Dutch AI Experts Sound Alarm as ChatGPT Performance Hits Plateau

2026-03-28 data

Amsterdam, Saturday, 28 March 2026.
Leading Dutch technology analysts warn that ChatGPT’s declining performance may signal we’ve reached artificial general intelligence’s limits. The system now produces hollow phrases and meaningless metaphors, with journalists identifying tell-tale AI writing patterns like ‘it is not this, but that’ constructions. This plateau coincides with Nvidia CEO Jensen Huang’s controversial claim that AGI has been achieved, creating uncertainty for Dutch innovation companies relying on AI advancement for their business strategies.

The Hollow Phrase Problem

Dutch journalists have identified a concerning pattern in AI-generated content that reveals the depth of ChatGPT’s performance decline. According to analysis published in De Volkskrant on March 27, 2026, current AI systems are producing increasingly empty responses due to over-training on existing data [1]. American journalist Jasmine Sun, writing in The Atlantic, documented the specific linguistic markers of this degradation: “Chatbots produceren betekenisloze metaforen, eindeloze ‘het is niet dit, maar dat’-constructies en een weeïge, kruiperige toon – en natuurlijk gebruiken ze mijn geliefde gedachtestreepje te veel” (Chatbots produce meaningless metaphors, endless ‘it is not this, but that’ constructions and a sickly, sycophantic tone – and of course they use my beloved dash too much) [1]. This linguistic decay has become so pronounced that NRC journalist Maarten Reijnders can now identify GPT-generated writing by spotting phrases like “Solidariteit is geen abstract gebod, maar een relationeel fenomeen” (Solidarity is not an abstract commandment, but a relational phenomenon) [1]. The phenomenon has even affected public discourse, with politician Ehsan Jami forced to deny using AI after his writing was suspected of being GPT-generated [1].

AGI Claims Amid Performance Concerns

The plateau in ChatGPT performance comes at a paradoxical moment when industry leaders are claiming artificial general intelligence has been achieved. On March 24, 2026, Nvidia CEO Jensen Huang declared in a podcast interview with computer scientist Lex Fridman: “Ik denk dat het nu is. Ik denk dat we AGI hebben bereikt” (I think it’s now. I think we have achieved AGI) [2][4]. Huang’s assertion centered on the capabilities of OpenClaw, an open-source AI agent platform designed to work autonomously on behalf of users, which is currently being acquired by OpenAI [2]. However, Huang himself tempered his claims, noting that while AI agents might create viral applications, “De kans dat 100.000 van die agents een bedrijf als Nvidia zullen opzetten, is nul procent” (The chance that 100,000 of those agents will build a company like Nvidia is zero percent) [4]. This contradiction between AGI proclamations and acknowledged limitations highlights the problematic nature of AGI as what many consider merely a marketing term [1].

Industry Response and Technological Pivots

The AI plateau is prompting major technology companies to recalibrate their strategies and investments. OpenAI announced on March 27, 2026, that its Foundation will deploy at least $1 billion in the coming year toward life sciences, economic opportunities, AI safety, and community programs, as initial steps toward a $25 billion commitment [5]. Key appointments include Wojciech Zaremba, co-founder of OpenAI, joining as Head of AI Resilience, and Anna Makanju as Head of AI for Civil Society and Philanthropy [5]. Meanwhile, hardware manufacturers are preparing for the post-LLM era with specialized solutions. Arm has launched its AGI CPU, designed specifically for AI datacenter workloads, with systems capable of housing up to 45,000 cores per rack in high-end configurations [3]. Meta has emerged as both a major customer and co-developer of these new processors, integrating them alongside their custom AI accelerators [3]. The pivot toward specialized hardware reflects industry recognition that current large language model architectures may have reached their performance ceiling.

Implications for Dutch Innovation Ecosystem

For Dutch technology companies and startups, the AI plateau presents both challenges and opportunities that could reshape the innovation landscape. SAP’s approach, as outlined by executives in March 2026, demonstrates how established software companies are adapting by building proprietary AI platforms that combine large language models with deep enterprise data [7]. The company’s strategy emphasizes that “Data is de brandstof, maar de context van een bedrijf is de motor” (Data is the fuel, but a company’s context is the engine), suggesting that competitive advantage will increasingly come from domain-specific AI applications rather than general-purpose systems [7]. Dutch companies working in the AI sector may need to shift focus from relying on continuous improvements in foundational models like ChatGPT toward developing specialized AI solutions tailored to specific industries or use cases [GPT]. This transition could actually benefit smaller Dutch innovation firms by leveling the playing field, as success will depend more on understanding specific business contexts and less on access to the most advanced general AI models. However, companies that have built their strategies around the assumption of continuous AI advancement may need to reassess their technological roadmaps and investment priorities.

Bronnen

artificial intelligence ChatGPT development