Text Generation and Machine Learning: What Will Google Learn From Its Feature Announcement? And What Will the New Language Model Tell Us About Us?
Google deserves recognition for reining in generative chatbots’ wild side like that. But in my tests, the new search interface felt incredibly tame compared to ChatGPT or Google’s own chatbot Bard.
As it prepared the experimental search feature announced yesterday that responds to search queries with chat style answers, it worked hard to tone down the chaotic streak of text generation technology.
The new technology is very new, it has challenges, and will make mistakes, according to the vice president of search at Google who gave WIRED a preview of the new features.
A machine learning model is used to predict the meaning of a string of text and digest large amounts of text on the internet. More training that humans give the bot made it good at answering questions and holding a conversation.
The smart version of search is narrow-minded, and won’t even use the first person. It completely avoids topics that might be considered risky, refusing to dispense medical advice or offer answers on potentially controversial topics such as US politics.
Even if the warning about faulty artificial intelligence is overblown, it will be interesting to watch how companies balance development of more powerful generative language models with the need to have them behave.
In March, some big names in Artificial Intelligence research signed an open letter to call for a pause in the creation of machine learning systems. Pichai was not a signatory and said in his keynote speech yesterday that the company is currently training a new, more powerful language model called Gemini.
A source at Google tells me this new system will incorporate a range of recent advances from different large language models and may eclipse GPT-4. But don’t expect to get to experience the full power or charisma Gemini can offer. It could appear like another clever autocomplete if the same chaos-taming methods are applied to it.