Editorial illustration for Nvidia CEO Jensen Huang says AI stops hallucinating, then hallucinates himself
Nvidia CEO: AI Hallucinations Remain Years Away
Nvidia CEO Jensen Huang says AI stops hallucinating, then hallucinates himself
Nvidia CEO Jensen Huang sat for a CNBC interview that aired April 25. He made a striking claim: generative artificial intelligence has stopped making things up. The technical term is “hallucination.” It remains a core, unsolved flaw in large language models.
No research breakthrough has been announced to eliminate it. Huang’s statement lands as his biggest chip buyers—Meta, Amazon, and Google—face investor questions. Those companies have all signaled they will pour tens of billions more dollars into AI infrastructure this year.
Huang literally says: "AI became super useful, no longer hallucinating."
Large language models work by predicting the next most likely word. That very architecture makes them prone to confabulation. Researchers at Google and OpenAI continue to list reducing hallucinations as a primary technical challenge.
For investors, the timing is everything. Huang’s statement arrived alongside quarterly reports detailing the staggering scale of planned spending by the cloud giants, Nvidia's primary customers.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic NVIDIA and Google Cloud let developers scale AI from prototype to production
Market Trends
Related Topic NVIDIA NeMo powers telco reasoning model for autonomous network workflows
Industry Applications
From Archives Claude’s initial constitution includes DeepMind’s Sparrow anti‑racist statements
From Archives Anthropic unveils Claude Opus 4.6 with multi‑agent code and large context window
From Archives NVIDIA's NVFP4 Training Recipe Boosts AI Speed and Cuts Costs
Industry Applications
From Archives Build Kimi K2.5 Multimodal VLM with NVIDIA GPU-Accelerated Endpoints
Common Questions Answered
What did OpenAI claim about citation hallucinations in GPT-5?
[nature.com](https://www.nature.com/articles/d41586-025-02853-8) reports that OpenAI claimed to have reduced the frequency of fake citations in GPT-5. The company specifically noted improvements in reducing 'hallucinations' and 'deceptions' where AI previously claimed to have performed tasks it hadn't actually completed.
Why are AI hallucinations difficult to completely eliminate?
[nature.com](https://www.nature.com/articles/d41586-025-00068-5) explains that large language models are trained to predict tokens from text corpora, which means factual knowledge is implicitly stored in model parameters rather than an explicit fact database. This inherent design makes it challenging to completely prevent hallucinations, especially for less common or more nuanced information.
What types of hallucinations do large language models typically produce?
[arxiv.org](https://arxiv.org/abs/2510.06265) describes hallucinations as AI-generated content that is fluent and syntactically correct but factually inaccurate or unsupported by external evidence. These hallucinations can range from slightly misremembered facts to completely fabricated references, undermining the reliability of AI systems in domains requiring high factual accuracy.
Further Reading
- Papers with Code - Latest NLP Research — Papers with Code
- Hugging Face Daily Papers — Hugging Face
- ArXiv CS.CL (Computation and Language) — ArXiv
