Cybernetic Teammates

Good morning,

If you have been reading this newsletter for longer than a few weeks you know that I keep struggling between the admiration for a technology that is revolutionizing the way we work and on the other hand a deeply-rooted suspicion of the secret motives of “Big Tech”.

It’s undeniable that there is progress to report on every week. This week for example - our image generation capabilities were upgraded once again with this time OpenAI adding next-level image generation capabilities in GPT4o. Users on X.com discovered the “Studio Ghibli” prompt and made images like this.

What is amazing this time is that for the first time text is displayed correctly - which spells the end for any graphic designers. My condoleances.

Some examples of AI generated drawings and pictures :

Or a different style (Pixar) of the same meme picture:

It can even do full comic books:

Anyway , back to the original posting.

Just like with any political debate, when you’re trying to gather information and even wisdom about a topic it’s extremely important I think to really listen to the arguments of both sides of the spectrum.

This sounds obvious but we all sit more or less in our own “echo chambers”. We watch the same type of news channels, surround ourselves with people with the same political views and I think some curiosity and empathy towards “the other side” is extremely valuable.

In the world of artificial intelligence i’m trying to seek out “authorities” that represent completely different views.

The problem is that when you listen to them - they sound like an authority and you tend to think what they say is actually true.

If you look like a glass of water that is filled to the exact middle then that glass of water can be considered half-full or half-empty. Both perspectives are 100% correct but the viewpoint is different.

So I’m being pulled from one side to the other. One side clamoring AI is going to change all our lives - it’s going to bring salvation basically. The other side thinks it’s a bubble that is going to pop in a spectacular fashion.

To be honest : the AI haters are a lot much more fun to follow and read. I love me some AI negativity and hate.

This week’s Ed Zitron column was once again a ball of negative fun that is difficult to pry you eyes away from.

After reading that article you’ll think AI is hot garbage.

On the positive side of things I follow people like Ethan Mollick. He’s an AI optimist and conducting studies on how AI is making its way into our work environments.

He conducted an intriguing experiment in cooperation with Harvard University.

To be honest - in the professional environments where I work I actually see this playing out so I tend to side with Ethan Mollick rather than Ed Zitron.

The purpose of the study was how the introduction of an “AI Teammate” would influence productivity, creativity and wellbeing.

They conducted their study at Procter & Gamble, where they asked employees from diverse fields like R&D and commercial strategy tackled actual business challenges, from product design to marketing strategies.

Some people where asked to work solo, others collaborated in pairs, and some had an AI teammate (powered by GPT-4).

The main finding of the study was that individuals working alone with AI matched the productivity and creativity of human teams working without AI.

That means one person equipped with AI performed as effectively as two highly experienced professionals collaborating traditionally. And when human teams of more than one person had AI by their side, they generated even better solutions, notably excelling in crafting exceptional, innovative ideas.

But it wasn’t just about efficiency.

AI erased traditional expertise boundaries. Specialists typically stuck in silos suddenly proposed balanced, cross-disciplinary solutions when paired with AI. Even employees with little to no experience started performing like seasoned pros.

Imagine how transformative this could be in your workplace—no more strict divisions, but instead, a fluid exchange of AI-enabled ideas.

Perhaps most surprisingly, using AI improved participants' emotional experiences, boosting enthusiasm and significantly reducing stress and anxiety. Rather than AI feeling like a threat, it felt like a supportive, energizing presence.

In short: adding a “cybernetic teammate” to your teams adds quality and creativity to the output of the teams and significantly reduces stress.

I’m not sure Ed Zitron’s blood pressure will be lowered by these findings but there you go.

Lots to think about this weekend once again.

AI News

• Anthropic has added real-time web search to Claude 3.7 Sonnet, finally matching competitors like ChatGPT and Gemini. The feature automatically triggers when more current or accurate info is needed and includes direct citations for verification. Though late to the game, Claude’s strong reasoning combined with web access could pose a serious challenge to search-focused tools like Perplexity.

• OpenAI has released new text-to-speech and speech-to-text models via API, allowing developers to generate customizable voices using text prompts and access top-tier transcription performance. The gpt-4o-mini-tts model can shift tone and character (e.g., bedtime voice, pirate), while GPT-4o-transcribe outperforms previous Whisper models. Though powerful, early demos still fall short of the realism offered by rivals like ElevenLabs and Sesame.

• Apple has reassigned Siri development to Vision Pro creator Mike Rockwell amid delays and internal concerns over the assistant’s weak AI capabilities. Rockwell will now report to software chief Craig Federighi, signaling a leadership shakeup aimed at fixing Siri’s troubled rollout. With competitors rapidly advancing, Apple’s long-hyped AI strategy is under pressure to deliver — fast.

• Researchers have developed ECgMLP, an AI model that detects endometrial cancer with 99.26% accuracy from microscopic images — far surpassing the 78–81% accuracy of human specialists. It also performs exceptionally well in identifying other cancers, including colorectal, breast, and oral. As AI continues to outperform human diagnostics, tools like this could dramatically improve early detection and expand access to expert-level screening worldwide.

• Tencent has launched Hunyuan T1, a new reasoning model that rivals DeepSeek R1 and OpenAI’s GPT-4.5, using a hybrid Transformer-Mamba architecture for faster performance and better efficiency. T1 excels in math and Chinese-language tasks while offering highly competitive pricing at just $0.14/$0.55 per million tokens. With China’s top labs now matching — or even surpassing — U.S. models, the global AI balance is shifting faster than expected.

• AI search startup Perplexity has submitted a surprise proposal to acquire TikTok’s U.S. operations, promising to rebuild the recommendation algorithm with full transparency, open-source oversight, and integration of its own AI search tools. The company also plans to use Nvidia Dynamo tech to boost TikTok’s performance and link videos to real-time citations. Whether serious or a headline grab, the April 5 ban deadline means we’ll soon see how this bold move plays out.

• Reve has burst onto the scene with Reve Image 1.0, a new text-to-image model that debuted as the #1 performer on Artificial Analysis’ Image Arena, topping giants like Midjourney v6.1 and Imagen 3. Known internally as “Halfmoon,” the model excels in prompt accuracy, photorealism, and advanced text rendering — and even includes tools for natural language editing and community sharing. With more features promised soon, Reve looks poised to be a serious new contender in AI image generation.

• DeepSeek has quietly released an upgraded version of its V3 model, dubbed V3-0324, featuring a 641GB Mixture-of-Experts architecture that runs on high-end personal devices like Apple’s Mac Studio. Despite the model’s massive size, only 37B parameters activate per token, allowing for smooth performance with lower compute. Now under an open-source MIT License, the update improves math and coding abilities and continues to fuel speculation about the upcoming R2 release.

• The ARC Prize Foundation just launched ARC-AGI-2, a new $1M competition aimed at pushing AI reasoning to the next level. The benchmark includes tasks that are simple for humans but still stump top models — with OpenAI’s o3-low scoring only an estimated 4%. The new challenge adds an efficiency metric to measure how cost-effective AI models are at reasoning, with $700K going to the first team to hit 85% accuracy within the resource limit.

• Google has unveiled Gemini 2.5, its most advanced AI model yet, with built-in reasoning capabilities and a 1 million-token context window (soon expanding to 2 million). The 2.5 Pro version tops the LMArena leaderboard and shows major improvements in coding, math, and science — especially in building web apps and agentic tools. With reasoning now a default feature, Google is raising the baseline for AI performance, though fierce competition means the lead may be short-lived.

• OpenAI has upgraded GPT-4o with native image generation, fully integrating visual output into its multimodal system and replacing DALL·E 3 in ChatGPT. The update delivers far more accurate images with readable text and improved context — especially for menus, diagrams, and infographics — while also supporting iterative editing through natural language. This marks a leap in how AI handles visual content, pushing it closer to real design and communication tasks.

• Apple is reportedly spending $1 billion on Nvidia servers to build out its first major generative AI infrastructure — a move suggesting a shift away from relying solely on in-house chips. The purchase includes around 250 of Nvidia's ultra-high-end GB300 NVL72 systems, with partners Dell and Super Micro helping scale the deployment. After delays with Siri and internal struggles, Apple is now playing catch-up in the AI race, finally committing big resources to stay in the game.

• Ideogram has released Version 3.0 of its AI image model, delivering major upgrades in photorealism, layout control, and especially text rendering. It now rivals top competitors in graphic design and even outperforms Imagen 3 and Flux Pro in human tests. With free access and features like style reference uploads, Ideogram 3.0 marks a big step forward — even if its launch is being overshadowed by the buzz around OpenAI’s new visual tools.

• Alibaba and BMW have announced a strategic partnership to bring next-gen AI assistants to BMW’s in-car systems by 2026. The collaboration will use Alibaba’s Qwen model to power voice-enabled features like dining suggestions, traffic updates, and gesture-based controls. It also includes two new AI agents for car diagnostics and trip planning, reflecting how automakers are increasingly turning to AI to redefine the driving experience.

• Alibaba also released Qwen2.5-Omni-7B, a new multimodal model that can process text, images, audio, and video simultaneously — and run efficiently on phones and laptops. With strong performance in real-time speech understanding and accessibility-focused applications, Omni-7B is designed for broad use and is now open-source. It’s another major step toward everyday AI agents that can truly understand the full spectrum of human interaction.

Quickfire News

• OpenAI released its o1-pro model via API, pricing access at $150 per million input tokens and $600 per million output tokens, roughly 10x the cost of regular o1.

• Perplexity is preparing to raise nearly $1 billion at an $18B valuation, potentially doubling its worth as it nears $100M in annual recurring revenue.

• Google’s NotebookLM added an Interactive Mind Map feature, enabling users to turn notes and source material into visual diagrams.

• Meta began rolling out its AI assistant across 41 European countries through Facebook, Instagram, WhatsApp, and Messenger, following a year-long regulatory delay.

• Hugging Face published a response to the White House’s AI Action Plan, urging support for open-source AI systems over closed commercial models to foster innovation.

• Anthropic introduced a "think" tool for Claude, allowing the AI to perform structured reasoning during complex tool use tasks.

• OpenAI and Meta are exploring a partnership with India's Reliance Industries, with OpenAI considering an up to 85% price cut on ChatGPT to gain market share.

• Kai-Fu Lee commented that "Sam Altman is probably not sleeping well," as his startup 01.AI pivots to DeepSeek's open models, operating at just 2% of OpenAI’s annual costs.

• Apple is developing an Apple Watch with a camera and AI Visual Intelligence features, using its own models and targeting implementation by 2027.

• Zapier launched its own MCP protocol, enabling AI assistants to take actions across 8,000+ apps without requiring custom API integrations.

• Studies by OpenAI and MIT found a correlation between increased ChatGPT use and greater loneliness and emotional dependence in users.

• Browser Use, which builds AI-friendly web interfaces, raised $17M in seed funding from Felicis, Paul Graham, A Capital, and Nexus Venture Partners.

• Sam Altman announced new leadership promotions at OpenAI, with Mark Chen named Chief Research Officer and Brad Lightcap’s role expanded as COO.

• Alibaba’s Qwen team open-sourced Qwen2.5-VL-32B-Instruct, a vision-language model with improved mathematical reasoning and enhanced visual capabilities.

• Netflix co-founder Reed Hastings donated $50M to launch the Hastings Initiative for AI and Humanity, focused on studying AI risks and long-term impact.

• Ant Group, affiliated with Alibaba, adopted a hybrid chip strategy using both Chinese and American chips, cutting AI development costs by 20%.

• Google began rolling out Project Astra features to Gemini, enabling advanced vision, live video input, and screen reading capabilities.

• Alibaba released LHM, a new AI model that generates animated 3D avatars from just one reference image.

• OpenAI announced upgrades to Advanced Voice Mode, adding new personality options and fewer interruptions for more natural conversations.

• Figure AI published research and demos of its Figure 02 humanoid, showing natural human-like walking after completing years of simulated training in just hours.

• H&M is partnering with 30 models to create AI-based digital twins for ad campaigns, offering usage-based compensation and ownership rights to the models.

• ByteDance released InfiniteYou, an open-source AI portrait generator that delivers consistent portraits with accurate facial detail and strong prompt alignment.

• Synthesia launched a $1M equity program for actors whose likenesses are used as AI avatars, becoming the first to offer stock compensation for avatar contributors.

• Otter AI introduced three AI Meeting Agents, including a voice-activated assistant, a Sales Agent for live coaching, and an SDR Agent for autonomous product demos.

• Perplexity added new answer modes, enhancing search results with images, videos, and interactive cards that include built-in commercial transactions.

• OpenAI will adopt Anthropic’s open-source Model Context Protocol, allowing ChatGPT and other tools to connect with external data and software.

• Microsoft 365 Copilot introduced Researcher and Analyst, two new AI agents that handle research and data analysis tasks directly within workplace tools.

• A federal judge denied UMG’s request to block Anthropic from using song lyrics to train Claude, ruling the company did not show irreparable harm.

• xAI integrated its Grok chatbot into Telegram, now available for Premium users at no extra cost.

• Amazon launched ‘Interests’, an AI shopping feature that scans the store and notifies users of new products based on natural language prompts.

• Midjourney announced its V7 model is expected to launch on Monday, March 31, following updates during its weekly Office Hours session.

• The U.S. government blacklisted over 50 Chinese tech firms, targeting those involved in advanced AI, quantum technology, and supercomputing.

Like newsletters?

Here are some newsletters our readers also enjoy:

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter, please pay it forward !

Thank you for being here !