Many of you will know we have an email list, The Week in Botany. The idea was to promote blog posts on Botany One, but we wanted to do something a bit more than simply mail a list of posts to add value to subscribers. So, we turned to Twitter.
People were sharing a lot of links on Twitter. We thought there was an opportunity for a Wisdom of Crowds approach to finding what interesting Botany was happening. Using a system called Nuzzel, we could scan all the people following us, and it would send daily alerts showing the most shared links. Taking two news stories and two scientific papers each day helped produce the newsletter, meaning that you got a snapshot of the wider botanical community and not just what was happening here. The reason we’re running a new experiment now is that we’ve had to adapt following changes to Twitter.
What’s the problem with Twitter?
Twitter bought out the company that made Nuzzel and shut it down – promising that the features would be rolled into Twitter. Maybe that will happen someday. We started using a few tools that were less targeted on our particular interests. They worked until shortly after the Musk takeover when changes to the API shut those systems down.
Since then, I’ve been using a sample of Twitter accounts and manually counting shares and favourites. This method isn’t so accurate, but has also highlighted another problem. Share counts are falling.
In the past, a link would have thirty shares or more to be in the email. It dropped after the API changes to fifteen or more. Since the rebrand to X, share counts have fallen further. Some tweets still get large numbers of shares, but now a tweet can stand out with fewer than ten retweets. This decline might be partly due to a summer slump, but we are getting close to any signal being close to the noise of tweets that two or three retweets. At that point, selecting a few for the email becomes entirely arbitrary.
It would make sense to look at what gets shared on other networks. Facebook is difficult to track, but Mastodon is not. However, Mastodon has some issues of its own.
What’s the problem with Mastodon?
Mastodon’s biggest problem, from the email’s point of view, is that there’s not the same volume of botanical links shared there as there is on Twitter. I’ve been scanning Mastodon for links, but most weeks it provides just one or two.
One reason for this is that botanical people can be hard to find. Mary Williams has a spreadsheet of plant people accounts, but there are probably many more around. Some links may be getting shared that aren’t very well connected to the wider community.
The big advantage Mastodon has over Twitter is that while the community is smaller, it’s much more engaged, and the Mastodon system doesn’t penalise messages for including links. So if links were to appear, there’s a chance they’d get more engagement, and it might be possible to sift through the boosts and likes to discover the most interesting links – which connects nicely with a personal grouch I have.
There is no shortage of interesting Botany, but there is a shortage of time to write about it
I have plenty of Google Scholar Alerts set up to find potentially exciting papers. When I get an email, I read them and add the most interesting articles to a database. This database expands at the rate of ten papers a day on average. There’s no way they could all be blogged, which means a lot of good botany doesn’t get promoted by us.
Tweeting it would have made sense in the past, but these days sharing links on Twitter has problems. Twitter penalises messages that include links because they want to keep you on-site looking at adverts. It takes time to produce tweets that offer more information than just the paper title and link address. It feels futile to do more work for the world’s richest man, knowing he will penalise you for it.
Mastodon has longer messages, so posts there could include the title, link and a micro-summary of the paper. But this writing also takes time. If I spend fifteen minutes reading the article, writing the summary and posting it, that’s still over two hours a day gone from each working day.
What we need is a system that can summarise long texts quickly. Our work here with AI language models has shown they cannot write long texts accurately, but something as short as a 500-character message should be perfectly achievable.
The Mastodon Experiment
When I add a paper to the blogging database, I’ll also ask GPT-4 to create a short Mastodon message based on the title and abstract of the article. I’ll use a model that doesn’t use the input for further training.
I’ll add the message to Buffer to appear on our account @email@example.com, with the tags #Botany so that people can find it and #BotanyAI. This second tag is because people might want to follow the #Botany tag without seeing a message posted from this site every two hours. Mastodon has the option to mute tags. In this case, tagging the messages #BotanyAI allows people to mute any messages containing that tag without having to avoid any other messages tagged #Botany.
I’ve been doing this for a week just to see if it’s manageable and if it helps identify interesting links to add to the email. Results from the first week show that sharing so many links is helpful for us. I’d guess that up to 40% of the news and scientific papers links in this week’s email were sourced from Mastodon. It’s a guess, as I didn’t keep tags on the sources. I will do this week.
What could have worked better was the message format. I asked the AI to write everything, including a short, snappy title for the messages. For future messages, I’ll use the paper titles. The paper titles are often long and sometimes flabby, but often they’re not and can have important nuance that the AI sometimes missed when writing its own titles.
What I hope to get out of it
I’m not an expert on all areas of botany. Sharing these papers with a wider audience allows me to tap into the expertise of many more people who know much more about plant science than I do. That’s not just true of Mastodon. It’s true of Twitter, too – but changes at Twitter are cutting down that flow of feedback while it’s growing from Mastodon.
This experiment has already had some positive effects. In a few cases, Mastodon users have responded to posts pointing out things that I might have overlooked if I read them just by myself.
I’d also like to be able to replace Twitter with Mastodon (and maybe Threads) as a source of shared papers for TWiB. Botany One will stick with Twitter for now, but it looks like the site is sinking. It makes sense to prepare for what’s next before Twitter evaporates. This change will take time as one community waxes and the other wanes, so I won’t expect a switch from using purely one site to using another instantly.
I also hope it will spur more people into sharing papers on Mastodon as a routine thing. A few people do at the moment, but selfishly, I’d like to see more interesting papers that I didn’t find myself.
You can keep track of what I’m finding at The Week in Botany, which you can sign up for at https://botanyone.substack.com/