+1 to using MUD as a learning tool. Learn about agent loops, prompting iteration, etc.
I vibe coded a few room proof of concept MUD. I added a tools called `oracle` (aka me) that it could ask me questions to help along its way; and the ability to interject in the loop with a hint. Just as I might want an llm to stop to ask me for help instead of plodding along...
Interesting to limit the number of turns and see where it gets stuck or how quickly it can finish.
I refuse to believe this is the only way to make it cheaper. Standardization and modular components could help. Might not be cheap enough. But surely we can be more imaginative!
I've spent countless hours searching for a small, good quality passive speaker. In a roughly 4 inch cube size. That comes with an good (acoustically) box/enclosure. Plenty exist in various forms (kits) but I never know if they sound as good as even laptop or phone speakers. Which is a low bar, but still can't find good reviews.
A 4-inch cube doesn't leave much room for a passive speaker to be very good (depending on one's definition of "good").
We've got hard physical constraints with loudspeaker systems, and they show up as tradeoffs.
Very broadly speaking, in practical terms, this takes the form of a tradeoff betwixt enclosure volume, efficiency, and low-end frequency extension. It's impossible to improve all three of these at once. This was formalized with Thiele and Small's work ~60 years ago.
We get by with amazingly good small, self-amplified, active speakers these days because -- as a system -- they don't have the same constraints.
At their root, they're still just passive speakers... but whole of an active system comprises more than that. We can use EQ to improve low-end extension. We can use bigger amplifiers to make up for lack of efficiency.
We can use combinations of dynamic EQ and bigger amps to make small systems sound pretty darned good even at low volume and hold together nicely at higher volumes, thanks in part of some of the work of Fletcher and Munson nearly a century ago and also to modern measurement systems and the inexpensiveness of implementing functions (with DSP in 2026) that would have been unthinkably complex in the consumer analog space at any point on the timeline.
But while active speakers can be pretty neat, small passive speakers suck. They have always sucked in some way. They must always suck in some way compared to their larger peers, and a cube shape always makes some aspects of the suck even worse. Them's the breaks.
---
Anyway, the market wants small, active speakers these days...so we get small, active speakers. Reviewers are people too, and they review what people want to buy.
If I may ask: What's driving your desire for a small speaker that is also passive?
A lot of words explaining how something is hard to do because physics but in my layman experience any speaker I heard that didnt cost $500+ sounded horrible regardless of size, while my M1 Macbook Air sounds better than all of those, while keeping tiny formfactor.
Clearly it can be done, its just noone seem to be investing in RnD enough.
It can be done on a Macbook because a Macbook is not a passive speaker.
A Macbook comprises an entire audio system, with amplification and DSP and everything else that a passive speaker -- by its very nature -- can never provide.
I explained this once. If you read through it, and take the time to understand it, and then think that you can explain it with superior brevity, then by all means: Feel free. :)
If you have questions, then: Please ask them.
If you think you can also defy what are commonly accepted to be real physical constraints and produce a tiny passive practical speaker that sounds great, then: By all means, do that as well! The proceeds of this kind of success can make you very wealthy -- maybe not Mars-mission wealthy, but probably at least beach house, private flights, and built-to-spec Porsche 911 GT3-level wealthy. You'll then be able to take time to decide what your next venture is, or perhaps decide to just live with that level of lavishness doing whatever makes you happy until your days run out.
My intuition is that a small radio from the 80s is sufficient. Has relatively small size and passive speaker. Small passive speaker and small enclosure are cheap aka not 100+ USD. A tonie box sounds fine. Can I diy reproduce one similar at low cost (less than buying 50 USD)?
I once read an article about SQL and how reordering the sections of a query would make it more ergonomic for users (iirc things like specifying what you want first and then how to present it last, more like a pipeline). I searched many times over the years but have not been able to find it again.
I mean that the body most won't respond to signals to reabsorb water from the bladder because of weak QVPR2 activity.
AVPR2 only shows up (significantly) in the kidneys. It pumps water back from the bladder to the bloodstream. AVPR1 and AVPR3 show up in the brain (and drive/control any hormonal response tied to water availability/quality/safety-to-access. Including territorial mammals marking territory.)
Vasopressin is actually a mirror-image of oxytocin, with a few hundred million years of divergent mutations. Unique to mammals. That's why mammals are the only vertebrates that independently, separately, regulate water and ions. (Thus, sweating, lactation, crying, uncalcified placenta vs egg, etc.)
So the brain keeps pumping out vasopressin (in response to dehydration-induced corticotropin-releasing factor). This leads to water "running right through". And high baseline vasopressin levels that go even higher with dehydration.
Less commonly, especially for men, the body overreacts to vasopressin. Excessive vasopressin 2 receptor efficacy or transcription leads to low baseline vasopressin. Low vasopressin 'magnifies' any oxytocin activity. Leads to "human-hyperstimulated" autism and high innate trust in unfearful situations. Also often leads to low territory/spatial mapping capacity.
A. Research dates back to the 90s. Thousands of researchers have added their perspectives.
B. Look at what vasopressin does, in humans and other mammals.
C. Consider that Vasopressin Receptor 2, is right by the X pseudoautosomal region, so is mutated 15x-20x more frequently. Compare how often you've seen other pseudoautosomal differences in people with Vasopressin transcription differences.
Or just start searching for random genes from around the pseudoautosomal region + autism. Transcription differences in one pseudoautosomal gene are so closely tied to differences in others, you see a very unusually high number of correlations between autism-uninvolved genes and autism.
(Just as so many genes around TNF-alpha/6p21.3 are spuriously tied to autoinflammatory issues; major histocompatibility complex issues; tenascin-X-tied disorders including Ehlers-Danlos; and 17-hydroxysteroid-dehydrogenase-8 variance [which deactivates androgens+estrogens, and synthesizes moderate estradiol].
Or how so many genes near the adjacent corticotropin-releasing-hormone-receptor-1 (CRHR1) and Tau protein (MAPT), in the same area as DNA-repair-gene breast-cancer-associated1 (BRCA1), are spuriously tied to irrecoverable oxidative cell damage and neurodegeneration.)
C...Some very important pseudoautosomal genes include the final stage of melatonin synthesis (ASMT & ASMTL); antiviral and anti-small-pathogen signal receptors, for interleukin 9 and interleukin 3; SPRY3 lymphoid-to-myeloid switch granulocyte-macrophage colony stimulating factor (GM-CSF); cytokine-like-receptor 2 (CLRF); glycogenin 2 (starting-point for muscle fibers); steroid sulfatase (activates androgens, estrogens, progestogens); sex receptor Y and protocadherin-11-Y (PCDH11Y) in men [causing heritable father-to-daughter changes in protocadherin-11-X and near-adhacent androgen receptor].
D. Also, check out the research from the last head of the Kinsey Institute. They hired her for that vasopressin+oxytocin triggers pair-bonding, and monogamy in monogamous mammals.
I have used a few of these for testing at work (I'm a developer). They're pretty solid devices, well built, audio quality is _subjectively_ pretty good and they have headset ports.
I dunno about "multiway merge". But your standard 2-way merge is SIMD-optimized by having a binary-search across the "diagonals" of the so-called mergepath.
Leading to O(lg(p * sqrt(2))) time for the longest comparison diagonal, where "p" is the number of processors. Including thread divergence, that's O(N/p * lg(p * sqrt(2))) total work done, or assuming constant-p, O(N) of work per 2-way SIMD merge (with a large "p" providing an arbitrarily large speedup: but in practice is probably limited to 1024 for modern GPU architectures. Since modern GPUs only really "gang up" into 1024-CUDA thread blocks / workgroups efficiently)
I vibe coded a few room proof of concept MUD. I added a tools called `oracle` (aka me) that it could ask me questions to help along its way; and the ability to interject in the loop with a hint. Just as I might want an llm to stop to ask me for help instead of plodding along...
Interesting to limit the number of turns and see where it gets stuck or how quickly it can finish.