Hallucinations have commonly been considered a problem for generative AI, with chatbots such as ChatGPT, Claude, or Gemini prone to producing ‘confidently incorrect’ answers in response to queries.
This can pose a serious problem for users. There are several cases of lawyers, for example, citing non-existent cases as precedent or presenting the wrong conclusions and outcomes from cases that really do exist.
Unfortunately for said lawyers, we only know about these instances because they’re embarrassingly public, but it’s an experience all users will have had at some point.
For enterprises, there are hopes that LLMs trained on proprietary data – and only proprietary data – may be less prone to these hallucinations than public chatbots, but the need to manually check outcomes remains in cases where accuracy is key.
At the 2025 Huawei Connect conference, however, Tao Jingwen, director of Huawei’s quality, business process & IT management department, suggested that this is a wrongheaded way to look at the technology. Instead, Tao said that businesses should embrace hallucinations as part and parcel of generative AI.
Speaking at the manufacturing summit within the conference, Tao spoke about some of the challenges facing the sector when it comes to using AI.
“AI hallucinations and the black box nature of AI make it challenging for businesses and enterprises, especially businesses from the manufacturing sector, to trust and control, raising new issues around predictability and explainability,” he told delegates.
“From my point of view, well first of all we need to embrace AI hallucinations,” Tao added. “Without hallucinations, AI wouldn’t be what it is today. But there’s still a need to find effective ways to control and mitigate hallucinations.”
Tao also spoke of the challenges facing manufacturers that want to integrate AI into their businesses successfully and usefully when they already have decades of digitalization, automation, and refined processes in place.
“AI implementation requires collaboration across business, IT, data, and other teams for effective implementation, but such a magic box from AI has brought new challenges to us,” Tao said.
“Especially in manufacturing after years of digitalization and process, how can we better integrate AI with good results?”
AI hallucinations aren’t going anywhere
Tao’s comments on hallucinations come amid an increasing resignation to – rather than embrace of – the fact these erroneous outputs are unavoidable.
In a paper published on 4 September 2025, OpenAI researchers said that hallucinations were a fact of how most LLMs have been trained and that they are “inevitable” for base models.
More specialist training that has a more focused data set and that penalizes incorrect outputs can help mitigate this, researchers noted.
Ultimately, though, businesses would need to decide if it’s worth the effort of training an LLM in this way or if there are other workarounds that could be used to, in the words of Tao, “control and mitigate” the problem.
Make sure to follow ITPro on Google News to keep tabs on all our latest news, analysis, and reviews.
MORE FROM ITPRO
- Generative AI attacks are accelerating at an alarming rate
- Only a handful of generative AI projects make it into production – here’s why
- Generative AI enthusiasm continues to beat out business uncertainty
-
Advania UK eyes further growth under new CEO James HardyNews Hardy will lead Advania’s UK business as it targets the underserved mid-market with integrated IT services
-
Co-op chief executive ‘very proud’ of cyber attack response despite huge financial lossesNews The company says operating profits are £80 million down, and that the financial effects may continue into the second half
-
The rise of the UK’s ‘invisible’ AI workforceNews Research from Multiverse shows AI skills are becoming common in non-tech roles
-
Why Nvidia’s $100 billion deal with OpenAI is a win-win for both companiesNews OpenAI will use Nvidia chips to build massive systems to train AI
-
Shadow AI can be a tool for AI innovation with the right controls, say Gartner analystsNews Data-driven messaging and a supportive approach to securing AI tools are necessary for security staff looking to balance AI risks and unlock better funding
-
Zoom CEO Eric Yuan thinks AI will pave the way for a three-day week – and he’s not the only big tech exec excited about reduced working hoursNews Yuan joins Nvidia CEO Jensen Huang and Bill Gates in touting the potential for AI to unlock reduced working hours
-
DeepSeek’s R1 model training costs pour cold water on big tech’s massive AI spendingNews Chinese AI developer DeepSeek says it created an industry-leading model on a pittance
-
Global AI spending is set to hit $2 trillion next year – here’s where all the money is goingNews Huge spending increases in AI services and associated infrastructure are expected
-
OpenAI just revealed what people really use ChatGPT for – and 70% of queries have nothing to do with workNews More than 70% of ChatGPT queries have nothing to do with work, but are personal questions or requests for help with writing.
-
This DeepSeek-powered pen testing tool could be a Cobalt Strike successor – and hackers have downloaded it 10,000 times since JulyNews ‘Villager’, a tool developed by a China-based red team project known as Cyberspike, is being used to automate attacks under the guise of penetration testing.
