Benwen Blog.

AI, AI, AI. That's the answer. Now, what was the question again?

Sat, 07 Feb 2026 00:00:00 +0000

When some new techno trend “foo” arrives, I find myself repeating the same chestnut: “Foo, foo, foo that’s the answer, now remind me again what was the question?” In my career there was an XML phase, a NoSQL fever, a Cloud boom. That was followed by Node.js, Docker, Blockchain, and for the last two years, AI. This AI one has lasted longer than the rest (barring the Cloud one).

Now with the arrival of OpenClaw and orchestrators like Gas Town, the AI trend looks to be here for another year or two.

These open source projects are early signals for the next area of intense growth in software: long-running multi-agent systems. Both are going to be a huge sink for tokens. They will burn a lot of tokens. OpenClaw author at the first Claw-Con this week: “Ya, more tokens. Someone has to support Anthropic.” Steve Yegge in his latest Medium post The Anthropic Hive Mind: “But there is a yellow brick road: spending tokens. This golden shimmering trail will lead your company gradually in the right direction.”

Clojure Conditional Breakpoint on Passed Function Name

Wed, 07 Jun 2017 00:00:00 +0000

Conditional breakpoint in Clojure to match a passed function’s name/class, using Emacs and Cider. Clojure 1.8.0. CIDER 0.13.0.

Insert this in front of the expression you want to break on.

      #dbg ^{:break/when (re-find #"draw_row" (str (class f)))}

Clojure recur and variadic functions

Thu, 01 Sep 2016 00:00:00 +0000

tl;dr: for a recur call to a variadic function, wrap the variadic parameters in a (list ..)

I am writing this for my future self because I’ll stub my toe on this again if I don’t. Future self, you’re welcome.

So, I was editing a function to use recur. It was my first time using recur with a variadic parameter.

WTF. The recur invocation threw an error. Walking in the debugger confused me further. What saved me from frustration was this in the middle of the community documentation for recur.

; Note that recur can be surprising when using variadic functions.

And in the official docs

The recur expression must match the arity of the recursion point exactly. In particular, if the recursion point was the top of a variadic fn method, there is no gathering of rest args - a single seq (or null) should be passed.

Well, good to know.

And here’s the function I was refactoring.

(defn cleave-repeat
  "Finds a repeating pattern in string s using r as the next attempt to try."
  [s & [r]]               ; http://clojuredocs.org/clojure.core/recur 
                          ; recur with variadic does not wrap r in
                          ; a list. So a normal function call and recur'd one will otherwise differ.
  (if (empty? s)
    r
    (if (empty? r)
      (cleave-repeat (apply str (rest s)) (str (first s))) ; This calls the variadic "normally"
      (if (empty?                       ; We have the pattern in r if this is true
           (clojure.string/split
             (apply str (drop-last (mod (count s) (count r)) s))
             (re-pattern r)))
        r                            ; Return the pattern
        ; 'recur' calls with varadic not wrapped by a list, so (list ...) is needed.
        (recur (apply str (rest s))
          (list (clojure.string/join [r (str (first s))])))))))

Damn you JVM.

Boot and Vagrant

Thu, 28 Jul 2016 00:00:00 +0000

Boot is a new-ish software build system designed for and built with Clojure and ClojureScript. Here are some notes on using it with a Vagrant session on a Mac.¹

This is very much a work in progress. I generally like working with Vagrant to keep my development environments packaged, isolated, and repeatably built. No exception while I learn Clojure / ClojureScript, and as I decide on the surrounding tooling like boot. The main hurdles were to get the REPL and reload connections to be exposed to Vagrant network forwards.

build.boot

;; define dev task as composition of subtasks
(deftask dev
  "Launch Immediate Feedback Development Environment"
  []
  (comp 
   (serve :dir "target")
   (watch)
   (reload :port (int 3001))            ;; ports 3000-3003 are forwarded by Vagrantfile
   (cljs-repl :nrepl-opts {:port 3002 :bind "0.0.0.0"})
   (cljs)
   (target :dir #{"target"})))

:port in reload needs to be cast to an int.
I failed to get bash to pass the EDN parameter {:port 3002 :bind "0.0.0.0"} to cljs-repl for the nREPL port on the command line, so creating a deftask is necessary.²
The :bind parameter in cljs-repl overrides the default localhost, which is not visible to Vagrant.
Run with boot dev.
Thank you to modern_cljs tutorial for the source task.

Vagrantfile

  config.vm.network "forwarded_port", guest: 3000, host: 3000 # web server
  config.vm.network "forwarded_port", guest: 3001, host: 3001 # reload browser signaling
  config.vm.network "forwarded_port", guest: 3002, host: 3002 # nREPL 
  config.vm.network "forwarded_port", guest: 3003, host: 3003 # bREPL 

Emacs

To connect to the guest nREPL for the CLJS environment

M-x cider-connect
localhost
3002

M-x cider-connect doesn’t (yet) bind the host cider environment to the guest REPL, so sadly a lot of the magic isn’t working yet, like cider-eval-defun-at-point

In the nREPL, to connect to the bREPL

boot.user> (start-repl :port (int 3003) :ip "0.0.0.0")
<< started Weasel server on ws://0.0.0.0:3003 >>
<< waiting for client to connect ... Connection is ws://localhost:3003
Writing boot_cljs_repl.cljs...
(open host browser at http://localhost:3000)
 connected! >>
To quit, type: :cljs/quit
nil
cljs.user> (js/alert "hello")
(dismiss popup in host browser)

Detailed use case

a Clojure application server in development mode on port 3000
the guest virtual machine is running Ubuntu Linux.
web application runs ClojureScript
web browser is on host Mac
Emacs running Cider is the development environment running on the host Mac
Cider to nREPL on guest, chained to bREPL to the browser for debugging / inspection.
Boot tasks: serve watch³ reload cljs-repl cljs
Guest OS: Ubuntu Trusty64. Host OS: MacOS 10.11.5. Vagrant version: 1.8.1. Provider: vm_fusion. Hypervisor: VMware Fusion 8.1.1. Clojure: 1.7.0. Java: SE Runtime build 1.8.0_91-b14 and HotSpot 64-bit Server VM build 25.91-b14, mixed mode. Emacs: Aquamacs 3.2 GNU Emacs 24.4.51.2

TODO: create a GitHub repo with an auto-provisioning Vagrantfile as a nice template. ↩
boot cljs-repl -n "{:port 3002 :bind \"0.0.0.0\"}": clojure.lang.ExceptionInfo: clojure.lang.PersistentVector cannot be cast to java.lang.String. ↩
The guest filesystem watch is blind to changes made from the host’s Vagrant synced files. Sadly this currently means that touching a file from the guest to kickoff a build. ↩

Multisensor Data Fusion

Fri, 13 Feb 2015 00:00:00 +0000

Recently, I’ve come across this technical term that’s crops up a lot in the engineering world of IoT and mobile devices: Sensor Fusion. In its nominal usage, it means the ability to combine the output of common multiple mobile sensors like gyros, compasses, and GPS into a synthesized, more application-useful signal. A more precise term for that would be “multisensor data fusion”. That is, data from multiple sensors being combined together.

More broadly, sensor fusion can include synthesis of outputs of a single sensor over time, or an array of similar sensors in an environment, which would include human stereoscopic vision.

I’m largely an application software kind of guy, so these terms are novel to me.

I like to take concepts to an usually absurd apex or nadir¹. So then, would the human brain then be an exquisite sensor fusion implementation?

“Multisensor data fusion” can be sung to Supercalifragilisticexpialidocious. ↩

Oculus Vertiginous

Thu, 12 Feb 2015 00:00:00 +0000

Unfortunately, I’m in the category of vertigo sufferers when donning the Oculus Rift. I tried one on tonight for the first time. I had it on for maybe 10 minutes before I had to take it off. I almost had to look around for a barf bag.

A bit of a disappointment.

Back in 1995 I had a prototype head-mounted display shipped from California for a Virtual Reality project we had undertaken. I distinctly remember the hand-wrapped package made from two boxes taped together haphazardly. The goggle’s plastic was coarse and a bright light green and black. The duckbill shaped front was unbalanced. The latency was on the order of hundreds of milliseconds and had terrible angular sensor resolution. I think the screen had poor color saturation and contrast. I don’t think it ever made it past the prototype phase.

We had other prototypes and products, like the Virtual I/O, which may have been the most popular at the time.

The Rift is far superior than those, but I still sense a very hobbyist kind of experience.

There is some kind of irresistible siren song of 3D virtual reality goggles for each generation of programmers. Prior to the Virtual I/O generation, there were Evans and Sutherland displays that had to be mounted on a articulated arm to support the weight.

I may no longer be in the demographic for these, but I understand the appeal. So here’s a toast to the next generation!

Apple's New Market

Wed, 11 Feb 2015 00:00:00 +0000

Apple’s New Market

Ben Thompson adds Apple’s recent astounding quarter to evidence against the Clay Christensen’s low-end disruptive innovation model… at least applied to Apple’s business. He ends with a sly reference to the further disruptive possibilities of the Apple Watch. To which, I agree heartily.

Makes me wonder: What other businesses could this dis-commodification apply?

Naming Things

Tue, 10 Feb 2015 00:00:00 +0000

In roughly the order they spilled out of my brain. There are lots of other considerations that go into a name of a company or product. Here are a few acid tests to throw on your names ideas.

Evokes your desired brand characteristics (e.g. strength, humor, compassion): +5
Is easy for most people to spell: +1
Is “scalable” meaning there are a variety of natural sub-names +1
You have to spell out your URL in phone conversation: -2
Your 15-second radio spot uses 10 seconds to get the URL out -2
You rely on interCaps as a pronounciation hint: -1
It’s trademarkable: +2
Is available across all relevant media +2
References an obscure Dennis Miller schtick: -25
Unless you’re marketing to Dennis Miller fans exclusively: +24
Uses phonetic dopplegangers (e.g. Lyft) -1
Unless you have a big marketing budget +1
Uses easily confounded letters / numbers -1
Uses 1eetspeak: -1
Has punctuation in the name (hypens, periods, exclamation points etc.) -1
Your mother pronounces it correctly on sight: +10
Your mother spells it correctly upon hearing it: +15
You overly agonize: -1

ATT Telegram

Mon, 09 Feb 2015 00:00:00 +0000

Apparently ATT still stands for American Telephone and Telegraph company. This is the back of an actual ATT u-verse Internet service order form:

To cancel this transaction… send a telegram to the address below..

Entire form

Robo Speed Limits

Sun, 08 Feb 2015 00:00:00 +0000

Seeing a Google self-driving Lexus on the highway, I thought, “Will they ever be caught speeding?”

And if so caught, or involved in an accident, will the memory or configuration of the computer or GPS be admissible evidence?

Would that be an incentive for manufacturers to install a speed governor?

Would a sufficent fraction of speed-governed self-driving cars on a stretch of highway cause median non-robotic drivers to otherwise slow their pace?

Would the governor be a deterrance for rushed or thrill-seeking drivers¹ from enabling the computer, wanting the freedom to exceed the legal speed limit?

With years of safety data, assuming any needed improvements in safety, suggesting that self-driving cars are safer at speed, would there then be a relaxation of the law thereby allowing robo-speeding, effectively penalizing human speeders? A side benefit would be higher vehicle densities (with greater road wear) per unit time.

Would that make the manual driving of a car seen as reckless as not wearing a seatbelt?

I, for one, welcome our new fast and furious overlords.

Not me, officer. ↩