grin).

Now, onto the news!

Welcome to the Blacklynx Brief!

(2 paragraphs of this intro were AI generated - can you guess which ones?)

By the way, there is now a referral program for this newsletter. For every 3 people that you get to sign up, we will gift you a 15€ Amazon gift card.

Scroll to the bottom of the newsletter for details!

AI News

• OpenAI has launched Canvas, a new interface for ChatGPT that enables collaborative writing and coding with editing features, shortcuts, and contextual knowledge. Canvas, which opens alongside the chat, offers inline feedback and targeted editing options, boosting output accuracy by 30%. It’s currently in beta for Plus and Team users, with broader access planned for the future.

• Google announced ads within its AI Overview search summaries and new AI search capabilities, including video understanding and voice input. These updates include in-text links in AI Overviews, AI-organized search results for diverse content, and features like song identification through video searches. As Google leans further into AI-driven search, monetizing Overviews with ads may raise questions about information bias.

• Black Forest Labs released Flux 1.1 Pro, a faster and more precise text-to-image model available via API for developers. The model, which surpasses competitors like Midjourney in quality, will be offered through partners at $0.05 per image. Flux 1.1 Pro showcases advancements in speed and image fidelity, pushing the boundaries of AI-generated visuals.

• Meta announced Movie Gen, a suite of AI models for creating and editing video and audio content, with capabilities for HD video generation, synchronized audio, and natural-language video editing. Set to launch on Instagram next year, Movie Gen aims to simplify content creation, positioning Meta as a key competitor to other video generation tools like OpenAI’s Sora. This rollout could revolutionize user-generated content by providing advanced editing tools accessible through simple prompts.

• OpenAI’s case study on Altera highlights the startup’s use of GPT-4o to create “digital humans” that autonomously play Minecraft, achieving significant in-game success. Altera’s agents, designed for prolonged and natural interactions, could soon expand to roles as AI “coworkers” and complex simulations, suggesting rapid advancements in AI agent capabilities that may arrive sooner than expected.

• Cleveland Clinic and IBM have developed LISA-CPI, an AI model that predicts interactions between drugs, gut microbes, and pain receptors, which could aid in identifying non-addictive pain treatments. This model has pinpointed FDA-approved drugs for potential repurposing and revealed new microbe interactions, offering a promising path for faster drug discovery focused on pain management amid the opioid crisis.

• Two Harvard students have developed I-XRAY, a proof-of-concept system using Meta’s Ray-Ban smart glasses to identify strangers and retrieve personal information, sparking privacy concerns. By combining facial recognition, reverse image search, and large language models, the system successfully identified individuals and their details on Harvard’s campus. This demo underscores how rapidly AI is transforming privacy and surveillance, highlighting the potential risks as such technology advances.

• Inflection AI launched Inflection for Enterprise, a new business-focused AI system built in partnership with Intel, featuring cloud services, an API, and an upcoming on-premises appliance powered by Intel’s Gaudi 3 accelerators. Inflection 3.0, the underlying model, includes versions for chatbots and productivity tasks, positioning Inflection to compete in the enterprise market after recent leadership changes. With Intel’s backing, the startup aims to deliver high-performance AI tools tailored for large-scale business applications.

• Researchers from the University of Oxford and Cohere introduced TICK, a novel method for AI model evaluation that generates checklists to assess performance. TICK improves evaluation accuracy by 5.8% over standard methods and can automate the checklist process, reducing costs. Additionally, the self-improving STICK variant enhances model performance on reasoning tasks, suggesting that simplified, checklist-based approaches could be a promising avenue for refining AI capabilities.

• OpenAI is reportedly seeking to reduce its dependence on Microsoft for computing power, planning to lease a data center from Oracle and develop its own AI chip. Tensions have surfaced over Microsoft’s speed in supplying resources and disagreements on their joint data center project, Fairwater. As OpenAI explores more independent infrastructure, its evolving relationship with Microsoft could significantly impact the AI landscape.

• Scientists Geoffrey Hinton and John Hopfield received the 2024 Nobel Prize in Physics for their foundational work in machine learning and neural networks, which underpins today’s AI advancements. Both have expressed concerns over AI’s potential dangers, with Hinton, a vocal critic since leaving Google, advocating for stricter regulations to ensure AI safety. Their shared Nobel win underscores the dual nature of AI’s progress—both transformative and, to some, unsettling.

• Adobe has launched a free web app, Adobe Content Authenticity, to help creators protect and attribute their work amid growing concerns about AI-generated content. The app uses digital fingerprinting and watermarking to attach content credentials, which can indicate whether creators permit AI training on their work. While promising, its effectiveness will depend on widespread adoption and cooperation from other tech platforms.

• AI startup Writer has launched Palmyra X 004, a large language model (LLM) designed for enterprise use, outperforming models from OpenAI, Anthropic, and Google in tool-calling accuracy. With a 128k context window, support for multimodal inputs, and advanced tool interaction capabilities, Palmyra can execute tasks like database updates and email automation. This model's success suggests that highly capable enterprise AI doesn’t always need the vast resources traditionally associated with top AI labs.

• At Zoomtopia 2024, Zoom introduced several AI features, including AI Companion 2.0, Zoom Tasks, and customizable AI avatars set to roll out next year. These tools offer deeper context and automation across the Zoom platform, aiming to reshape work by enabling AI-driven task management and even avatar participation in meetings. This points to a future where AI seamlessly integrates with daily workflows, making virtual work environments more autonomous.

• Penn State scientists have developed an AI-based 'electronic tongue' capable of detecting subtle differences in liquids with high accuracy, such as distinguishing between spoiled and fresh juices. The device uses a neural network to mimic human taste processing and has potential applications in food safety and quality control. This technology represents a new frontier in AI's sensory capabilities, allowing for deeper understanding and analysis of how AI makes complex decisions.

Quickfire News

• Google released Gemini 1.5 Flash 8B, a cost-effective version with a 50% reduction in costs and twice the rate limits compared to the 1.5 Flash model.

• Fourier launched GR-2, its second-generation humanoid robot, with enhancements in battery life, hand dexterity, mobility, and a new developer kit.

• The U.S. Commerce Department announced a $100 million grant program for AI semiconductor research, aiming to develop more sustainable materials.

• OpenAI secured a new $4 billion credit facility from major banks, raising its total liquidity to over $10 billion for continued growth and innovation.

• AI coding startup Poolside raised $500 million in a Series B funding round, bringing its valuation to $3 billion as it accelerates work towards AGI.

• Apple is expected to release its Apple Intelligence features on October 28, coinciding with the iOS 18.1 update, according to Bloomberg insider Mark Gurman.

• Google started rolling out new AI anti-theft features for Android, including Theft Detection Lock, Offline Device Lock, and Remote Lock, initially showcased at Google I/O.

• Cohere launched enhanced fine-tuning features for its Command R LLM, offering longer context support and a "bring your own fine-tune" option.

• AI startup Otherside AI’s Reflection 70B model failed to meet performance claims, as noted in a post-mortem by the team after initial promotion as the "world’s best open-source model."

• North Carolina musician Michael Smith faces federal charges for allegedly using AI to generate thousands of songs and bots to stream them billions of times, netting over $10 million in royalties.

• Former Google CEO Eric Schmidt stated at the Washington AI Summit that AI advancements should take precedence over climate goals, arguing that current efforts aren't organized enough to meet those targets.

• Northrop Grumman introduced an AI-powered upgrade to its Forward Area Air Defense system, designed for rapid decision-making against drone swarm attacks.

• Grindr is developing an AI "wingman" feature for its dating app, which can scout prospective partners, set up dates, and interact with other AIs to find matches.

• Nvidia and Peking University researchers launched EdgeRunner, a model that enables high-quality, detailed 3D mesh generation.

• Enterprise GenAI startup Writer is reportedly set to raise $150-200 million at a $1.9 billion valuation, doubling from its previous $100 million Series B valuation last September.

• Security researcher Harish SG published findings showing that LLMs can achieve advanced reasoning capabilities, similar to OpenAI’s o1 model, through a combination of sophisticated prompt techniques.

• OpenAI and Hearst announced a strategic partnership to incorporate content from over 20 magazine brands and 40+ newspapers into OpenAI's AI products.

• Hugging Face released OpenAI-Gradio, a tool for quickly creating AI-powered web apps with OpenAI models using minimal code.

• Uber plans to launch an OpenAI-powered AI assistant in early 2025 to assist drivers with electric vehicle-related questions, aiming to boost EV adoption on its platform.

• Anthropic introduced the Message Batches API, which allows developers to submit up to 10,000 queries for asynchronous processing within 24 hours at a 50% discount compared to standard API rates.

• Google added drag-and-drop functionality to its AI Studio, enabling direct file uploads without needing to first import files into Google Drive.

• KoBold Metals raised $527 million to expand its AI-driven mineral discovery technology, which analyzes extensive data to locate deposits of critical minerals like copper, lithium, and nickel.

• Google DeepMind’s Demis Hassabis and John Jumper were co-awarded the Nobel Prize in Chemistry for their work on AlphaFold, an AI system that predicts and designs protein structures.

• Amazon launched AI Shopping Guides for over 100 product categories, using generative AI to help users research products and receive tailored recommendations within its U.S. app and mobile site.

• Chinese startup MiniMax’s Hailuo AI introduced a new image-to-video feature, along with enhanced style controls and processing capabilities.

• Meta expanded the availability of Meta AI to six additional countries, including the EU, and is rolling it out internationally in Ray-Ban Meta smart glasses; however, the EU version will exclude multimodal features due to regulatory restrictions.

• Stripe announced an expanded partnership with NVIDIA, allowing global access to NVIDIA’s AI cloud services and using the platform to enhance fraud detection capabilities.

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter. Please pay it forward and receive 5€ for every person you get to sign up in the form of an Amazon gift card. What’s keeping you ?