Sources: 
Google's AI struggles with spelling accuracy, recently misspelling 'Trump' as 't-r-p-u-m.' As counting letters in words remains a persistent issue for large language models (LLMs), Google acknowledges this challenge, stating,
"Counting within words has been a known challenge for LLMs". The limitations of the token-based architecture used by LLMs contribute to these spelling issues.
Moreover, Google's AI previously experienced a significant 'disregard' issue, where it would ignore instructions in its AI overview responses. Fortunately, as of today, searching for 'disregard' yields a traditional featured snippet with its definition, resolving the previous confusion. These developments highlight the ongoing struggles and adaptations of AI technologies as they strive for improvement in language processing tasks.
The challenges faced by AI in accurately perceiving language units underscore the limitations inherent in their tokenization approach. As one expert noted,
“My guess would be that there’s no such thing as a perfect tokenizer due to this kind of fuzziness.” However, Google remains dedicated to addressing these issues.
Despite the setbacks, Google's efforts to refine its AI's spelling capabilities continue as they navigate the complexities of natural language processing.
Sources: Google's AI has been criticized for spelling errors, notably rendering 'Trump' as 't-r-p-u-m.' Moreover, Google's AI has faced challenges in adhering to instructions, particularly regarding the term 'disregard,' which previously generated incorrect responses, but now offers a standard definition instead.
