news.today
Hello · Register today | LoginHello · My Account[—] [32]
Tech

AI image generators close in quality, while chatbots struggle with accuracy

Summarised from 2 outlets · Updated 15 Oct, 00:20 · Archive

Google's Nano Banana 2.1 image generator rivals ChatGPT's competitor, but AI chatbots frequently provide inaccurate information.

Google has released an upgraded version of its Nano Banana image generator, now integrated into Google Gemini. TechRadar tested Nano Banana 2.1 against ChatGPT's Images 2.5 by giving both identical prompts and comparing the results. The two generators proved surprisingly evenly matched, with both handling complex tasks like lighting, reflections and text rendering competently, though each approached instructions with different artistic instincts.

Meanwhile, a fact-checking study by Full Fact tested five AI chatbots over ten weeks from mid-July, asking them identical questions daily. Claude, ChatGPT, Gemini, DeepSeek and Grok were questioned on topics including voting requirements and migrant arrivals. The report found the chatbots often gave correct answers but also confidently provided wrong responses. One example cited Gemini incorrectly reporting 12,210 small boat arrivals to the UK in the first half of 2026 after making a calculation error. Another found a chatbot falsely asserting that Sir Keir Starmer was still UK prime minister in September, weeks after his resignation.

Full Fact noted that chatbots sometimes calculated incorrect figures rather than citing official statistics, presented outdated information as current, and answered the same questions differently on different days. The report said responses were often 'incomplete, lacking context, or poorly sourced'.

AI image generators close in quality, while chatbots struggle with accuracy
Image: The Independent

Comments (0)

How it is being reported

In this story: Google · OpenAI · Full Fact · ChatGPT · Sir Keir Starmer

This summary was written by AI from the headlines and standfirsts above, and states as fact only what at least two outlets report. How we use AI · Report a problem