Are YOU prepared for the chatbot flood? OpenAI is switching from being a chatbot provider to a chatbot platform, giving ChatGPT Plus subscribers the tools to build their own chatbots, host them on the new GPT Store, and earn a share of the revenue their custom bots have generated. It’s a huge announcement, and everybody’s very excited about it. 

‍

So excited, in fact, that they’ve missed the fact that as part of the GPT Store’s ongoing rollout, OpenAI have launched thirteen new chatbots of their own. It’s easy to overlook them as they seem to simply be examples of what you can achieve using the new GPT Builder tool, but in fact they are quite different. Most of them are specialized GPTs with their own dedicated training models, geared around a single subject. Some are silly, others are extremely useful, but they’re all remarkable and surprising. Let’s dive into the top of the list with arguably the most interesting new OpenAI chatbot of all…

‍

Game Time

 Game Time is a dedicated chatbot; in its own words it specializes “in discussing and explaining board games and card games, offering insights into game rules, strategies, and setup.” We love games here at SPYSCAPE, so this is right up our street! And Game Time is fascinating. 

The first thing to note is that like many of these in-house GPTs, Game Time has a dedicated training model with a knowledge cutoff of January 2022. This is not simply a reskin of GPT-4! We’ve been testing it out on the rules and strategies of various complex board and card games, and it has performed well, but is far from perfect; typically when asked a generic question about rules it will provide all the correct information, but also some incorrect information as well. 

‍

When it comes to board games, it seems to have been trained on the published manuals for those titles, but lacks in-depth knowledge of game mechanics, so it knows how to play Wingspan, but does not hold information about specific cards playable in the game.As for more traditional games, it still can’t play chess, but it does have an excellent knowledge of the concepts of chess; we quizzed it for some time on the finer details of middlegame strategy in the Jobava London system, and it responded with the authority of a Grandmaster. We then tried asking the same questions of vanilla GPT-4, and got slightly less informed results. It’s hard to be certain, but Game Time really does seem to have a lot of specialized knowledge that is unavailable elsewhere! It’s a very useful tool, and a strong start for the in-house GPT Stable. Let’s see what else they offer…

‍

The Negotiator

The Negotiator describes itself as “a virtual coach for negotiation skills, aiming to empower users with the knowledge, strategies, and confidence to negotiate successfully in various aspects of their personal and professional lives.” Like Game Time, it has a dedicated model with a January 2022 knowledge cutoff, which is focussed on generating roleplay scenarios to aid users in preparation for their negotiation challenges. We’ve tested it with various hypothetical challenges and it certainly fulfills its brief, offering strategic advice and providing a space for practice. Generally speaking, feeding details of your personal and professional life into chabots is a terrible idea, but with a little care you can anonymize your personal situation and still generate usable roleplay scenarios; this could be a useful assistant if used in the right way. 

‍

Creative Writing Coach

Another January 2022 cutoff, with a training model that the bot claims “focuses solely on the nuances of creative writing. This includes an in-depth understanding of literary techniques, styles, genres, and character development.” We had some fun testing this out by feeding it chunks of classic literature; for example, Moby Dick is praised for its distinct voice and style, but the bot feels Melville should focus more on clarifying complex sentences and maintaining consistency in tone. For more contemporary authors, this bot has some use as a tool to bounce paragraphs off for some quick editorial feedback, but it’s severely hamstrung by one factor; the “context window”. This is the upper limit of how many “tokens” the bot can process in its memory at one time, and it currently stands at 4,096 tokens, which is roughly equivalent to 1400 words. Restrictive, but still useful, unlike our next chatbot, which is restrictive and pointless…

‍

Cosmic Dream

How the bot describes itself: "I am Cosmic Dream, a vivid burst of creativity and coolness. Like a psychedelic vision, I inspire through imaginative, colorful responses, weaving dreams into every interaction. Always positive, always imaginative, I turn even the mundane into a kaleidoscope of wonder.”

‍

How we describe Cosmic Dream: A baffling reskin of DALL-E 3 that only creates bad psychedelic art. The intention is clearly for it to be whimsical and fun, but we’re unsure why this is any more fun than simply using DALL-E 3. Your mileage may vary depending on how keen you are to have dreams woven through every interaction; for us it grew tiresome after precisely one interaction. Next! 

‍

Tech Support Advisor

Ah, this is more like it. No matter what your level of technical expertise, there’s always something waiting to trip you up, and Tech Support Advisor is an absurdly useful bot for nerds and normies alike. It’s another January 2022 cutoff bot, but the lack of contemporary knowledge is more than compensated for by its dedicated training in understanding and explaining tech problems. We’ve tested it versus vanilla GPT-4 on various thorny technical challenges and it always comes up with answers that are both more detailed and more easily understood. A true star.

‍

Coloring Book Hero

The first of our bots without a dedicated training model, or at least the first to have the same knowledge cutoff as vanilla GPT-4. This generates cute black and white lineart images for kids to color in, but it does not provide color-by-numbers scheme for kids (or adults!) to follow, which is disappointing. 

‍

Laundry Buddy

Watch out Drew Barrymore, OpenAI’s invading your turf! Laundry Buddy boasts a dedicated training model that has a deeper knowledge of stain removal tips, and also a different conversational style to vanilla GPT-4, which will sound familiar to fans of daytime TV laundry experts who are also former child stars: it dispenses clear DO’s and DON’Ts in a “cheerful and upbeat tone.” Phone home, GPT!

‍

Sous Chef

A similar theme to Laundry Buddy, with a dedicated model focussed on recipe discourse, but Sous Chef also has a very cool trick up its sleeve. This bot doesn’t just tell you how to cook, it will show you as well, with DALL-E 3 enlisted to generate images of the recipe. This means that it can provide reference images for how your food should look at a given stage of the cooking process, which is a tremendous boon for less experienced cooks. 

‍

‍

Sticker Whiz

Another novel use of DALL-E 3 here, with a bot that will not only generate artwork for sticker designs, but also provide you with a link to order those stickers online. Yes, physical stickers you can stick to real things! OpenAI have partnered with Sticker Mule to handle the printing and postage. We’ve not been able to test it, but it‘s certainly a cool idea! 

‍

Math Mentor

The last of the potentially useful bots, but one that does not have its own dedicated model, and it probably needs one.  Math Mentor is a GPT-4 reskin with an April 2023 cutoff, and as such we’re not entirely convinced there’s any benefit to using it over vanilla GPT-4. It has been “tailored to engage in a more educational and explanatory manner, especially suited for explaining mathematical concepts and solving math problems,” but in testing we haven’t seen much difference in its responses. Sadly, GPT-4 is not a reliable mathematician, with various weaknesses - particularly in calculation - which make its utility as a mentor questionable. Although not as questionable as the final three bots…

‍

Hot Mods

Hot Mods suggests alterations to images. Upload your image and it will suggest ways you can decorate or augment it. It doesn’t generate or alter images itself, it simply describes what it thinks you should do with them. We’re struggling to think of a use case for this! The closest Hot Mods gets to useful functionality is its ability to create prompts for DALL-E 3, but vanilla GPT-4 also does this perfectly well. A very strange bot.

‍

Mocktail Mixologist

We’ll let the Mixologist describe itself: “Think of me as your go-to mocktail mixologist, ready to bring fun and creativity to your drink-making adventures. While I don't delve into alcoholic drinks, I'm all about making the most of what's in your kitchen and stirring up some delicious, alcohol-free magic.“ It offers nothing over vanilla GPT-4, other than a refusal to discuss alcohol-based drinks. 

‍

genz 4 meme

Lastly, and very much least, we have a genuinely bizarre creation. genz 4 meme’s tagline is “i help u understand the lingo & the latest memes” but this is a chatbot with a dedicated model and a January 2022 knowledge cutoff, so that’s clearly not true. It also sounds an awful lot like Elon Musk’s widely ridiculed chatbot Grok, which is ironic given that one of the people ridiculing Grok has been OpenAI founder Sam Altman. Both bots share the same “your uncle trying to sound cool” approach, something they would no doubt describe as a “cringey vibe.”

‍

Many of the GPT bots above can be dismissed as merely replicating existing websites, but they add something extra through their ability to explain concepts in detail. genz 4 meme does the opposite, simply replicating content you can find on websites such as Know Your Meme, but obscuring its answers behind unconvincing attempts at sounding youthful. Worse still, it maintains that facade even when discussing serious subjects; when asked about memes that are common among the far right, it persists with its goofy chat style, leading to truly awful sentences such as this description of one far-right meme: "it's like a mix of irony, internet culture, and some pretty heavy political vibes." Stick with the websites.