Quick update on @CallDRCope: - Support for Calendar and health data integration is going well and should be ready for internal testing today - Login is simplified to use Magic Link, which automatically inherit security from your existing email (2FA, etc) is done It's likely that I'm going to hold off the release to early access participants for a day or two. Here's why: - The current implementation for E2E encryption is too cumbersome when switching devices and with people changing their phone annually sometime, this will become a pain, especially if they forgot their PIN since that means all data has to be decrypted and reencrypted after recovery. The adjusted approach we are taking will not just make it more secure, but also eliminate the need for having to touch the data even if user has to go through recovery. In case you were wondering, this is the security scheme we will be using: "A random 256-bit master key encrypts all data and is only stored/synced as an AEAD-encrypted blob wrapped by a PIN‑derived KEK (Argon2id/PBKDF2), with no plaintext keys persisted and a human‑readable Recovery Key for disaster recovery." Security and privacy are super important to us so I want to make sure that we have the right approach from the get go, especially when it will not delay the full launch. Which takes us to the discussion of the full launch. The paperwork is almost done. Based on the current estimate, the launch window will be within August. The only thing that could hold us back is the D.U.N.S. number, which should now be expedited with the help from Apple. Our participation in the research project will kick off mid-August regardless of the full launch. We will just have to distribute it the existing way. Not much we can do about it. Anyway, we are almost there. $COPE

In most interesting part I found with ChatGPT-5 is consistency regardless of where I use it. I asked it for a brief on how to create a web page and I took it to Lovable and turned on GPT-5 and the result was basically identical on both end. The only dumb thing it's continuing to do is overselling its capability. For example, I asked if it can under a GIF and read all the content so that it can see all the animation, behavior, etc, since it cannot understand the MP4 recording, it will tell me that it can and make me go convert MP4 file to GIF and then fail to produce. This alone is quite crucial and potentially will end up causing a lot of time wasting for developers who rely too much on AI to code. @sama is there a way to get rid of this behavior?

Just had a chat with the team. If all goes well, these will likely be released Monday night to early access participants. Health and calendar data will be fully encrypted and no one accept the user and @CallDRCope agent can see it. We are taking our usual approach of using scientific/mathematical formula to make sense out of the data instead of having to rely on private data. $COPE

This is happening more often than not. I evaluated 500+ decks in the past 4 years and also evaluated competitors to our products. What I found is many are just a prototype or even a clone + minor customization of something (Project Lyra) being the most common. Never afraid of competitions. If you think you have the right idea and the right team, you can likely outpace them.

Had a couple of people DM'ed me about using MindExtension (ME_) Agent Creator Platform for their AI agent just like @CallDRCope so I thought I share the reason why you should consider as well: - ME_ is specialized in memory management. There are currently 3 architectures that compliment each other. For general use cases, these should be sufficient. We have used it for cases like Dr. Cope, e-commerce website, customer support agent, etc, without having to add another layer on top. This helps simplify development significantly since you just have to configure instead of developing it yourself. Imagine having a more superior memory on top of any LLM. - You can enable thinking even on models that don't support thinking. - Scalability. You build and we help you scale. - Personalized support. ME_ is still young and is open to new ideas and requirements. You may find that working with a bigger platform means you are stuck doing things the way they tell you to. With ME_, we can help support a more personalized approach when it makes sense. - Free credits for you to explore while going through ideation phase. We are looking to support any interested builders. - If you need help with development, our subsidiary, @SiloTechAI, can also support you so that you don't need to hire your own developers until you are sure about what you want to do. Anyway, if you are interested, let me know. The platform will become self-service and fully open later this quarter, but for now, we will do a more personalized onboarding. @saakuru_labs $SKR

Since someone asked about what the UI will look like in the @CallDRCope app, I thought I share these 2 new sections in the daily view. There are also changes to year and month views as well so you can view them based on overall assessments or specific ones like health/bio data only. Looking to push these out very soon. $COPE

Quick update on @CallDRCope: - We got 4 devs to help speed up the dev work since we are looking to get a few more features in because we kick off the research project in 2 weeks in Singapore - Data from smart devices and calendars (not just Google Calendar) are available and @SerenYenikent has been tasked to come up with scientific formula to help preserve privacy while providing insights into users' situations so that the app will be very helpful to users in their mental health improvement journey - On the paperwork side, it's progressing slowly, but we are getting there. We are definitely looking to push the app our for public release as soon as we are able to. $COPE

We have been quietly working behind the scene on @CallDRCope. TL;DR : - 3-month pilot cohort in motion. Working directly with supervisors and students in practicum settings - Partnership with Cura: Therapist tooling for session notes, automation & burnout prevention $COPE

After testing out tons of image gen tools for character consistency starting from a single image, I found a surprising solution. Care to take guess? Yea...turns out Veo3 (Quality) was the answer. You can then take screenshots and generate LoRA out of them and use it with other tools.

This has to be one of the most interesting outcome ever...LOL No one knew @astronomerio before, but it would be hard to find anyone that doesn't now. Whoever runs crisis management should get a major raise.