I have a lot of OpenAI API credits to use by end of the month, so I’m using 4o. I’ll probably switch to a more sustainable model afterwards, consider this a request for API credits, all! Email’s on website :)
On a side note, I love the illustration style, some sort of colorful mixed-media collage of UI widgets and physical setting. Does anyone know any similarly-styled documents?
> There’s no official API for ChatGPT yet, so I’m using the unofficial chatgpt NPM package to wire up the app in the demo video. OpenAI has their Cloudflare protection cranked up high; to get around that, it runs a full Chrome instance (!) in the background to shuttle messages back and forth between the little REST API server I wrote and ChatGPT’s internal API.
This is quite a workaround, have you tried the official Davinci003 api? It's rather capable now and probably faster response times. Very cool experiment regardless!
I'm less bullish on specific deadlines for these kinds of notes. I'm planning to do a weekly summary notification - "you made X notes last week, Y of them were high-priority" https://github.com/maxkrieger/voiceliner/issues/20
I suspect Google's limiting this because they don't want devices "freeloading" their cloud transcription service, since most phones can't do it on-device.
on iOS, it's on-device. on Android, Azure - so far I haven't hit the free limit of 5hr/mo. Might start charging Android users if we hit the free limit.
It’s certainly awkward! I’m personally willing to stomach it because it’s kinda cool to “invent a new social primitive”, as dorky as it appears. There’s a very real tradeoff between mere awkwardness and fumbling with a text entry/forgetting the thing altogether.
I also recently rolled out a “create text note” escape hatch in the menu.
Hey HN! A few months ago, all I wanted was a voice memos where I could hold-to-record. Holding and releasing lets me “burst” quick chains of thoughts together. It’s especially useful when walking with friends, to capture stray references and ideas.
After testing the initial burst interaction, I realized I wanted to transcribe them, and relate notes together into a hierarchy. Other features came naturally, like geotagging each note and swiping during recording to change the “temperature” (importance) of a note.
An analogy would be the weights of pretrained machine learning models. They're still serializable, though it's true they're less semantically trackable than a well-written TikZ snippet (which, by the way, I haven't seen many of in practice!)
I work on the SVG renderer for Penrose and would love to reach out and see if there's anything we can do to improve this! The renderer is written declaratively in react so it's easy to add supplemental tags or even a more readable DOM hierarchy.
There is something very whimsical about a pocket-sized lisp machine.
I wonder what the experience of (junior?) high school would have been like if students were given something like this instead of a TI-84. Math and science classes could focus on constructing playable models that answered the questions posed by the instructors a la "Structure and Interpretation of Classical Mechanics" [0]. Obviously there are some ergonomics issues but I'm still going to daydream about the concept.
If people are interested in these kinds of ideas I highly recommend "Mindstorms" by Seymour Papert [1].
Great concept. It seems like the graph view is begging to convey more information. At the very least it seems appropriate for the nodes to have an icon based on what kind of data they contain, for instance, a paperclip or document for an attached paper, a histogram for univariate data, a regression plot for bivariate, etc.
I have some API credits but I intend to make it sustainable. If any LLM provider wants to sponsor some credits, hit me up.
Otherwise I’ll switch models and add user accounts.