One of the great blessings of living through the generative AI chatbot revolution is getting to see massive tech corporations wage their eternal epic battles against one another. Bard is Google’s response to Microsoft’s highly-praised Bing Chat and OpenAI’s ChatGPT, though its general reception has been tepid at best.
Bing runs on GPT-4, the newest large language model from the creators of the ChatGPT (whose powerful comedic potential we wrote about here). But Google Bard runs on Google’s own large language model LaMDA, which had been in cautious development for years. You may remember LaMDA from that time a guy declared it sentient and got fired.
But all these technical details are ultimately tangential to the main question we care about: how can we use Google Bard to be funny?
TL;DR: Google’s Bard is the Google Glass of the 2020s.
As we put Google Bard through its paces, let’s start off with the first test of any aspiring short-form comedy writer: coming up with jokes, plain and simple.
These make sense logically, but that’s about it. If punchlines should be related enough to make sense but unexpected enough to be surprising, these are somehow both too obvious (calling a college student who can cook a “unicorn” makes logical sense but isn’t surprising) and too random (calling a college student who can cook a “unicorn” doesn’t relate the punchline to anything about college or cooking). Basically, Bard won’t be submitting Daily Show jokes anytime soon.
Google Bard does seem to have less reservations than ChatGPT–which refused to follow my request to make jokes at the expense of certain dogs’ appearances–when asked to make jokes about particular topics:
But again, these are at best underwhelming, and at worst nonsensical (how is a chihuahua in a snowstorm a dust bunny? If you can figure this out, please drop me a line).
Let’s give Bard the benefit of the doubt, though. In the real world, most of the time we’re writing to a specific tone and format, so jokes like this often don’t really land. Why don’t we throw Google Bard something a little more specific and, hence, easier–
Oof.
Yet again, Bard somehow comes up with punchlines that make a little bit of sense while also making no sense at all (“The Congressional hearings on TikTok were so ridiculous, even the TikTok users were laughing” exemplifies this almost-sense of Bard’s jokes).
“But surely,” you say, “this is but one example!” And you’d be right. So, here’s another:
0/2, Bard. And to think I gave you a second chance.
“But surely!” you say, intentionally repeating yourself to show your indignation, “this is an impossible task for a generative AI! They’re chatbots, not people!”
And to that, I would mournfully present the results of the very same prompt given to ChatGPT.
Granted, these aren’t winners, but they’re orders of magnitude funnier and more creative than Bard’s. More importantly, they (mostly) involve comedic ideas, or at least provide a unique set of jumping-off points for the writer to fix up the punchline, setup, or both.
Alright, let’s put aside all of that joke writing stuff above. Bard wasn’t created to be a comedy writer, and that’s fine. It was created to help as a brainstorming tool, so let’s use it that way to work on some sketches:
There’s a little bit here–#4 seems somewhat fruitful to me–but on the whole, it’s not particularly comedic. This tracks with Bard’s other reviews, which have referred to it as “dry and uncontroversial”, likely to the detriment of its comedic instincts.
Again, these types of chatbots are not specifically designed to have the best idea of what’s funny and what’s not, but ChatGPT still provides much more specific juxtapositions that offer more comedic ideas right off the bat.
Some of these are honestly pretty good ideas for sketches (or at least beats of sketches). It seems that Bard is able to identify the ideas behind an ironic juxtaposition, but it isn’t able to select for those particular ironies that get a laugh (or at least a wry “that’s funny”).
But what about once you’ve already got an idea? Can Bard help you come up with the actual execution of the sketch?
To Bard’s credit, it did come up with things that could happen in my sketch idea, but the problem here is that it didn’t really use the humor of the original idea – it reads more like a drama than comedy, and while it escalates in intensity of plot, it doesn’t escalate in terms of the comedic behavior (it doesn’t escalate the game).
I really don’t mean to beat the point to death, but again, ChatGPT does a much better job in this respect:
So, it seems that while ChatGPT has some instinctive understanding of comedic game and escalation, Bard lacks that understanding, and instead focuses solely on plot.
As we discussed in our post on ChatGPT, one way to use generative AI as a writing tool is to come up with some funny linguistic ideas based on a particular tone or point-of-view.
Unfortunately for Google CEO Sundar Pichai, Bard continues to falter in terms of tone and voice help (since we’re sure comedy writing was his main intention).
Bard doesn’t integrate information about the pirate’s prior knowledge with the Woodstock festival’s specifics, but rather just crudely replaces regular Woodstock words with pirate words or, worse, just shoves pirate phrases after normal phrases.
Essentially, Bard is able to regurgitate facts or talk in a specific tone, but it can’t really integrate the two.
Bard’s only real ability over ChatGPT for comedy is the same one that it has for other disciplines: it’s totally up-to-date.
This is ChatGPT’s Achilles heel.
A feature which Bard surpasses with flying colors.
For comedy writers (and, I suspect, for most users), there’s the real value of Bard over ChatGPT. It is continually up-to-date with everything on the Internet, integrating current information in real time.
However, Bard’s present information is not infallible (and can even be outright wrong), as I discovered while trying to find news stories for monologue jokes. Bard seems to have a somewhat loose relationship with time and truth.
Bard even began to contradict its own answers once pressed to provide accurate information:
Bard’s immediate and apologetic retractions may be the funniest part about it. This leads me to conclude that, while Bard might be an alright choice for up-to-date information regarding extremely specific questions, even for topical jokes you’re probably better off with ChatGPT, Bing Chat, or just Googling “odd news”.
Poor underdog Google. I hope you bounce back soon!
–If you’re interested in how AI can help comedy writers, check out our related article on ChatGPT (we didn’t know how good we had it!). Or, if you’re looking for a good Late Night comedy writing manual, we recommend Joe Toplyn’s book! Or, for more information about our work at Hello SciCom, check us out here.–
