You know how Google’s new feature called AI Overviews is prone to spitting out wildly incorrect answers to search queries? In one instance, AI Overviews told a user to use glue on pizza to make sure the cheese won’t slide off (pssst…please don’t do this.)
Well, according to an interview at The Vergewith Google CEO Sundar Pichai published earlier this week, just before criticism of the outputs really took off, these “hallucinations” are an “inherent feature” of AI large language models (LLM), which is what drives AI Overviews, and this feature “is still an unsolved problem.”
Genuinely curious, what pieces do you suggest we can keep from LLM/GenAI/etc?
?
Have you never used any of these tools? They’re excellent at doing simple things very fast. But it’s like a word processor in the 90s. It’s just a tool, not the font of all knowledge.
I guess younger people won’t know this, but word processor programs were very impressive when they first came out. They replaced typewriters; a page printed from a printer looked much more professional than even the best typewriters. This lent an air of credibility to anything that was printed from a computer because it was new and expensive.
Think about that now. Do you automatically trust anything that’s just printed on a piece of paper? No, because that’s stupid. Anyone can just print whatever they want. LLMs are like that now. They can just say whatever they want. It’s up to you to make sure it’s true.
Font of all knowledge sounds like an excellent font. I assume it’s serifed?
snort
facepalm
The only good response! 😄
The main field where they are already actively in professuonal use are rough drafts in creative fields: quickly generate possible outlines for a text, a speech, an art piece. Visualize where something could be going, in order to decide which direction to pick.
Also, models that work differently from the GPTs are already in use in science, scanning through huge amounts of texts in archives to help analyzing or search for something in particular. Help find patterns in things for studies. Etc.
The “personal assistant AI” thing obviously isnt quite working yet. I think it will take some time and models with a different technological structure (not GPT) to achieve progress in that regard.
Using it to generate things that you double check. Transforming generative work to review work is a boost in productivity. So writing of any kind, art, etc. asking the llm for facts without context is a gross mistake. Prompting it to generate a specific paragraph in a larger, technical or regulator document is useful.