Why choose when you can have both OpenAI and Anthropic? - An evening with Poe - part 1

The only downside of so much happening in generative AI: I might need to consider using it to generate content (rest assured, I won't). My original post was supposed to be about prompt engineering, but then came Poe, Quora's generative AI chatbot. A brief look turned into another evening with generative AI (sorry husband! :-). Or rather, with three, because Poe is more of a collective, with bots Sage and Dragonfly powered by OpenAI and Claude...yes the Claude from Anthropic!

Quora's goal is not so much to introduce yet another generative AI, but rather to give users an easy-to-use chat interface that allows us to easy access technology that's already there. Really cool idea, if you ask me! Poe is still free, and you can download it as an iOS-only mobile app. Android will follow soon.

A first look

efore you can start, you need to sign up and verify, both by phone and mail. Once done, you get access to an app that doesn't sport one, but three chatbots: Sage, Dragonfly and Claude.

The chat interface looks really clean and easy to use. Something that's new compared to ChatGPT is that, apart from the generated answer, the bots also show suggestion chips that offer topics and further questions.

In addition, you can also connect with other Poe users, and access a feed with conversations that they shared. This is a great feature: how often don't you find a really cool conversation and post it on social? I guess this will also give the platform developers a sense of which conversations are useful, interesting or worrying.

OpenAI's bots: Dragonfly and Sage

Sage and Dragonfly are both powered by OpenAI, and as such, don't have access to any information after 2021, just like ChatGPT.

While at first they look pretty similar, there seems to be a difference between Dragonfly and Sage. Whereas Sage is positioned as a chatbot, Dragonfly profiles itself as an AI assistant that's capable of more complex tasks, like staying organised, and doing research and data analysis.

So Dragonfly seems to have a pretty clear picture of who or what it is, how it's positioned and what the differences are.

Let's turn to Sage for a while.

Okay...Sage seems to have a hard time remembering what it is, who Dragonfly is, and that both are here in Poe, which is not really a bot but an aggregation platform with the last name of a famous writer. It puts its name in parenthesis, almost as if to remind itself who it's supposed to be. Or is it GPT-3?

Hm, don't know about you, but at this point I'm already feeling my suspended disbelief pooring back in through the cracks of inconsistent persona. Of course, when you're chatting with just one Generative AI bot, it's quite easy to uphold the illusion of human-like conversation. But if there's more parties involved, and they don't know each other, even though their, well, family, I guess? That's kind of weird. Let's see if Sage knows where it is, and who else inhabits the Poe space (aaargh...nice piece of bias creeping in, I confess...I start to see Sage as female. Oh dear).

And that brings to light an interesting aspect of generative AI bots: they're pretty much brilliant on their own (or at least, that's what they can make us believe), but as soon as you take them out of their 'isolated in time, isolated in space' context and connect them to the internet or put them in a social context with multiple actors, it becomes so blatantly obvious how clumsily these interfaces stumble through a conversation. We still have a long way to go.

Claude

Claude is Anthropic's bot that's based on constitutional AI. I'm still on the waitlist for the actual Claude, so it was a nice suprise to find it available in Poe already! Like Sage and Dragonfly, it can't access the internet, but its training data is a bit more recent, and reaches up to 2022.

What makes Claude special in my book, is that it's specifically designed to be increasing less harmful (which in this case means that it's designed to reach increasing levels of truthfulness, i.e. it aims not to bullshit). To do so, it uses constitutional AI, with a core set of guidelines ingrained in its little bot brain.

What still suprises me is that Google choose to go live with a generative AI that was based on Lambda, rather than with Claude. Claude is supposed to be more cautious, and while that may make a more boring conversation, its conservatism might actually have helped Google to look like the more reliable party in this generative AI playing field.

Wow, that turned pretty elaborate pretty quickly. Stay tuned for the next installment where I'll take all three bots on a test drive. And of course, after last week, there's only one question that matters...Who took the very first image of a planet outside our solar system?

Join me at Convoclub Language & Linguistics!

Monday February 13th, 18:00 CET

About Maaike Groenewege

Maaike is a conversation design lead, linguist and DesignOps coach with?Convocat conversational expertise. She helps both starting and more experienced conversational teams in optimising their conversation design practise, NLU analyses and team communication. Her main focus right now is on how LLMs can benefit enterprise conversational AI.

• Join?Convoclub, my boutique community for conversation designers, linguists and language lovers. We meet every other Monday at 18:00 CET, find the invites in the Convoclub Events section, or keep an eye out on LinkedIn!
• Find your next book on conversational AI in my?Convocat library