ryebread777

Here’s a rough estimate. Though of course this was a side project (~10hrs / week) and much of the time was spent learning.

V1 - Wordpress and jupyter notebooks (2 months)

V2 - Django (7 months)

V3 - Django + React (3 months)

Don’t worry, If you are interested in buying $6k plants I will create a new site just for you lol

Web scraping or using site APIs!

Right now either they are from Wikipedia, or they are from one of the stores selling plants. I have been reaching out to many stores to request permission to use their images in this way and luckily a few have been kind enough to grant it. Those are the images I use. If you scroll down far enough though you will see there is still a need for many more.

Good idea!! I’ll admit I came at this having much more experience with Houseplants than gardening, but I definitely want it to serve many communities. Thanks for the suggestion! Is that basically the same thing as “zone?” I think that should be doable - I will look into it! Would love to know any other features that would be valuable from the gardening perspective.

Ah, yes you are right about that. The search sucks right now. An even bigger issue is that search only matches in the scientific name but most people are entering common names. Yours is the first user feedback to mention it though so thank you for sharing! I will make it a higher priority to fix.

Sorry! I forgot to include a disclaimer that most of the stores are US based right now. I am excited to expand to different countries but there are a few difficult questions to figure out in terms of user experience. I’m hoping to nail the core functionality and then focus on expanding to other countries. Thanks for the comment!

This is neat. It worked really well for my image. I assume it is just generating an image in the style of a mini figure?

It would be really incredible if it could generate a lego mini figure using real components (ie. this combination of hair, hear, torso, etc). I can see how this would be much more difficult. But it would also make it a really interesting product imo. People already pay a lot for mini figures - would probably pay a LOT for one that looks like them, or for a set that looks like their friends/family.

My wife and I tried the game out over the weekend. Overall the game does a lot really well. I'd say it's a well-done game, in the same vein as one of those games where you select dialogue choice options. My wife quite enjoyed it (gave it 8/10) but I expected something a bit different (5/10). In particular, I was hoping there would be a lot more freedom to guide the game, or at least guide conversations, through language. In reality, most of the dialogue was about accomplishing a pretty specific and simple objective. Ie. compliment this person, agree with their viewpoint, etc. The game was just about figuring out the objective of each challenge and then completing it. There was no balance (ie compliment him but also seem genuine. Don't offend him, but also don't be a pushover) or opportunity for creativity (ie. you can lie your way out, or appeal to their pity, or distract them, any of these could work). If someone likes games where you select dialogue choices this might be a good game for them. But I was hoping for something that felt more natural and self-driven. At the very least, I think making the conversations longer and more open ended would be nice, even if the outcome doesn't affect the story. One of my favorite challenges was booking the dinner reservation, just because I ended up having a short conversation with the man about the pasta bolognese that didn't feel like it was part of any script.

Everything else (graphics, voices, dialogue, interface, story, transcription) worked great and I don't think need to be improved. But the main mechanics of the game feels like a thin layer of LLM on top of a rigid skeleton of scripted scenarios with fixed outcomes. I'd love to see the language model on a longer leash.

Thanks for sharing and huge kudos for creating a playable game that was legitimately enjoyable! I look forward to the next parts of the story.

This is brilliant! I think a game with LLM NPCs would be crazy cool. Agree with the other commenter I would probably need a playable demo to pull the trigger though.

Also, I didn’t realize from the website that it actually uses voice? That should probably be highlighted, unless I just missed it.

One other thought, your title says it’s a game about social persuasion. In the examples on the site it seemed like many of the challenges just involved coming up with lies. If there are no consequences to just making up lies it seems to kind of flatten the strategy. Though I’m just speculating. Thanks for sharing!

That’s pretty neat. I can see a bunch of those from my balcony. I always wondered where squirrels lived!

This is very cool! One comment - I don’t love the on-hover effect on the images, at least for mobile. It keeps triggering as I scroll the page. Might be worth rethinking. I imagine on desktop it would be great though.

Seems useful as I do occasionally get broken JSON from LLM responses and would like an easy solution to that. However, I don't see this as something I would use an API for. Especially seeing the supported fixes are relatively simple. Seems like it should be a python package. Maybe if you could somehow handle a comparison with a pre-specified json schema that would make it more valuable.

Not familiar with similar libraries so not sure how this stacks up. But a few things that I see as helpful improvements would be: 1. using templating engine like jinja to allow for more flexibility in customizing prompts 2. sending multiple rows per request. If I want to do sentiment classification, I may have a fairly large bit of text explaining exactly what I want the classifier to do. If I include that instruction along with every individual sentence I want to classify, I'm wasting a lot of tokens. Instead I'd rather send the instructions along with 100 sentences and ask it to classify each one. This sort of functionality would work well with what you're trying to accomplish using dataframes.

I’m confused by the naming of their models. Is this equivalent to or associated with one of the existing turbo-preview models?

Very helpful! I will come back to read more deeply. But in general, what approach would you recommend?

Very interesting, I wonder if there are studies on the effects of removing downvotes / upvotes / both from a platform. I tend to associate my dislike with the types of content on certain sites with the fact that they don't have down-voting, but hard to say what impact it actually has.

This seems like a good policy to me. Though it makes me wonder why we would treat an image made with AI differently from an image that has been edited digitally via photoshop or other means.