The Countdown to AGI šŸ“…ā±ļø

AGI predictions move to 2040, OpenAI’s leadership rift surfaces, and OpenFold3 opens cutting-edge biomolecular research to all through open-source innovation.

The AI landscape is heating up with breakthroughs that push intelligence, transparency, and accessibility to new heights. Experts now predict AGI could arrive as soon as 2040, internal tensions at OpenAI shed light on the power struggles shaping AI governance, and OpenFold3 opens the door to a new era of open science in molecular biology.

🧠 AGI Arrival Accelerates Countdown to Singularity – A sweeping global survey of over 8,500 AI experts suggests Artificial General Intelligence may emerge by 2040, far earlier than earlier forecasts of 2060. With rapid progress in models like GPT-5 and Gemini, many now foresee the singularity arriving in the 2030s, signaling a profound transformation in human–AI collaboration.

⚔ Ilya vs. Sam The Hidden Rift at OpenAI – Newly revealed deposition documents expose a 52-page memo from co-founder Ilya Sutskever detailing conflicts with CEO Sam Altman, citing manipulation and governance issues. The fallout from the 2023 leadership crisis continues to reverberate, intertwining OpenAI’s internal politics with broader industry debates on AI safety and corporate transparency.

🧬 OpenFold3 Open-Source AlphaFold3 Challenger – Enter OpenFold3, a community-driven model rivaling DeepMind’s AlphaFold3 in biomolecular structure prediction. Supporting proteins, RNA, DNA, and small molecules, it democratizes access to cutting-edge folding AI, accelerating discoveries in drug design and molecular biology through open collaboration.

From predicting human-level intelligence to decoding the molecules of life, AI’s evolution is becoming more transparent, transformative, and globally inclusive than ever before.

AGI Arrival Accelerates: Experts Predict Singularity by 2040

Most AI experts agree that Artificial General Intelligence (AGI), the point at which machines achieve human-level intelligence across diverse tasks, is inevitable. Recent surveys and predictions from around 8,590 AI researchers, scientists, and industry professionals show a trend toward expecting AGI sooner than previously thought. While early predictions around 2019 suggested AGI by 2060, the consensus estimate in 2025 is around 2040, with many entrepreneurs even more optimistic, predicting it as early as the 2030s. Some prominent figures foresee it happening between 2026 and 2035, fueled by rapid advances in large language models like GPT-5 and DeepMind's Gemini, exponential growth in computing power, and scaling of current AI architectures. However, there is no single agreed-upon path to AGI, and some experts believe new methods beyond scaling are necessary. Challenges include defining and measuring AGI, ethical concerns, and high resource demands. Despite uncertainties, the general view is that the singularity will likely occur within the next two decades, reshaping AI and society profoundly.​

Ilya Unveils Hidden Conflict in OpenAI Leadership

The deposition document reveals that Sam Altman’s ousting from OpenAI was part of a longer brewing conflict, documented extensively by co-founder Ilya Sutskever who prepared a detailed 52-page memo outlining patterns of dishonesty, manipulation, and leadership issues attributed to Altman. Most of the evidence for this memo came from former CTO Mira Murati. The internal friction involved multiple board members and executives, with additional memos criticizing other key figures like Greg Brockman. During the crisis, there were also talks of a potential merger with Anthropic, led by Dario Amodei, which Sutskever opposed. The dramatic board decisions in November 2023 reflected deep disagreements over AI safety, company direction, and governance, and led to a temporary removal of Altman followed by his rapid reinstatement after widespread employee pushback. This saga has continued to influence the AI industry’s leadership dynamics and is now a focus of ongoing litigation initiated by Elon Musk challenging OpenAI’s restructuring.​

Open-Source AlphaFold3 Alternative for Biomolecular Structure Prediction

OpenFold3 is an open-source biomolecular structure prediction model developed as a close reproduction and competitor to DeepMind's AlphaFold3. It accurately predicts the 3D structures of proteins, RNA, DNA, and small molecules, supporting complex multi-chain and molecular interactions essential for drug discovery and biochemical research. While it matches or exceeds AlphaFold3 in several modalities like RNA modeling, it is slightly behind in antibody-antigen docking but offers broader accessibility due to its open-source license. OpenFold3 accelerates biomolecular research by democratizing access to advanced protein folding AI technologies and supporting a wide range of applications in computational biology and drug development.

Hand Picked Video

In this video, we'll look at MCP Update by Claude.

Top AI Products from this week

  • Jinna.ai - Jinna.ai is built to get you paid faster. Talk, type, or upload a file to create an invoice. Make it yours: add a logo, photo, video, signature, even music. Include bank details or a Stripe payment link for instant payment. Jinna sends it via link or crafts a friendly email to your client.

  • Softr Workflows - Build automations, user interactions, and AI agents that power your Softr apps and business — all in one place. Set up triggers, actions, and logic with a visual builder, or start fast with templates and an AI Co-builder.

  • Firecrawl v2.5 - Firecrawl v2.5 is the world's best Web Data API, powered by a new Semantic Index & custom browser stack. We deliver the highest-quality,

  • MCP Playground- MCP Playground is a web-based developer tool designed to inspect and test Model Context Protocol (MCP) servers. It provides an interactive environment for exploring tools, resources, and prompts exposed by MCP servers, making it easy to debug and develop MCP integrations.

  • Lovelace - Lovelace is a browser-based AI IDE for developers who code from anywhere. It delivers AI-powered code completion, generation, and an integrated AI Agent across any device. The tool provides cloud-based workspace management from tablets, phones, or any browser.

  • Expertise Booking - Expertise Booking is a powerful and intuitive scheduling tool designed to streamline how you book meetings, whether you’re working alone or as part of a team.

This week in AI

  • Apple to Integrate Google Gemini AI in Siri Revamp - Apple’s new Siri will ā€˜lean’ on a custom Google Gemini AI model in a major 2026 overhaul, enhancing voice interaction and AI-powered web search while ensuring data privacy by running on Apple’s servers.

  • Valley ByteDance's Multimodal AI Powerhouse - advanced multimodal AI excels in text, image, and video tasks, ranking top in e-commerce and video benchmarks with open-source demos under Apache-2.0 license.

  • OpenAI & AWS $38B AI Infrastructure Deal - AWS and OpenAI partnership boosts AI with $38B, scaling compute via EC2 UltraServers and NVIDIA GPUs to power advanced AI models like ChatGPT and GPT-5 efficiently.

  • Odyssey-2 Real-Time Interactive AI Video - Odyssey-2 streams instant, interactive AI video at 20 FPS, letting users shape multi-minute videos via text prompts in real time, simulating physics and dynamic scenes.

  • MCP 1st Birthday - Hugging Face celebrates Model Context Protocol's first year with community projects, hackathons, and resources to build innovative AI apps using MCP technology.

Paper of The Day

The paper proposes a Hybrid Retrieval-Augmented Generation (RAG) Agent for trustworthy legal question answering in judicial forensics. It combines a legal knowledge base with multiple large language models to ensure accuracy, traceability, and up-to-date responses. When reliable legal data is found, the system uses retrieval-based grounding; otherwise, it generates answers via an ensemble of LLMs and selects the best output using a scoring model and human review. Tested on Chinese legal QA datasets, the approach significantly reduces hallucinations and outperforms standard LLM and RAG models in answer quality and reliability.


To read the whole paper šŸ‘‰ļø here