I measured this on my past claude code sessions. I have a dataset with full LLM requests and responses and I ended up to an estimated 5-7% savings on tokens. Also with my primarily Opus 4.8 usage, only 5.9% of unique reads even qualify for the bulk-read operation and 58% of the reads were already targeted.
No favorite in single spark usage really. Glimmer is interesting because it scales well under load. Qwen3.6 MoE was fast enough on low concurrency. Qwen3.8 works but slower than the DS flash on the two sparks.
How do you like the AMOLED screen? I have a Fenix 6, which has the old screen. While it doesn't have a flashlight, I'm very happy with the ability to toggle the backlight on with a single button, which is more than enough for me to see in the dark where I'm going.
I feel that I'm lucky as my Fenix 6 still works as I would not want to switch to an AMOLED screen, which draws too much power to be kept on 24/7.
I would have preferred the old one because of battery life. I wear watch to know the time by just glancing on it, so always on is a must for me. It looks nice though.
Still, even with always on, I do 4-5 days between charges with normal use. I recommend watch face 'portal hybrid', which seems to have a slightly lower refresh rate (who cares), but you gain a day of battery compared to stock.
Curious, what have you built? I just wibe coded a packing/checklist app, a llm-powered meal planner and a very specialised app for tracking LAN party network building.
I vibe coded a battery monitor for my wireless mouse. It doesn't have Linux support and I wanted to know when to charge it.
I built an AllTrails equivalent (iOS app and all). It allows me to scan my backpacking gear to load it into a library (with metadata attached), build a weight profile for my trip, has offline maps, and more. I wanted to make something that was more suited to long backpacking trips and will eventually add itinerary planning. I was originally planning on trying to monetize it but I don't know how to sell software and honestly the QA process is so time consuming (hike, fix bugs, hike more, fix those bugs) that I just gave up. I plan on redeploying it to my homelab since I previously ran it on Hetzner.
That's so cool. I also thought a bit what it would mean to monetise the checklist/packing list app, but I just don't see much business case with it, with all the QA and support requirements.
Well, the monetization would be that you would have to have a subscription to use it over a long term - similar to AllTrails. Checklisting/packing/gear tracking was one dimension. The other dimensions would be curated backpacking routes and AI analysis of whether or not you were prepared for your trip.
The idea would be that you would scan all of your gear in and then create a map (or select one) and then Gemini would tell you whether or not you were prepared for that trip - does your sleeping bag insulate well enough, do you have the right shoes, etc. You could even do neat tricks like calculate expected route completing time based on assumed weight. I was also planning a "ask a human" feature that would snapshot all that data and post it to an internal forum.
Ultimately though, outdoor people hate AI (understandably). Selling it as "AI powered" is a no-go.
And I genuinely do not understand how to distribute software. Getting people to use your software is extremely hard unless it is very polished and has no bugs. The feature set was too similar to AllTrails so people will expect AllTrails quality. And something that I'm struggling with for all software that I make is I don't know how to get beta testers to get a better feedback loop. Going "hey my software isn't perfect but can you try it out and let me know what you think" is genuinely met with disdain for bothering the person, apathy, blank stares, or whatever. Doesn't really matter if it's offered for free or not - or even if the person expressed a desire for a solution. A person can say "wow that is a great idea I would love to try that" and I would follow up with them with practically no success.
At the end of the day I think I just want to give up and go back to big tech and optimize ad distribution pipelines to deliver weight loss ads to children or whatever tech is up to these days.
Have we reached the capability of a local STT+LLM system being constantly listening for normal speech in a room and being able to understand when the human is addressing the system instead of talking to another human?
Great article! After reading I started to think there must be some purchable items which would demonstrate colours outside P3 colorspace. It would be cool to hold one on your hand and experience how a photo of it just cannot do justice.
Anybody know any links to webshops for such items?
There's a few dyes (mostly pinks and cyans) that are outside sRGB space (though once you apply them to a material, that combination of material and dye might be fine), Tom Scott had a video a few years ago, though I don't know if that particular dye/seller is still going:
There's also Vantablack and similar ultra-black dyes which are a somewhat adjacent phenomena but in the same category of "colors your screen can't display". The most black a pixel can be is "off" (and that's only achievable if the display is on if your screen is an OLED), and the back coating of your screen is probably not as black as those dyes.
reply