nijave

Hmm interesting I have 4-5 of them and quite like them. They've all stayed connected over a year except 1 which I just replaced the battery in and I think that fixed it. It was reporting 40% battery but would drop off shortly after connecting.

Aqara might be the ones with a crappy ZigBee implementation that doesn't roam right so if you move them or your topology changes, they'll drop off.

Nothing at the moment. Knowledge is a graph. You can try to shove it in a tree but it won't fit right.

Consider tomatoes.

Legally and culinarily, they're vegetables. Botanically they're fruit.

Tomatoes are members of the nightshare family which includes tobacco, potato, and chili peppers.

They are used in popular recipes like Mexican salsa and Italian pasta sauce. Pasta sauce commonly uses the San Marzano variety. Here's a <picture> of San Marzano tomatoes from our Italy trip.

Rhett likes tomatoes in all dishes. Link only likes tomatoes if they're blended in a dish like pasta sauce or tomato soup. Frank is allergic and can't have tomatoes.

Tomato prices are up due to a bad <year> yield in <country>.

---

This is all memory. Where does it go?

There are some systems that do that, but I don't know how effective they are. I've been meaning to evaluate AWS AgentCore Memory which has strategies you can apply for reconciling and organizing memories.

This ignores the lifecycle involved with memory. It doesn't appear to have a concept of time or reconciliation (both issues we're having at work). A memory might be relevant for a week or a month, but it might no longer matter after that ie "we're migrating systems so keep <thing> in mind"--that doesn't matter after the migration is complete. I guess the offered solution implicitly supports reconciliation since the agent can keep history of its memories and rewrite them, but it doesn't appear to be very first-class and it doesn't seem like there's anything intentional unless you prompt it "go check all the docs and make sure everything is consistent".

It also doesn't seem to have a way to separate preference from factual memory which is useful when you have various humans interacting with the same agent. One human might prefer a certain output style over the other and that's something a memory framework can also address.

Like the MCP articles a few months ago, it also seems to assume all agents are cli coding harnesses running on your local machine. We have a handful of other things like chat bots, event-driven agents running on servers, chat driven agents running on servers in sandboxes--there's not a single filesystem and even if there were one, having multiple agents try to edit it at once would corrupt it.

Also nice with tmux. I can split into a bunch of panes/windows, toggle through with keyboard shortcuts, connect from other machines

Performant and robust are relative. Better than browser or click automation.

>It was fairly obvious why MCP would be needed once AI engineering and uptake moved beyond the solo developer and single harness stack

This was already happening back in March. Lesser technical users were looking to hook up web LLMs to various SaaS apps so they could pull data in.

It was quite surprising people were forming such strong opinions when there were very clear and objective differences in use case.

Why is that the right tradeoff? It seems like the Plex/Jellyfin ecosystem has landed on keeping high quality persisted and hardware accelerated transcoding for everything else. Put another way, transcoding capable caches.

I'd be interested in seeing data that supports this.

Netflix reported in 2025 something like 30% of traffic was av1. I think smartphone and PC support are fairly robust but under the impression streaming devices are still quite terrible unless it's a Chromecast.

AVC/h.264 is the only thing I know of that's pretty much universally supported at this point. I found a Facebook/Meta report citing some consultant provided data that VP9 was a close second. Limiting to devices from the last few years, HEVC might work.

Quick search makes it look like there were some politics and ReVanced devs left and created Morphe. Seems more active but I'm also just learning about it

Sponsorblock is still super useful even with sponsorships left on. It's more like a "cut everything except the core content" tool.

Ref https://watchwithmira.com/blog/sponsorblock-categories-expla...

Cutting intro/outro/preview is nice if you're watching a bunch of videos on the same channel in a row.

That works if you fix everything to a single quality and codec. YouTube videos tend to have something like 20+ variants that can get dynamically switched between while the video is playing (adaptive streaming).

The client can pick the optimal encoding and quality which makes that quite a bit more difficult. A user on a modern device that can hardware decode HDR av1 will have a file cached completely useless to a browser that can only do SDR h264.

I think there is a sort of network effect needed to make it become reasonable.

Can't you just do "ota" updates like codepush used to do? ie shove a JavaScript asset with whatever is needed and allow the app to pull it at runtime?

Smarttube on Android TV just pulls its own APK so you just need to click Install after it downloads and opens it.

Long lived websocket seems to work okay. Iirc that's what the Home Assistant app uses

A place I worked back around 2020 was running a Grafana instance per customer that got embedded on the web dashboard. We had 110 pods per GKE (kubernetes on gcp) 4 CPU node because that was a network imposed pod limit at the time. The nodes were usually idle--could have shoved a lot more on if not for the IP limit.

I think around that time Grafana changed their license tho so you couldn't host OSS Grafana as part of your service.

For AWS, moto

That makes sense for the "make sure this thing is working correctly" case but there's still "I want to run this thing locally and it requires this other thing". The second one you really only need the happy path to work and you can run your test from the first thing against it (contract test).

A lot of apartments specifically forbid outdoor radio equipment outside which might make Starlink difficult. 5G should be fine

Rev sharing agreements can be quite lucrative for the building owner.

Here's an old example https://www.biggerpockets.com/forums/432/topics/554933-cable...

There's also https://github.com/ankane/strong_migrations for Ruby

The checks/explanations are fairly simple and straightforward so it also makes a good reference regardless of whether you're using the library

I think you could automate this with 2 transactions

- connection A, lock timeout=0, acquire unwanted lock

- connection B, lock timeout=0, run migration

- collection A, rollback

Then connection B will fail if it tries to acquire an undesirable lock since it will conflict with A. You'd be adding a very small window when you're actually holding the undesirable lock, though

Personally, a combination of low tech and semi scientific tests based on what you normally do (redo the same task you used an older model for with a newer one).

So far, Simon Willison's pelican bike "benchmark" is the only one I've found that shows Fable 5.1 beating Opus 5.5. My personal experience has been Opus is unusable on design work it's so terrible. Evaluating whether we should consolidate AWS DMS tasks (Postgres full load and change data capture) into fewer tasks with more tables, Opus 5.5 was factually wrong and needed correction roughly every other turn.

On a "help me find a sandbox solution for agents embedded in a web app to run untrusted code" research project it kept misrepresenting security boundaries and ended up recommending DuckDB which ironically specifically says it does not provide a strong security boundary in its own documentation. GPT 6 (can't remember if it was Sol or Astra) and Fable 5.1 both recommended FaaS like Cloudflare Workers and AWS Lambda which fit fairly well with the requirements.

I switched from Opus to Fable in the session going badly sideways and told it to "Review the previous conversation and come up with a correct comparison table and corrected recommendations grounded in objectivity supported by citations. Do research as necessary to understand the current ecosystem" and that was a full 180 back to coherency...

>What was hard there was to train the model head without ML expertise, and considerable amount of data

Which has become a lot easier the big, expensive general purpose LLMs

Some of the use cases seem interesting but a lot of them also seem like the "old" stuff can do it faster/cheaper

For the privacy use case, there's Stanford NER and wrappers around it. That is likely cheaper and faster than Jev (it'll run on CPU)

JSON or XML in with json schema for the response?

https://developers.openai.com/api/docs/guides/structured-out...

Not an ML guy so maybe missing something important here. It seems like many use cases could be covered by even dumber, simpler models created with the aide of general purpose LLMs.

Like, can't I just tell Claude or GPT to help me create a BERT or <insert old boring ML framework model> that's even cheaper and faster?

For instance, Frigate runs 4.6mb yolov9 image object detector on a $50 USB Coral stick. It takes like 1-2 watts of power, responds in 15ms, and is completely local. A pay-per-token cloud hosted not-really-an-LLM-but-probably-based-on-one that runs in 150+ ms still seems terribly inefficient.

Zigbee, Zwave, and Bluetooth (although a bit to a lesser extent) are easy mode of you want local only. They're not IP-based networks so no Internet access to begin with.

Bluetooth--local only but setup can be a bit trickier since they don't automatically form a mesh. You may need to configure relays if your existing devices have that capability or buy dedicated relays

My switches go up and down just like they always did. I can also toggle them from the app (ok so sometimes the switch is up and the thing is off but multi-way switches have been like that forever), track their on/off state, and use them as event sources to do other things.

There is a single magic switch that does "extra" stuff and that one's in the master bedroom. All of them except that one fail open and continue working like regular switches if Home Assistant is unavailable.

The livingroom has a Shelly dimmer that replaced an analogue dimmer and it also turns 2 lamps on and off.

The basement switch was wired to a single bulb and it triggers an event that powers all the smart bulbs in lamp holders spliced into arbitrary circuits on and off so all the basement lights work with the existing switch instead of just the 1 physically wired to it.

I think there's plenty of dumb things you can do with smart home tech but there's also tons of subtle and convenient enhancements as well. Phone alarm emits an event that opens the curtains and turns on a lamp is also a nice one.

Having a dashboard that shows whether (many) things are on or off and their power consumption is super handy. Dropping an Emporia circuit power monitor widget in the dashboard shows if the washer and dryer are running (a little bit of power, a lot of power, a lot of power + a little bit more power)

>I don't know if bragging about being AI-native is something any of us should be doing

Me neither but we now have a likert scale on our performance review for it...