I agree with the diminishing returns on spend, but worth noting that when on an Enterprise seat and paying API rates, I'd say that you can easily spend above 200/mo before seeing the curve begin to flatten
Obviously there are a ton of ways to spend money / tokens and people have different levels of experience that will put this ceiling at very different levels for different people.
I think that we are starting to see that API inference prices for US labs are excessive and subscription prices are closer to real costs so 200/m can be realistic longer term.
I'm quite curious. Friends I've talked to have been very pessimistic about accurately measuring blood sugar via a watch.
If they are actually able to get an optical measurement which is good enough for MoM or WoW trends, I'd agree it's pretty big even if they can't provide anything precise enough for real time / daily use. Just being able to roughly validate diet changes for women with PCOS could be very impactful and that's the smallest of the 3 populations you called out.
I don't see the point of focusing the conversation on Mexico City. You may be far enough from the fault line to have plenty of time to react, but not everyone does (as you acknowledge). You may not have tsunamis, but that doesn't change the fact that one needs to know WHAT to react to.
You are welcome to think it is ridiculous for needing 5 seconds to process what alert is going off. But the simple reality is no good disaster plan should assume that everyone will immediately understand what is happening.
In most cases that would indicate an organizational failure. However, if something is trillions of miles away, and wrong commands could break communication etc, I'd probably just stick with the most experienced person as well.
Not likely. When the price goes down, some people take jobs doing other things pretty quickly. When it goes back up, they don't all come back immediately. The result is that the price tends to stay high longer than it stays low.
Parakeet is certainly faster on my machine (m3 max), but I can't stand using it over Whisper for dictating my prompts. It makes a lot more mistakes, possibly because (like you mentioned) large portions of the speech will pause / stutter as I think about what to include.
With whisper v3 turbo, I can almost always live with the few mistakes because the overall stream-of-thought word-salad I provide is still clear at a high level. The bits and pieces of context seem to help, that I might leave out if typing and focused more on traditional conciseness / clean writing. With parakeet, I needed to do frequent editing even for shorter bits of speech.
I realize some applications prioritize the latency.
The near-instantaneous nature of Parakeet has led me to keep using it and occasionally simply re-dictating a second time as needed.
For round two after a typo-laden transcript, I’m dictating and annunciating with great passion as I read the first transcript to make sure I don’t miss a beat. It’s kind of fun because it’s a little performance.
It sounds like post processing should be the job of an LLM. I would like the voice model to be faithful to what was said and then that output can be smoothed over or postprocessed as needed for the use case
To be clear, I'm talking about high word error rate with parakeet vs whisper, not post processing and cleaning up my speech. Re: being faithful to what was said, one small example, Whisper will often put ellipses when I pause.
I used to work at BigTech in a role where I spent a lot of time reviewing new open source projects. It was always a fun game to guess which single person in which department tried out a piece of software that caused our logo to end up on the "as used by..." description on the tiniest of projects.
Fair haha. I'm no trademark lawyer, but I have to assume there's some wiggle room when you're not actively stealing their trademark and just referencing it... but still, the combined value of endorsements from Google/Nvidia/Amazon/ByteDance/Tencent/Alibaba/SalesForce/IBM etc has to be worth... a lot. That "not company endorsements" is rather load-bearing. I don't think I've seen anyone just say that, so maybe they're being a bit less bold haha.
You might as well say Herdr runs on 3 billion devices, go all in! 16 billion devices! Every human on earth installed Herdr twice! (source: study amongst users of herdr, n=5)
I get your point completely. I was (naively, I suppose) pretty surprised when I realized how low most people's threshold is to add a logo and claim a client/brand.
I've seen VCs very much encourage this (and shadier) behavior. There's a line between hustling / being an entrepreneur and actively trying to mislead people... clearly we don't all see that line being in the same place (:
edit: to be clear, I didn't mean to make such a broad statement about VCs. I'm sure plenty of people do this on their own
When I was in Italy last summer, I couldn't figure out how to pay with my card at the machine in a small town, where you'd park to walk into an ancient city on a hill*. I asked two Italian woman for help and even with being able to read the Italian + having paid with coins themselves, they struggled to help me understand the combination of steps required to pay with card.
And that's the beauty of open source and code. You can share it freely and easily. There's no thing that can be made for everyone.
Maybe some irony is everyone tells me they "just care that it works". Yet it can work and you'll always have the comments like above because it works for the reason it was made but not for things it wasn't made for. But it's open, so modify the code and put in what you want ¯\_(ツ)_/¯
I guess I should clarify its source available, not "Open Source". There's no license on the project so it's default theirs. But no harm if you're just editing it yourself. @OP should put up some license to let others know how what is allowed and what isn't
Agree on the source-available clarification — this exact
distinction matters on HN and I learned it the hard way recently.
I just went through the licensing decision for my own project
and landed on BSL 1.1 with a 4-year conversion to Apache 2.0.
Framing it as "source-available, auto-converts to Apache 2.0
in 2030" reads as transparent intent rather than "fake open
source."
That said, BSL/FSL really only make sense if you plan to
monetize a hosted version yourself. For wrapper tools like
Claudraband that sit on top of an existing product ecosystem,
MIT or Apache 2.0 might fit better — you're not protecting a
competing SaaS, you're just sharing code.
Obviously there are a ton of ways to spend money / tokens and people have different levels of experience that will put this ceiling at very different levels for different people.