Business EconomyPosted at: Sep 7 2025 11:07PM


Hallucinations persist partly because current evaluation methods set the wrong incentives: OpenAI on AI hallucination

New Delhi, Sep 7 (UNI) GPT-5 has significantly fewer hallucinations, especially when reasoning, but they still occur, said OpenAI in its research paper themed ‘Why language models hallucinate.’
‘Hallucination’ in relation to AI tools refer to the situation when an AI model produces outputs that are inaccurate, misleading, or entirely fabricated, despite often presenting them with high confidence.
Sharing insights on hallucination, the OpenAI research paper said “Hallucinations persist partly because current evaluation methods set the wrong incentives. While evaluations themselves do not directly cause hallucinations, most evaluations measure model performance in a way that encourages guessing rather than honesty about uncertainty.”
“Think about it like a multiple-choice test. If you do not know the answer but take a wild guess, you might get lucky and be right. Leaving it blank guarantees a zero. In the same way, when models are graded only on accuracy, the percentage of questions they get exactly right, they are encouraged to guess rather than say, “I don’t know.”
The research paper also noted that the older OpenAI o4-mini model performs slightly better. However, its error rate (rate of hallucination) is significantly higher.
The ‘OpenAI o4-mini’ is a compact, cost-efficient and high-throughout reasoning model from OpenAI’s ‘o-series.’ This model is specifically designed for applications requiring fast and budget-friendly AI performance including high-volume tasks.
“When averaging results across dozens of evaluations, most benchmarks pluck out the accuracy metric, but this entails a false dichotomy between right and wrong. On simplistic evals like SimpleQA, some models achieve near 100% accuracy and thereby eliminate hallucinations. However, on more challenging evaluations and in real use, accuracy is capped below 100% because there are some questions whose answers can’t be determined for a variety of reasons, such as unavailable information, limited thinking abilities of small models, or ambiguities that need to be clarified.” OpenAI said.
Pointing towards a solution, OpenAI said there is “no straightforward fix” for this problem. The research paper noted the key finding that “ Accuracy will never reach 100% because, regardless of the model size, search and reasoning capabilities, some real-world questions are inherently unanswerable.”
OpenAI, a US-based AI research and deployment company that creates and promotes artificial intelligence (AI).

UNI SAS RKM
Please log in to get detailed story.
Tags: #Hallucinations persist partly because current evaluation methods set the wrong incentives: OpenAI on AI hallucination
More News

Market valuation of five of top-10 most valued firms spikes by Rs 1 54 lakh crore

19 Jul 2026 | 12:09 PM

New Delhi, Jul 19 (UNI) In a week, the market valuation of five of the top-10 most valued firms jumped Rs 1.54 trillion. IT major TCS emerged as the biggest winner, in line with a positive trend in equities. TCS added Rs 72,072.3 crore, taking its market valuation to Rs 8,20,672.70 crore.

see more..

AP: India's largest fashion destination, 'TRENDS' opens store

18 Jul 2026 | 8:14 PM

Guntur, July 18 (UNI) India's largest and fastest growing apparel and accessories specialty chain of Reliance Retail, TRENDS, launched its 7th new Store in Amaravati Road at Guntur in Andhra Pradesh.

see more..

Revanth Reddy announces land for FICCI, reaffirms push for women-led entrepreneurship

18 Jul 2026 | 6:58 PM

Hyderabad, July 18 (UNI) Telangana Chief Minister A Revanth Reddy on Saturday reiterated his government's commitment to women-led development, announcing the allocation of 1,000 square yards of land near Shilparamam in Hitech City for the Federation of Indian Chambers of Commerce and Industry (FICCI) to establish its headquarters and one acre in Bharat Future City to make the organisation a partner in the State's flagship development project.

see more..

Kotak dismisses margin pressure concerns after 26 pc Q1 profit jump

18 Jul 2026 | 5:43 PM

Mumbai, July 18 (UNI) Kotak Mahindra Bank on Saturday sought to allay investor concerns over the quality of its June-quarter earnings after reporting a 26 per cent rise in standalone net profit to ₹4,123 crore, saying its net interest margins (NIM) had stabilised following the Reserve Bank of India's repo rate cycle and that earnings were driven by underlying operating strength rather than merely lower provisioning.

see more..

Kotak Mahindra Bank Q1 profit rises 26 pc as NPAs fall

18 Jul 2026 | 4:17 PM

Mumbai, July 18 (UNI) Kotak Mahindra Bank reported a strong start to the financial year with a 26 per cent jump in standalone net profit to ₹4,123 crore in the first quarter of FY27, supported by robust credit growth, higher deposits and a sharp improvement in asset quality.

see more..