The internet's largest chat app is going all-in on AI. At an "AI day" virtual event hosted by Discord for press, the company announced three experimental AI tools soon entering testing on select servers. Chief among them is a bot utilizing OpenAI's prolific chat interface, ChatGPT, that Discord one day plans to integrate into every server.
Discord's ChatGPT bot is named Clyde, and you'll be able to ask him anything by "@" mentioning him in a server just as you would an actual person. Except, instead of politely ignoring you or waiting [[link]] 15 minutes to respond like a normal human, Clyde gets back to you in seconds with answers.
In a demonstration I attended with Discord, a user asks Clyde for the current time and weather in Tokyo. Easy enough, but then they ask Clyde for a "good super hero movie gif that captures the joy of watching movies with friends." That's a much taller order—when I'm seeking out a gif that captures the joy of watching movies with friends, I could be there for five minutes before selecting one that makes me feel funny or clever for having found it. Clyde experiences no such inhibitions, and immediately offers a gif of Rizzoli and Isles from the hit 2010 TNT network crime drama Rizzoli & Isles laying on a bed, with the caption "WATCHING MOVIES." Clyde adds that "Watching movies with friends is always super enjoyable!" and I cannot disagree.
A year ago I might've thought Clyde's suggestion was kinda cute, but the "magic" of AI chat bots has eroded at record pace. Talk with ChatGPT long enough for the novelty to wear off and you realize you're having a back-and-forth with a somewhat conversational search engine. There is no personality there to connect with (despite what some say) and I simply don't need a chat bot to do basic googling for me. Clyde appears no different.
If you don't love the idea of Clyde butting into your conversations, you can always turn him off within server settings. Clyde will be available as early as next week as a "public experiment" on select servers.
AI moderation
Discord already has an Automod tool that automatically flags potential rule-breaking messages, but the company tells me it will soon get an AI upgrade that allows it to not only follow the letter of the law, but interpret the context of messages.
The idea is to reduce false positives and catch offending messages that are trying to skirt the rules. Discord demonstrated with a G-rated scenario where, in a server dedicated to cats, a user tries to circumvent the rules by talking about fish but spells it "fi$h" to avoid an Automod flag. AI Automod catches the not-so-clever alternate spelling. Discord says Automod is also able to tell the difference between, for instance, a user calling for "death to all hamsters" (against the rules) and a user sharing the leading causes of hamster death (permitted).
Interestingly, Discord also said Automod's AI enhancements means it even enforces rules that don't explicitly spell out to it. Discord claims that Automod is capable of reading the rules laid out in a server's official "#rules" text channel and upholding them.
That particular claim wasn't clearly demonstrated in the video shown to me. It sounds too good to be true, or at least too ambitious to actually work well. Having moderated a chatroom full of people before, I can see the value in an AI assistant that can read between the lines, but it also sounds super risky to give an AI the keys to moderate outside strict parameters set for it. What if your rules section includes a fake rule like "Absolutely no Coca Cola discourse" that humans understand to be a joke, but Automod takes seriously? Will I get the boot for declaring a love for Pepsi? Flagged sanctions shown in Discord's demo still required a human to confirm their [[link]] legitimacy, at least.
Clyde and Automod are the furthest-along AI happenings for Discord, but it also showed a "sneak peak" at a few other tools in the works, including a few Discord-backed projects from outside companies:
- A collaborative whiteboard app in the "Activities" library that can turn quick sketches into AI-generated art
- An "Avatar Remix" bot that can add objects and features to profile pictures based on prompts
- A third party chat bot focused on "safety and research" called Anthropic
- A third party tutor bot designed for high school and college students called Juni (which already seems to be publicly available)
Also at the event was a demo for Midjourney, the AI art generation bot active in over 2.9 million Discord servers that can spit out pretty pictures at will (as long as you don't mind seriously messed up hands). Discord spent a lot of the presentation lauding Midjourney for fostering an art [[link]] revolution and giving people a new way to be creative. I used to feel something close to that about the possibility of AI art, but nowadays, generating jpegs with keywords sounds like the opposite of being creative.
I think we're all starting to develop a radar for when a piece of art we scroll past is probably AI-generated, and when that radar goes off in my head, my eyes just glaze over it and move on. I do think the Midjourney bot is a convenient tool for rapid impressions—generating a quick picture of a forest to set the mood for a D&D session comes to mind—but otherwise it's hard to care any more about AI art than I do the default wallpaper for Windows 11.
Meanwhile, a feature that will probably mean a lot more to the average Discord user just rolled out yesterday: official voice chat support for PS5.