Dilip

Dear Concerned Indian Doctors, I see your rage against Temple on X. I have a few points to make: 1. First and foremost, a business lesson: if Temple were a scam, and any of what we said weren't true, people would know it in just a few days of using Temple. We have 14-day, no-questions-asked returns (which even Apple doesn't offer in India). If the product doesn't work, it fails out of the gate and the company dies before recovering a tenth of its R&D. 2. We have already spent $30m building Temple, $20m out of which is my own money. Selling a few thousand devices recovers a small fraction of this investment. A scam that pays nothing is not a scam anyone would run. 3. I left the Eternal CEO role to chase purpose with Temple. This is not about money. I have enough of it to last for generations. My incentives are fully aligned with getting the science right. I am not staking a reputation earned over 20 years on making a quick buck. 4. Our employees bought into our round a few months ago. The round we're raising now is mostly our customers. Scams don't build that cap table. 5. Most wearable companies start with internal studies. Whoop shipped its band in 2015, and the first published validation followed a few years later. We hold ourselves to a higher standard; pre-prints are a step prior to peer reviewed publications, and they are coming up. 6. At some point you have to stick your neck out and say: we believe in our science, and the product is now valuable enough for people to pay for and use. We believe we are there. More validation continues in public. 7. We are confident that we are building something that the world is going to be positively astonished with. We stand by our science and our methods. Each of our claims is easily verifiable by individuals and institutions alike, and you will see a lot of independent studies come up on their own when the device becomes available to use. Last but not the least, the most ground breaking features in Temple are not even public yet. And some of them, thankfully, have nothing to do with doctors 🤭 Hope this addresses some or all of your concerns. If not, keep raging 😘🤗

Alright! I found time this weekend and today to reproduce these results. I used your harness completely (your replay tool, C adapter for libghostty, oracle, etc.). Thank you for providing all of it. The headline is I was able to reproduce most of your results, with one major asterisk: your harness configures libghostty-vt explicitly in zero-scrollback mode, which eliminates a lot of the optimizations we make. It is a fair comparison for alt screen, but not fair for "normal" terminal usage, I'd say. To cover the gap, I modified your harness to run with zero scrollback (yours), libghostty-vt defaults (10KB), and Ghostty GUI's defaults (50MB). I attached those results. With any scrollback, libghostty is significantly faster. I also noticed Hotaru gets continuously slower as scrollback increases, which causes terrible benchmark skew. I suspect you have an O(N) somewhere screwing you. Since Hotaru at zero scrollback had the best results, my graphs show that. Ghostty did not exhibit the scrollback slowdown problem. I spent my nights over the weekend closing the gap for the styled, hyperlink, and cursor/erase stuff, and I rebenchmarked main vs main just an hour ago. So this includes that. I want to be clear there WAS a gap here, and I fixed it. I really appreciate the great harness and competition. I hope you can now mog these results and we can continue! I attached three graphs. The first two are throughput numbers. One is your data set plus your runner. The other is `ghostty-gen` data plus your runner. I include `ghostty-gen` because it produces more random data with more knobs than yours. Its all in our repo, so you can reproduce (I'll show you mine if you show me yours kind of thing lol). The last shows your slowdown on your own input data. I didn't include Ghostty here at all since I can't get apples-to-apples timings here because your harness for this includes command startup and so on. Ghostty doesn't exhibit any throughout number changes though for any increase in history size, its constant. Feel free to reproduce, I look forward to hearing any feedback, and enjoy the high quality competition. I'm still working on more improvements, so just include the SHA of Ghostty if you benchmark again.

Little preview of something I've been working on! You can connect to any provider, or through any harness like Hermes Agent, and the setup is instant with no configuration required, unless you want to. Proactive, self-aware, and will be fully open source

A lot of Anthropic employees have been leaving within the past 2 weeks. This is not a good sign

i had 12 Claude agents running in parallel for 6 hours and i managed to reverse engineer @supabase i'm making it available here so that you can save your tokens - 100% free & open-source: github.com/supabase/supabase

We have worked closely with regulators, industry and operators in India for 5+ years, and built a network of dedicated systems, gateways, and other infrastructure all in-country to exclusively serve Indian traffic safely and securely - and all with the requisite government approvals. Starlink is ready to connect India with high-speed internet.

Ultrafast is so fast I do not ever want to go back

POV: you switched to a low reasoning effort model