Didn't Make the Cut

Cybersecurity researchers identified a critical 'by design' vulnerability in Anthropic’s Model Context Protocol (MCP), potentially enabling remote code execution (RCE) across thousands of AI-integrated software packages.

The Hannover Messe industrial trade fair opened with a significant focus on 'Industrial AI' and humanoid robots. NVIDIA and its partners showcased AI-driven manufacturing solutions, including agentic design and real-time digital twins. German Chancellor Friedrich Merz underscored AI's pivotal role in achieving Germany's goal of doubling its computing capacity by 2030, signaling a national commitment to industrial AI infrastructure.

SK hynix announced the mass production of 192GB SOCAMM2 memory modules, specifically engineered to eliminate memory bottlenecks in the training of large language models (LLMs) on NVIDIA’s forthcoming Vera Rubin platform.

The world's largest industrial trade fair, Hannover Messe, opened with industrial AI and humanoid robots as central themes. German Chancellor Friedrich Merz stressed AI's role in boosting productivity and competitiveness, announcing plans to double Germany's computing capacity by 2030. NVIDIA also demonstrated "agentic design" and Industrial AI Cloud applications at the event.

A significant update to the OpenAI Codex app now enables the AI to directly control macOS and Windows operating systems, allowing it to perform complex tasks across multiple applications autonomously. This marks a major leap in AI autonomy and its integration into daily computing.

Google has initiated the global rollout of a native AI Mode within its Chrome browser, integrating Gemini capabilities directly for real-time article summarization and contextual Q&A. This move brings advanced AI features to millions of users, fundamentally altering web browsing and information consumption.

Mira Murati's new AI venture, Thinking Machine Labs, has raised $2 billion, signaling significant investment in cutting-edge AI research and development.

Fei-Fei Li's spatial intelligence startup, World Labs, achieved a valuation of over $1 billion, indicating strong investor confidence in its innovative AI applications.

Reports indicate that Elon Musk is initiating a joint venture between Tesla and SpaceX to establish a 'Terafab,' a facility designed to supply 1 terawatt of annual computing capacity. This ambitious project aims to produce custom semiconductors specifically tailored for the development and deployment of humanoid robots.

Apply Digital launched TORQ AI, an "agentic accelerator" on Google Cloud, designed to provide CPG, retail, and media enterprises with a production-grade path to functional AI within 30 days. This platform aims to significantly reduce the time and complexity of deploying AI solutions in businesses.

The GSMA (telecoms industry association) warned that the massive demand for AI-specific memory chips is diverting resources away from standard networking hardware, hindering efforts to connect the 2.2 billion people worldwide who still lack internet access.

Canada's New Democratic Party (NDP) has called for a federal ban on 'surveillance pricing,' where AI algorithms set different prices for the same item based on an individual's personal data and search history.

NVIDIA announced the world’s first family of open-source quantum AI models, named NVIDIA Ising. These models are designed to accelerate the development of useful quantum computers by improving processor calibration and error-correction decoding up to 2.5 times faster than traditional methods.

Google.org committed an additional $15 million, bringing its Digital Futures Fund to a total of $35 million. The fund supports independent global think tanks and academic institutions researching AI’s long-term impacts on the global economy, workforce opportunities, and cybersecurity.

Intel announced "NeuroBridge," a revolutionary chip architecture designed specifically for highly energy-efficient training and inference of next-generation large AI models. The new design promises a 5x improvement in compute-per-watt efficiency compared to current leading AI accelerators.

Z.ai has released its GLM-5.1 model under an MIT License. This model is specifically optimized for 'long-horizon autonomous engineering,' designed to handle extended autonomous tasks rather than short-burst reasoning. Its open-source nature aims to foster broader innovation and development in the field of autonomous AI systems.

Accenture acquired Keepler Data Tech, a Spanish cloud-native AI firm. This acquisition is intended to strengthen Accenture's offerings in 'agentic AI' and data foundation services for its European enterprise clients, signaling a growing focus on integrating autonomous AI into business operations.

A multi-center study published in eBioMedicine showed that large language models (LLMs), specifically GPT-4o, can effectively identify drug safety signals and adverse events from clinical notes with high accuracy using zero-shot learning.

Over 512,000 lines of source code for Anthropic's Claude Code CLI were leaked, sparking industry-wide discussions on AI intellectual property security. The incident highlights the growing challenges and vulnerabilities in protecting proprietary AI technologies.

Anthropic's AI model, Claude, surged to the #1 spot on the U.S. App Store, surpassing ChatGPT for the first time. This shift is attributed to a 'QuitGPT' movement, where users are reportedly migrating to Claude due to Anthropic's public refusal to allow its models for mass surveillance or autonomous weapons, contrasting with OpenAI's controversial partnership with the Pentagon.

NVIDIA announced a $2 billion investment in Marvell Technology, forming a strategic partnership to integrate Marvell into NVIDIA's AI factory and AI-RAN ecosystem. This collaboration leverages NVIDIA NVLink Fusion to enable semi-custom AI infrastructure and advance silicon photonics for 5G/6G networks.

Google DeepMind announced the release of Gemini Ultra 2.0, a significant upgrade to its flagship multimodal AI model. The new version boasts enhanced reasoning capabilities across text, image, and video, demonstrating superior performance in complex scientific problem-solving and creative content generation benchmarks. Analysts predict its widespread adoption could accelerate advancements in various AI-driven applications.

Innovation Council Action, a new political group backed by former Trump aides and David Sacks, announced plans to spend over $100 million in the 2026 US midterms to support pro-deregulation candidates and advance the 'AI race against China.'

Industry reports indicate a growing gap between soaring AI budgets and actual ROI, with only 25% of enterprises successfully moving more than 40% of their AI pilots into full production due to challenges in proving tangible returns.

**Date:** 2026-03-29

**Date:** 2026-03-29

**Date:** 2026-03-28

**Date:** 2026-03-28

**Date:** 2026-03-28

**Date:** 2026-03-28

**Date:** 2026-03-27

**Date:** 2026-03-27

**Date:** 2026-03-27

**Date:** 2026-03-27

**Date:** 2026-03-26

**Date:** 2026-03-26

**Date:** 2026-03-26

**Date:** 2026-03-26

**Date:** 2026-03-25

**Date:** 2026-03-25

**Date:** 2026-03-25

**Date:** 2026-03-25

**Date:** 2026-03-25

**Date:** 2026-03-25

**Date:** 2026-03-25

**Date:** 2026-03-25

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-24

**Date:** 2026-03-23

**Date:** 2026-03-23

**Date:** 2026-03-23

**Date:** 2026-03-10

**Date:** 2026-03-10

Regulatory pressure forces Meta to allow third-party AI assistants inside WhatsApp in Europe.

**Item:** Open Question 1 — "Will AI models converge in capability?" (Status: Partial)

**Item:** Timeline mentions LLaMA release (Feb 2023) but misses the actual catalyst.

**Item:** Timeline entry #35 — "2024-12 — Context Windows: 128K-200K Becomes Standard"

**Gap:** The timeline covers model releases, funding rounds, and technical milestones but has almost nothing about AI's societal impact beyond two policy entries (Biden EO and UK Safety Summit).

**Item:** Open Question 13 — "Why don't we have reliable agents yet?" (Uses Moravec's Paradox framing)

**Item:** Timeline entries #38 and #39:

**Item:** Timeline entry #32 — "2024-05 — Gemini 1.5 Pro 1M Context Window"

**Gap:** The timeline shows ChatGPT (Nov 2022) but doesn't explain WHY it worked. InstructGPT and RLHF are the answer.

**Item:** Timeline entries #33 (June 2024) and #34 (August 2024) both list "xAI raises $6B"

**Item:** Open Question 8 — "Are hallucinations a fundamental problem?" (Status: Resolved — No)

**Item:** Open Question 3 — "Is RAG the right architecture for AI memory?" (Status: Resolved — Mostly Resolved)

Two foundational research milestones are missing from the timeline:

**Gap:** The timeline has zero entries for AI in scientific discovery. AlphaFold 2 is arguably the single most impactful real-world AI application to date.

**Gap:** Timeline lists GPT-1 (June 2018) but completely omits BERT (October 2018).

**Item:** Timeline entry "1969-01 — First AI Winter Begins"

**Status:** 🔴 Still Open | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

**Status:** 🔴 Still Open | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

**Status:** 🔴 Still Open | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

**Status:** 🔴 Still Open | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

**Status:** 🔴 Still Open | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

**Status:** 🔴 Still Open | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

**Status:** 🟡 Partially Resolved | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

**Status:** 🟡 Partially Resolved | **Inspired by:** [Dwarkesh Patel](https://www.dwarkesh.com/p/questions-about-ai)

If AI handles most coding, how do juniors learn?

Google's dominance was supposed to be threatened by AI chat. How's that going?

AI could go either way:

Opinions range wildly:

The biggest meta-question of this era.

**Status:** 🟡 Partially Resolved

**Status:** 🟢 Emerging Yes

**Status:** 🟡 Partially Resolved

**Status:** 🟡 Partially Resolved

**Status:** 🟢 Resolved — No

**Status:** 🟡 Partially Resolved

**Status:** 🟡 Partially Resolved

**Status:** 🟡 Partially Resolved

**Status:** 🟢 Resolved — Yes

**Status:** 🟢 Mostly Resolved

**Status:** 🟡 Partially Resolved

**Status:** 🟡 Partially Resolved

**Date:** 2026-02-28 | **Category:** Policy

**Date:** 1966-01-01