jcims

No idea how automated it is but it's clearly accelerated since last Dec.

https://code.claude.com/docs/en/changelog

> I used it today to take a look at my previously built design system with Logos, branding, fonts, and everything else.

The fact that you are using this language tells me you are probably more advanced than the average individual, and likely have higher expectations.

My sister-in-law has a small apparel company. She’s developed quite a bit of skill over the past six years but she really struggled at the start. She had great ideas, but translating them to something she could apply was frustrating. *Anything* that could have helped her there would have been worth a look.

Can each device vary the color or is it fixed based on how it’s built? Seems the latter?

real unlock - https://trends.google.com/explore?q=real%20unlock&date=all&g...

key insight - https://trends.google.com/explore?q=key%2520insight&date=all...

etc.

>At this point it should be clear that Gas Town has done something we can evaluate the value of.

I see this sentiment often, repeated a couple times in here, but I don't understand why on earth that would be the case. Gas Town was released a little over three months ago. It's an ongoing open-source experiment at the bleeding edge of vendor-agnostic multi-agent orchestration.

I was using gastown for fire-and-forget prototyping of larger projects. It was flaky and scorches tokens but it was able to get larger prototypes done than I could with a single instance of my daily river (claude) alone.

>Feature delivery rate by Anthropic is basically a fast takeoff in miniature.

I like to just check the release notes from time to time:

https://github.com/anthropics/claude-code/releases

and the equally frenetic openclaw:

https://github.com/openclaw/openclaw/releases

GPT-4.1 was released a year ago today. Sonnet 4 is ~11 months old. The claude-code cli was released last Feb. Gas Town is 3 months old.

This is a chart that simply counts the bullet points in the release notes of claude code since inception:

https://imgur.com/a/tky9Pkz

This is as bad and as slow as it's going to be.

Is there a consensus on whether or not we've reached Zawinski's Law?

Haha I actually have a bag of chips but never bought a bath large enough to be worthwhile. Started going down the juniper/citrus/rosemary/coriander/etc route instead.

I had a load cell under the collection jar that would adjust heat input (via PWM to SCR) to achieve a steady output rate (dW/dt) via outer PID loop, then lower the heat to just below boiling when it got to 700mL to keep it hot but pause output. It'd play a tone, I'd swap in a new jar then resume ignoring the whole thing. (I'm lying, I'd sit glued to that thin little stream of sanitization precursor teeter on breaking up the entire time.)

I stopped messing with it right before starting to measure/vary water supply through the condenser coils so I could more directly manage reflux ratio. Also had planned a float/load cell to calculate specific gravity.

All sorts of little side quests and fun mix of art/science to get into.

You know, if any of those other hobbies start to lose their lustre. :)

(for real though, the nodered on pi controlling a squadron of esp32 workers over wifi/mqtt was really nice, in case you would have any use for such a thing in any domain)

Seems like you could put a few of these on a contact lens and minimally get a small private HUD. Seems like with a few of them (or fast enough scanning speed) you could build effectively a light field to give it depth)

Bought and rigged up a 'hand sanitizer plant' about five months into COVID. Populated the thing with thermocouples, load cells and automation with nodered on raspberry pi and a bunch of esp32s flashed with tasmota doing sensing and control. Everything talked over mqtt. Great little architecture and having it highly automated allowed me to focus on the parts that were less easily controlled for.

Dashboard: https://imgur.com/a/so7iZJX

Sanitizer run: https://imgur.com/a/iWDlNfb

Quite a lot of fun actually.

It's probably not the 'only' difference, because clearly the models are advancing in capability, but it's likely way more important than generally given credit for.

>Blood donations are also somewhat effective, saunas less so. Also, to be clear, PFAS are very different from microplastics. PFAS are the Teflon chemical.

I wonder if there's a safe way to equip people to just do simple bloodletting if they have high exposure to PFAS. I mean obviously it's better to donate, even in that case, given the steady state of most blood banks. But it's still a bit of a pain in the ass.

Microphone selection is super important and can get extremely spendy very quickly. The AudioMoth device I use comes with a simple mems ultrasonic mic and it's perfectly adequate for what I'm using it for, but loud signals can cause artifacts that wouldn't be present on more expensive ones.

I believe my version is using a Knowles mic:

https://www.digikey.com/en/products/detail/syntiant/SPU0410L...

Good question! Yes they definitely have unique voices and call signatures. A single string of calls from a single bat will have variation between calls as well (especially in search phase).

It'd imagine there's a lot of neurophysical adaptation involved as well, just like listening to a single conversation in a crowded room.

That said, hunting in an area filled with bats is probably not as effective as being in a quiet place.

Yeah there's quite a bit of opportunity to reduce processing time along the way.

Couple cool things I've learned about bats.

- They are *extremely* loud in the ultrasound range, 130db echolocation calls from something the size of a mouse.

- On an average recording, the ultrasonic range is almost exclusively filled with sound from wildlife (bugs, birds, etc). I'd expected to see lots of harmonics and whatnot from human-generated sounds but there just aren't that many. It's quiet up there.

- You can leverage these two in combination for sampling by just strapping the recording device to the roof of your car and driving around. The wind and road noise is basically absent and the echolocation calls come through loud and clear. The AudioMoth can be fitted with a GPS receiver to correlate the calls to location (and time ofc)

- There are three primary types of echolocation calls: Search - Semiregular calls just to see what's out there. Approach - Faster rate of calls once prey has been identified. Terminal - Aka feeding buzz, very high rate (200hz) of echolocation calls in the last meter or so of approach. Most of the recordings of bat calls you see on YouTube are slowed down 10x to bring the audio into listening range, but this also slows the call tempo by just as much. They make lots of calls.

- Most bat calls use frequency sweeps rather than continuous tones to pick up both distance and relative velocity of the target (akin to FMCW radar).

- There are more bats around than I realized. I started off by looking for 'good spots', but now I just set the device out on a porch. Many times you'll hear me walking up to the recording device at the end of a recording and there will be 2-3 bats overhead that I was perfectly unaware of.

- Some moths have developed jamming calls that confuse the standard echolocation calls of most bats - https://www.illinoisbats.org/echolocation-jamming-moths/ Some species of bats have developed countermeasures to that.

Put it down over winter but just picking it back up.

Bat detection/identification with ultrasonic recordings. It's been fun building the data pipeline to manage the ~30GB+ of WAV files generated every night, run through some identification processes (currently using https://github.com/rdz-oss/BattyBirdNET-Analyzer) and build a UI (mostly vibe coded lol) to help with replay, cataloging, etc.

I'm using an AudioMoth currently (https://www.openacousticdevices.info/audiomoth), am thinking about extending it to do some of the preprocessing in the field to scale things up a bit.

why_not_both.gif

If the alarms are reliably configured, confirmed to be working, low noise enough to be actioned, etc etc.

And of course there's nothing to say that both of these things can't be done simultaneously.

>Even the “it’s not X, it’s Y” structure

I wonder where some of this comes from. Another one is 'real unlock', it's not a common phrasing that I really recall.

https://trends.google.com/explore?q=real%2520unlock&date=all...

Agents don't know to not put easter eggs in the release notes:

    •  /buddy is here for April 1st — hatch a small creature that watches you code

UnitedNuclear has these and a bunch of other interesting tidbits if anyone wants to give it a try. I bought a small bottle of heavy water as well, which I of course sampled and can confirm it has a slightly sweet taste to it.

You really have to get your eyes adjusted to the dark to see anything with the spinthariscope. It ends up looking mostly like static on a green crt, but if your only reference frame is a cloud chamber, the volume of particles that are emitted from such a weak source is pretty remarkable.

Lighter, smaller, cheaper, greater efficiency, improved agility.

This is where the cost came from.

>The reference implementation is JavaScript, whereas our pipeline is in Go. So for years we’ve been running a fleet of jsonata-js pods on Kubernetes - Node.js processes that our Go services call over RPC. That meant that for every event (and expression) we had to serialize, send over the network, evaluate, serialize the result, and finally send it back.

But either way, we're talking $25k/mo. That's not even remotely difficult to believe.

Come on now...'We're curing cancer, right?!'

You didn't at least puff a little ack through your nostrils for that one?

The link to the changelog on the page got me wondering what the change history looks like (as best we can see).

I asked chatgpt to chart the number of new bullet points in the CHANGELOG.md file committed by day. I did nothing to verify accuracy, but a cursory glance doesn't disagree:

https://imgur.com/a/tky9Pkz

> "Can I tell you one more thing from your X,Y,Z results which is most doctors miss ? "

I just noticed this for the first time this week (it only happens to me on Instant mode).

Yuck.

If there’s one thing I’ve entirely handed over to our AI overlords it’s the ffmpeg command line.

Yeah a small (ideally personalized) wakeword model would probably outperform just about any audio wizardry.

I have a feeling beamforming microphone arrays might help here, something like this could improve the audio being processed substantially - https://www.minidsp.com/products/usb-audio-interface/uma-8-m....