PicoBlog

When I signed my latest contract to write for Lived Places Publishing, I had to agree to not use AI-enabled content creation tools like Google’s Bard. I was happy to agree. Large Language Models are helpful in some superficial ways. But they’re absolutely garbage for research and academic writing.

Testing my thesis, that Google Bard is currently worthless, I asked a few questions related to my current research needs. I wanted to find research on autistic gestalt language processors not assimilating their language when migrating to another country. I’m aware of the topic of “language maintenance,” and how gestalt language processors hang on to the gestalts as they “recorded” them, as such, we tend to sound foreign for quite a lot longer than our typical processing migrant peers.

Bard (barf!) was happy to return results to my query. Unfortunately, it was entirely made up. Google features a button that allows users to search the entirety of Google to verify Bard’s results.

In the case above, Bard had hallucinated a research paper, title, authors’ names, and abstract. It did this not once, but thrice for that particular search. Nothing in the results was verified.

I changed up the search terms a bit and tried again. Again, complete garbage.

Certainly, Simon Baron-Cohen exists. But of the seeming thousands of articles with his name on it, none have the title returned by Bard.

I pressed on with more modifications. More garbage results.

I can find an Anabel Quan-Hasse who is an academic author, but no Annabel Quan. Is Bard hallucinating by randomly altering actual events, concepts, and people? Who knows.

A final tweak, I asked a more complex question asking it to opine based upon what it found in the research. Again, complete garbage. Thankfully, it has the disclaimer at the bottom of the screen that notes that it may display inaccurate information.

Not may … will.

ncG1vNJzZmiZpanAqrDEZ6qumqOprqS3jZympmegZLSwu8alnGaakaez