# Preview Prior to actually using the thing, here's what I want / expect / hope for. If the pendant doesn't meet this specification I'm going to be disappointed. - All-day, always on use - The whole idea of this thing, as interpreted by me, was that you could record _everything_ you say and hear on a daily basis. As such, I will be disappointed if this thing runs out of batteries or somehow doesn't actually have an always on mode. - Data access - I want to be able to access transcripts so that I can incorporate them into my other memory systems, such as my Obsidian notes. - I want to be able to back up my data with zero dependence on Limitless. I expect they will go out of business or pivot, as happened with Rewind, their last product. - Multi-lingual - It should support Chinese as well as English. This is not asking for much given modern transcription technology, so this is essentially my hope that they use modern transcription technology rather than something less effective like built-in on-device transcription[^1]. Non requirements, but nice to have: - Actually useful. Rewind proved not-useful unfortunately, it was too hard to find what I wanted. Hopefully the pendant will do better. - Ability to revisit the actual audio. - While I'd like to be able to go back to the audio recording, I suspect this won't be available due to the size of the data. Once transcribed _i expect_ they will discard the audio. That seems like a worthwhile tradeoff. - Still, Limitless requires a subscription so they could afford to store a lot of data per user. --- # [[2025-06-02|First day of use]] Here are my initial impressions. - There is an API!: https://www.limitless.ai/developers#setup - This is great. Of course its downstream of useful recordings, but its a great sign. - Tap to favorite a moment. Also great. Really appreciate the option for manual curation, even if limited. - How can I add custom vocabulary? Especially for Chinese, I'd like to be able to add people's names so that the transcript isn't constantly off. - I'm currently a bit disappointed in the lack of plain search. - There's only Ask AI right now, which is, of course, going to be using Search somehow under the hood. But there's no search endpoint where I can just go find things. Actually, maybe it's not fair to say there's no endpoint. I haven't checked. But there's no feature inside the app. There's no search box. There's just an Ask AI box. - The lack of search box is unfortunate. - This would be a pretty egregious omission if not for the fact that there's an API, so you can roll your own search if you want. - Even so, they must have a search feature to support the RAG AI responses, so the fact that they don't expose that search functionality to the user is very off-putting. - The _automatic_ AI usage will be unwanted for some - This comes in two flavors. The first of which is the summarization of individual transcripts, which happens all the time automatically. - The second is the full day summary and 'prompts' as I call them. The system will automatically ask for a daily summary but also automatically prompt with things like "how can I improve?". That's not something I want to happen automatically, but as far as I can tell there's no configuring it. - Stress tests - Works in the elevator where there's no connectivity but the phone still thinks there is - Works in the gym with low connectivity and background noise - Seems to work fine whiles witching from wifi to cellular - Completely fails multi-lingual overlapping speech. - This is not surprising though. As far as I know there is no transcription provider that can handle this case. - In my household there is always a mix of Chinese and English, because I speak English to my son and my wife speaks Chinese to him. Thus there is often _overlapping_, _multilingual_ audio. Unsurprisingly this is an extreme challenge for transcription software. - On top of this we have two more confounding factors: - My own Chinese is not perfect, and my accent is neither fully Taiwanese nor Mainland Chinese. - My two-year-old son's English and Chinese are still far from perfect. His voice is that of a child, his pronunciation is often incorrect in novel ways, and his grammar is not perfect. - He also often mixes English and Chinese in the same sentence. - Interestingly it properly transcribes my wife's Chinese as Traditional Chinese. This is interesting because it must have pivoted on her accent. It seems to assume my Chinese warrants simplified characters. ## Do we really want always-on recording? This question arose in my mind while running my tests. When every single thought could become a voice note with almost zero additional effort, there's always the question of whether or not you should speak out loud. I've only been running the experiment for a single day so it's too early to judge, but I could imagine the added mental friction of deciding whether or not to speak could become a burden. I've also had one of my friends express trepidation over the prospect of being recorded by others. The Pendant makes it very easy to stop recording, but I can see his point. If these devices became ubiquitiosu one might assume that everything you say might be recorded. ## Overall I'm excited about this device. So far it has met my expectations and even provided a nice surprise in the form of the "star this moment" functionality. None of the drawbacks are deal breakers so far. I haven't yet had a chance to test that it really can work all day. Another stress test I'd like to put it through is attending an all-day conference, with lots of dialogue from different people. I don't have any such event on the calendar though so I won't be able to run this test any time soon. [^1]: Apple's built in transcription on iOS is simply horrible, so it will be a huge disappointment if they use that.