# Jarvis End goal: build a second brain you can talk to. The perfect UX is one where it acts seamlessly without you needing to tell it to. It just understands you perfectly and only notifies you when you need to know. # Mark 1 Record audio on a device which is switched on the whole time you're awake. Video would take up too much space, and is not as discreet. It has to be discreet enough not to raise questions, so that it can have widespread adoption. Video recording could still be viable in the future if the cost and social stigma issues are reduced. ### Workflow Clip on a lapel mic in the morning, connected to a custom device. Hide the mic so that people can't see it, and it doesn't interfere with normal daily life. Click a button on the device, it is now recording. Recording can be started and stopped at any time. Before bed, the mic is taken off and the box is placed in its charging dock. This stops the recording and charges the battery. While this happens, the box uploads the recordings to be stored safely in cloud storage. To preserve privacy, these recordings are encrypted. Computation can be performed on these recordings via [[Secure Multiparty Communication (SMPC)]] to preserve privacy. This gives the convenience of cloud servers, whilst maintaining the privacy of self-hosting. Jarvis will process the audio and train its AI on it. It will also recognise different voices and keep track of them. ### Why? What we hear and what we say contains a rich amount of data that is currently untapped. It is as important as what we see, perhaps even more so. It is the closest input we can get to a person, besides directly tapping into their brain. Audio has much lower storage requirements than video, and can also be captured discreetly. It is also quite forgiving in terms of recording quality (as opposed to video where camera angles are important, and video doesn't capture everything you see, it is only pointed in one direction, whereas a mic picks up everything you hear and say). It is through audio, that we can capture (almost) all the input that influences a person, and capture (almost) all the output that a person uses to influence the world. ### Why now? The world has woken up to the power of ChatGPT and is ready for integrating it with daily life. Even though there are huge privacy concerns with using it, people still use it because it is so powerful. Social stigma around recording is on the downtrend due to the prevalence of social media. Privacy ML enabled by zk proofs is an emerging technology and enables even more data to be captured and used for AI. ### Why would someone want this? Going to the hassle of recording audio every day and uploading it isn't something most people would do, unless there was a huge benefit to doing so. There are several overarching reasons which I think would compel enough of a market to use this product: - self-improvement: improve your conversations, how you speak, what you get hung up on, bad patterns in your thinking, analytics - posterity, saving a record of who you are, what you did, what you thought, as well as those you interacted with as well - perfect recall, like a constant "hey siri remind me to ..." ### Use cases - after listening to hundreds of pitches, it can generate the perfect pitch deck for you, exactly how you would do it but better - save anything you want to remember - location (can infer from audio, or use gps) - reminders - birthdays - set alarm naturally, mention in conversation "oh I need to leave by 5", sets an alarm for you - adds stuff to your calendar - gives you information when you need it (brought up in conversation) - ask the model questions about the analytics on your audio, chatgpt style. Doesn't need to be graphs and stuff - how can I improve my conversations - which situations make me most nervous - which people do I interact with the most - how much does this person like me? - who do I like the most? - how kind am I to X? - what do I believe that is incorrect? (tells you which facts you've said that are wrong, can go even deeper to interpret your underlying mental models of the world) - replay conversations to you - coach you on how you can better perform at X. e.g. teaching maths, comforting someone - act as a digital clone for someone. Can talk to others on your behalf ### Things to fix - get a cord reel thingy so that the cord can extend easily without causing a mess in the pocket - make the recording device more convenient - a clip you could attach somewhere - wireless - mask it as necklace, headband, earrings, glasses, button, pen, badge, airpods - make as small as possible, whilst keeping good battery life - set file name as the date of recording ### Future features - hardware - charging dock - connect to wifi, automatically upload when in the dock/connected to wifi - location tracking - software - ask questions directly to the mic and get push notifications with the answer - fingerprint each person's voice and connect the conversations in a graph db, linking data sets between people - clone people for other people to talk to. (give yourself infinite of your best sales person) - privacy, [[Secure Multiparty Communication (SMPC)]] - marketplace for selling data. People can take bids on their data from companies, researchers, etc. Data stays private because of ^ SMPC ### Growth Strategy Wear a mic on person and go up to random people saying "is it ok if I record this conversation?" and give them the pitch. Track the penetration per location (using a Plague Inc-like map), move on to a different city once there's a critical mass going. Referral program for people to get 2 free months of the service for every person they refer. Advertise with productivity influencers, sponsor a whole video with them: - Nathaniel Drew - Matt D'avella - Struthless ### Risks! Ethical concern over recording audio without someone's permission. - Should be ok as long as it isn't used for any nefarious purposes, but only personal use. Privacy on the cloud hosting side, but ultimately if the recorder decides to use the data nefariously, that's their responsibility. Zk machine learning is too hard and therefore privacy is compromised. People use the audio for malicious purposes e.g. to manipulate others - mitigated by preventing requests with malicious intent. Ultimately, the fallback is the social layer. People can stop hanging around those who are malicious to them. Too much of a social stigma to record in public/ask for permission to record. [[Anti wiretapping laws]] The very act of recording corrupts the quality of the data due to the [[Observer Effect]]. - Mitigated by recording secretly. - Perhaps familiarity ameliorates this over time People overanalysing their conversations - Potentially mitigated by sheer volume of data and the asynchronicity of it ### Similar ventures ##### Amazon Halo A fitness health band which also has a microphone to record voice tones. All computation is done on the phone. [Amazon Store link](https://www.amazon.com/Amazon-Halo-Fitness-And-Health-Band/dp/B07QK955LS?th=1) [User review](https://www.amazon.com/gp/customer-reviews/R17M1N5PQCSVI6/ref=cm_cr_dp_d_rvw_ttl?ie=UTF8&ASIN=B07QK955LS) ##### [Rewind.ai](https://www.rewind.ai/) View everything you've ever done on your Mac. [Replika.ai](https://replika.ai)