Transformer Co-Creator: I’m ‘Absolutely Sick’ of the Tech | Microsoft Overhauls Copilot & Enterprise AI Faces Leadership Crisis

Transformer Co-Creator: I’m ‘Absolutely Sick’ of the Tech | Microsoft Overhauls Copilot & Enterprise AI Faces Leadership Crisis

A frustrated AI developer looking at complex AI algorithms and a Microsoft Copilot interface, symbolizing the current challenges and overhauls in enterprise AI.

Key Takeaways

  • A pioneer of the transformer architecture, Llion Jones, declared he’s abandoning the dominant AI tech due to dangerously narrow research and calls for exploring new breakthroughs.
  • Microsoft unveiled a massive Copilot update with 12 new features, including a character “Mico,” collaborative “Groups,” deeper OS integration, and a strategic pivot to its own MAI models.
  • Writer AI CEO May Habib warned that nearly half of Fortune 500 executives believe AI is “tearing their company apart,” blaming leaders for delegating transformation to IT.
  • OpenAI’s GPT-5 was quietly revealed to be powering a new multi-agent research assistant, signaling continued advancement in core models.

Main Developments

Today’s AI news paints a picture of an industry grappling with profound contradictions: a leading architect of the technology powering every major AI model is abandoning his own creation, while simultaneously, established tech giants are making sweeping advancements built upon that very foundation, and businesses struggle to integrate the new paradigm.

In a striking act of self-critique, Llion Jones, co-author of the seminal “Attention Is All You Need” paper and coiner of “transformer,” delivered a candid assessment at the TED AI conference. He declared he is “absolutely sick” of transformers and drastically reducing his work on them, arguing that AI research has become dangerously narrow. Jones, now CTO of Sakana AI, warned that unprecedented investment is stifling creativity, leading to an “exploitation versus exploration” paradox where researchers chase incremental gains on existing architectures, potentially missing the “next big thing” just around the corner. His call to turn up the “explore dial” and openly share findings challenges the industry’s competitive, fast-paced environment.

This critique arrived as Microsoft unveiled its ambitious Copilot 2025 Fall Update, a testament to the ongoing power and practical application of generative AI. Headed by Mustafa Suleyman, Microsoft’s AI division CEO, the update introduces 12 significant features, deepening Copilot’s integration across Windows, Edge, and Microsoft 365. Highlights include “Mico,” an expressive AI assistant character reminiscent of Clippy, real-time collaborative “Groups” for up to 32 participants (a direct challenge to Anthropic and OpenAI’s team workspaces), and specialized tools like “Copilot for Health” and “Learn Live.” Notably, Suleyman emphasized a strategic shift towards Microsoft’s homegrown MAI models, such as MAI-Voice-1 and MAI-Vision-1, alongside those from OpenAI. This move signals Microsoft’s intent to position Copilot as a contextual AI infrastructure, driven by “the best models – both those we build and those we don’t.” OpenAI quickly responded by expanding its Shared Projects feature to all users, highlighting the escalating competition among these “frenemies.”

Microsoft CEO Satya Nadella reinforced this enterprise-focused vision in his latest shareholder letter. He outlined five key takeaways for technical decision-makers, emphasizing security and reliability, a hybrid and sovereignty-ready AI infrastructure, a shift towards AI agents (not just chatbots), the necessity of unified data platforms like Microsoft Fabric, and mandatory responsible AI practices. Nadella’s letter positions Microsoft as the builder of the foundational AI infrastructure for decades to come, moving beyond mere productivity tools to full operational AI.

Yet, despite these advancements and strategic blueprints, many enterprises are struggling. May Habib, CEO of Writer AI, revealed at the same TED AI conference that 42% of Fortune 500 executives believe AI is “tearing their company apart.” Habib blames leadership for making a “category error,” delegating AI transformation to IT departments rather than driving it strategically. She argued that AI inverts the traditional model where execution is expensive, making it programmatic and abundant. True “AI-first leaders” must take a “machete to complexity,” manage the fear of changing career ladders, and embrace unbounded ambition. For CIOs, their role shifts from gatekeeper to enabler, building the “stadium” for AI agents while business leaders design the “plays.”

Amidst this blend of profound critique, aggressive productization, and organizational friction, a subtle yet significant detail emerged: OpenAI’s GPT-5 is already being deployed. The research assistant Consensus announced it now uses GPT-5 and OpenAI’s Responses API to power a multi-agent system, accelerating scientific discovery for millions of researchers. This quiet reveal underscores that while the debate over AI’s future architecture rages and enterprises face adoption challenges, the underlying models continue to evolve and become more powerful.

Analyst’s View

The AI landscape today is a fascinating study in contrasts. On one hand, Llion Jones’s urgent call to broaden research beyond transformers echoes a growing sentiment that the field might be prematurely optimizing within a local maximum. His perspective, coming from a co-creator, lends significant weight to the argument that true breakthroughs require exploring genuinely novel architectures, not just scaling existing ones. Simultaneously, Microsoft’s comprehensive Copilot rollout and Nadella’s strategic letter demonstrate the immense potential and ongoing commercialization of current AI, with a clear focus on enterprise integration and Microsoft’s own growing model capabilities. The quiet emergence of GPT-5 further confirms that model advancement continues unabated. However, May Habib’s stark warning about enterprise failures highlights the critical gap: technology alone isn’t enough. The next frontier in AI may not just be about smarter algorithms, but about smarter, more adaptable organizational leadership capable of dismantling complexity and navigating profound cultural shifts. The industry must balance radical exploration with responsible, business-led adoption.

Source Material

阅读中文版 (Read Chinese Version)

Comments are closed.