1,067 karma · joined August 20, 2022
Ambient ScotRail Beats: https://matteason.co.uk/scotbeats
matteason.co.uk Twitter: @MattEason Email: hn@matteason.co.uk
It would be nice if you could change the block type after creating it. For example I don't think there's a way to change an h1 to an h2, or a bulleted list to a numbered list.
The collaborative editing is neat too, are you using a library for that or is it bespoke as well?
> If your work isn't ready for users to try out, please don't do a Show HN. Once it's ready, come back and do it then. Don't post landing pages or fundraisers.
If the user marks it as correct it could have a verified tick or something on the admin side, and then form creators would only have to manually check over the ones without a tick (with the option to see the original text of the user's response and a form to correct the data)
Obviously it would also be useful for the form creator to author the question so that the possible options are given when they're non-obvious, but users don't always read :)
I'd have a couple of concerns if I was looking to adopt this:
- Is it actually quicker than filling in a form, especially for questions which would normally be radio buttons or checkboxes? For example on the conference signup example I have to type the days I'll be attending rather than just clicking a couple of checkboxes
- The validation doesn't seem very strict. For example on the conference signup again, I entered 'John Smith, Saturday and Sunday, 4XL' and it recorded my t-shirt size as XXL. The fact that there's no user-visible validation step is problematic because in real life, if I submitted my response by email I'd have no way of knowing that the AI had misclassified that response. As the form creator I don't want to have to look through the raw text of every response to check if it's been recorded correctly.
I think the site feedback example is the best use case; at the moment I wouldn't trust it to work with anything with more concrete real-world effects like getting the wrong size t-shirt or being registered for the wrong days
I agree that HTML -> PDF can be a really powerful tool. I worked on the UK government's tool to generate energy efficiency labels for consumer goods [0] and we ended up doing PDF generation with SVG templates, using Open HTML to PDF for the conversion. That ended up working very well, though as you allude to there can be some gotchas (eg unsupported CSS features) that you need to work around.
A few questions:
- Do the rendered documents support PDF's various accessibility features?
- How suitable is this for print PDF generation? For example, what version of the PDF spec do you target? What's your colour profile support like? Do you support the different PDF page boxes (MediaBox, CropBox, BleedBox, TrimBox, ArtBox)?
[0] https://github.com/UKGovernmentBEIS/energy-label-service
> If your work isn't ready for users to try out, please don't do a Show HN. Once it's ready, come back and do it then. Don't post landing pages or fundraisers.
Either way, there's not nearly enough information on the page for me to trust this with anything. No privacy policy, nothing to show how information is stored or protected, nothing showing how it actually works - just a checkout page asking me to pay $2 for an unspecified service
The pricing does seem a little steep in terms of the number of conversions per month. At the low end (basic, monthly) it's $0.90 per conversion, but even on yearly essential it's $0.66
The main call-to-action button says 'Start listening - it's free!' but then the pricing details don't include any information about a free tier or trial
It looks like you need to listen via the website, is that true? Ideally I'd get a private RSS feed (options for one per newsletter or a combined feed) that I could use in whatever podcast app I wanted
It would be great if the frame adapted to the aspect ratio of the uploaded image. At the moment the image is cropped to fill the frame. Even reordering the templates to prioritise those which match (or are closest to) the aspect ratio of the image would be a start, or the ability to filter portrait/landscape/square
I find the 'bass rumble' and brown noise particularly useful for masking external noise, which I guess is pretty similar to a thunderstorm frequency-wise
- on the 'Report a bug' demo I'd expect the 'Did you discover a bug?' text to be in a <label> connected to the textarea with for="id", and the 'Please let us know what happened' text to be connected with an aria-describedby on the textarea.
- on the CSAT demo the happy face/sad face buttons don't have an accessible text alternative to the icons, for example using aria-label, so the buttons will be unlabelled to someone using a screen reader
At least HN tolerates reposts if you don't get any attention the first time round [1]; if your launch bombs on ProductHunt, you don't get a second chance (ask me how I know!)
On gsky's about HN traffic lasting two days - that was definitely my experience, though if it's interesting enough it'll likely get picked up by other sites too. Blogs, podcasts etc are always picking stuff up from Reddit and HN.
I guess lots of competition means there's definitely an audience for it at least.
The HN post made the front page and got me the most traffic and feedback I'd had so far (about 6k visitors). It also resulted in articles on BoingBoing and Metalfloss, among others, which also sent a decent number of users.
The 1440 newsletter [2] also picked it up - just a small one-liner in a list of interesting links - and sent about 8k visitors in one day.
On Twitter it got picked up, unexpectedly by quite a lot of DJs. Turns out that a lot of DJs get tinnitus and white noise can be really helpful to mask it. So I got much less traffic from that but some really nice feedback on how useful it is. That also led to The Economist's podcast team contacting me and I ended up featuring in a piece about ambient music and white noise apps.
I did try ProductHunt and it got nowhere. It probably didn't help that I didn't really promote the launch, but it also got buried under a lot of AI projects. Ho hum.
User numbers have stabilised now at about 600-800 a day. I'm not really sure where to turn next - I hate constantly banging on about it on my personal Twitter and reposts of things that have got traction are discouraged on HN and Reddit (rightly so). I'm happy that it's got a decent userbase though.
[0]: https://ambiph.one
[1]: https://news.ycombinator.com/item?id=38856999
[2]: https://join1440.com/newsletter/florida-drug-imports-starsky...
- What resolution are the generated images? If I want to use them as actual posters they need to be a decent resolution (ideally 300 pixels per inch)
- Its understanding of text seems about on par with other AI I've seen (that is, not very good). There are obvious errors in the collage on the login screen ('DELIVVERY', 'BUSY PEOPLE FOR TO TIME'). So I'd need to do some work to get these to actually be usable. You mention conversion to SVG in the FAQs but I don't really trust that to get me something easily editable - and then how do I match fonts, text effects etc with the rest of the image (especially if the AI has just made up a font).
- The 'Pro' plan (the only plan available) includes 'Unlimited uploads' - uploads of what?
A lot of the best work I've done has been a bit under the radar. It's easier to ask for forgiveness than to get permission.
And this one precipitation estimates: https://go.nasa.gov/49a68Nq