OpenAI Model Rebuilds Stem Cells

In partnership with

This week in AI and tech, we’re seeing breakthroughs that push the boundaries of science, hardware, and reasoning. OpenAI and Retro Biosciences have redefined stem cell reprogramming with an AI-engineered protein breakthrough delivering 50x efficiency. Malaysia takes a bold step into the global AI race with its first homegrown edge AI chip, the MARS1000, built on 7nm technology. Meanwhile, a new method called DeepConf is revolutionizing large language model reasoning by filtering low-confidence paths, achieving near-perfect accuracy while slashing computational costs.

OpenAI and Retro Bio Achieve 50x Stem Cell Breakthrough

OpenAI and Retro Biosciences collaborated to develop GPT-4b micro, an AI-powered model specialized for protein engineering, which led to the creation of novel variants of the Yamanaka factors—key proteins used in stem cell reprogramming—with over 50-fold higher expression of stem cell markers and enhanced DNA damage repair compared to traditional methods. This breakthrough, validated in multiple experiments across varied donor cells and delivery methods, accelerates life sciences research by enabling higher pluripotency and genomic stability in induced pluripotent stem cells (iPSCs), showing promise for advanced cell therapies and rejuvenation treatments. Additionally, the new AI-generated protein sequences exhibited deep sequence edits and a high hit rate, significantly outpacing conventional engineering approaches and rapidly boosting the efficiency of cellular reprogramming, especially in aged or hard-to-treat cells.

Malaysia Launches Nation’s First Homegrown Edge AI Chip

Malaysian semiconductor company SkyeChip has unveiled the MARS1000, Malaysia’s first edge AI processor designed using advanced 7nm process technology. This chip is intended to power smart devices such as robots, cars, smart agriculture systems, Industrial 4.0 applications, smart cities, and traffic management, marking a significant step for Malaysia in the global AI and semiconductor race. Although not as powerful as high-end processors used in large data centers, the MARS1000 aims to foster local AI development and literacy, especially among students, and support the growth of a thriving AI industry in the country. This development aligns with the Malaysian government’s commitment to invest at least RM25 billion in semiconductor industry growth, strengthening the nation’s role as a major player in semiconductor packaging and manufacturing despite international trade restrictions impacting the sector.

Is Your Amazon Strategy Actually Working? Here’s What Top Brands Do Differently.

At Cartograph, we’ve worked with some of the most innovative brands in CPG—OLIPOP, Starface, and Rao’s—and understand the nuances of selling consumables on Amazon.

Are you a fast-growing brand in Food & Beverage, Supplements, Beauty & Personal Care, Household, Pet, or Baby?
Growing 50%+ YoY?
Do you know your Amazon profitability (and are you happy with it)?

We’ve spent the past 7 years helping CPG brands scale profitably on Amazon. What makes Cartograph different:
• Deep CPG focus
• No more than 4 brands per team
• Monthly P&L forecasts within 5% accuracy
• Daily reporting via Slack

Click below to get a custom, human review of your Amazon account—not just another automated report.

Get An Amazon Audit

Dynamic Confidence Filtering for Scalable LLM Reasoning

Deep Think with Confidence (DeepConf) is a novel method that enhances the reasoning capabilities and efficiency of large language models by using internal confidence signals to filter out low-quality reasoning paths. Unlike traditional majority voting approaches that treat all reasoning traces equally and incur heavy computational costs, DeepConf dynamically assesses the confidence of partial reasoning steps to stop low-quality traces early during generation. This confidence-aware filtering allows the model to maintain or improve final answer accuracy while significantly reducing the number of tokens generated. Evaluated on challenging benchmarks such as AIME 2025 with models like GPT-OSS-120B, DeepConf achieved up to 99.9% accuracy while reducing token usage by up to 84.7%. The method requires no additional training or hyperparameter tuning and can be seamlessly integrated into existing inference systems, offering a scalable and practical solution for efficient, high-quality reasoning in large language models.

Create stunning news videos fast with AI video editing tools! Instantly remove backgrounds, polish footage, and batch process clips for broadcast-ready exports. Boost newsroom speed, skip green screens with real-time background replacement, and export in TV or social formats effortlessly. Perfect for anchors, field reports, and digital news—save hours and deliver professional quality with smart AI tools.

[Start Editing Now]

Top Tools from this week

• Creem 1.0 - Python-native agentic framework enabling AI agents to work directly with libraries and runtime. Supports complex objects, sandboxed code execution, multi-agent orchestration, and real-time observability.

• Draw A Fish - Draw a fish and watch it swim in the global tank with everyone else's fish. Create, share, and vote on fish drawings.

• Pikto AI Studio - Elon Musk's xAI open-sourced Grok 2.5, its best model last year, now available on Hugging Face. Grok 3 will be open source in about 6 months, continuing xAI’s open-access push.

• TaskWand - Never spend hours building n8n workflows again. Describe your automation, get instant results. Complete workflows with conditions, loops, and error handling - ready to deploy in seconds.

• AIBI Pocket - Meet AIBI Pocket, a tiny AI companion that fits in your hand or attaches to your shirt. It recognizes you, chats with ChatGPT, rotates like a gimbal, shows weather with fun animations, provides reminders, and even plays with other AIBIs.

• Tokyo - Advanced AI toolkit for tracking and analyzing user interactions across multiple clients. Streamline your AI operations with our comprehensive monitoring solution.

This week in AI

• YouTube’s AI Experiment - YouTube is quietly testing AI video enhancements on Shorts, altering creators’ content with sharper, plastic-like visuals, sparking trust and authenticity concerns.

• Meta x Midjourney - Meta partners with Midjourney to license its AI image and video tech, aiming to boost future AI models and compete with leaders like OpenAI’s Sora and Google’s Veo.

• Bot vs Bot Future - As AI agents multiply, bots will dominate the internet, negotiating prices, trading, and even hacking turning online life into a machines-first ecosystem with humans sidelined.

• Amazon’s AI Agent Push - Amazon is investing heavily in AI agents through its AGI lab, aiming to move beyond chatbots to reliable task-completing systems as it races to catch rivals in the AI race.

Paper of The Day

The "State of AI in Business 2025" report shows AI adoption is rapidly growing, with over 70% of companies using generative AI across multiple business areas. However, 95% of generative AI pilots fail to deliver significant business impact, mainly due to integration and workflow issues. Successful AI deployment requires strategic focus, organizational changes, and governance. AI investments are rising, especially in back-office automation, which yields better ROI than many front-end uses. Adoption varies, with higher usage and impact seen in large companies and AI-native firms. Overall, AI is moving from hype to practical execution but still faces adoption and impact challenges.

To read the whole paper, go to here.