XRG

Love how the 5 PCBs slot into each other to form the final shape.

With respect to the assembly: do yourself a massive favor and let the PCB fab also assemble the LEDs. With a simple bill of materials like this it will add very little cost, it saves you a _lot_ of time, and it might also reduce the chance of component damage from adding too much heat or from accidental ESD strikes. (I don’t know how much these Neopixels can handle, though.)

Heh, when I read the symptoms:

“Some exhibited “haunted” behavior, seemingly jumping to random sections of the mcu program code, outputting messages on the display that made no sense given the context. One of them appeared to work in slow motion, with LED blinking and display updates noticeably more sluggish than normal,”

my first thought was that it smelled like a clock issue.

Some of the nastier issues I have had the pleasure to debug included (a) traces that had microcracks which affected analog readings when the PCB heated up after prolonged usage (QC issue from the PCB fab) and (b) a (suspected) ESD strike that gradually took out several components in the weeks following as I was investigating the device while new problems kept popping up. Marginally stable composite amplifiers have also caused some headaches over the years.

Hardware really is hard.

As someone who does both hardware design (mostly high precision analog electronics) and software engineering, the difference between both fields in AI capability is striking. Models like Fable and Astra can provide lots of interesting insights into the design of specialty electronics when prompted correctly. Yet they can hardly put that knowledge to work if instructed to generate a schematic or lay out a PCB.

Of course, for readers here with a solid grasp on how LLM's operate it's not exactly surprising. Still, I feel this is exactly the kind of thing that drives home that the way we reason about (general) intelligence does not map cleanly to LLM operation, no matter how advanced the current models are.

Here’s one crucial difference: there’s overwhelming evidence that human emissions have an effect on our climate. The mechanisms are generally well-understood and the research is widely disseminated and easily available to anyone that’s interested.

With the ‘dangers’ touted by these insiders, it’s all “trust me bro”, hyperbole, and very little hard evidence. As such, a skeptical mind would question their motives.

Boredom may cause teenagers to commit acts of vandalism; it’s not a typical cause for revolution. (That would be things like famine, gross inequality, severe injustices and so on.)

From the article: “For well over a decade, the reliability data that exists (and I think this is backed up by the anecdotal experience that mechanics who work on Volvos have) is that Volvo reliability is mediocre to poor, but of course Volvo forums are full of people who insist that Volvos are among the most reliable cars and that the data are all wrong.”

The author mentions one explanation, which is that most people driving a Volvo won’t experience a breakdown (and then generalize from that). Another one is of a very different kind known as post-purchase cognitive dissonance (https://en.wikipedia.org/wiki/Cognitive_dissonance#Consumer_...).

It is not necessarily untrue (source: am currently involved in battery R&D). However, the degree to which it is true will vary strongly based on the particular choice of chemistry, what sort of additives/agents are used (e.g. scavengers/getters), cell/pack engineering choices (protection against ingress of outside atmosphere and such), and so on and so forth.

Goes to show that having a creation date placed somewhere near the top of an article or blog post is a good idea.

Dear ChatGPT, how can I patch a hole in my sock?

As a side-gig I taught within the doctoral education program for several high-ranking universities in Europe for about a decade (over a thousand PhD candidates). My impression is that nearly all funding for PhD projects flows to fields like medicine, physics, chemistry, computer science, electronics, and so on. Spending on humanities is absolutely minuscule compared to those.

My first guess was that they used an external brick for the power supply with a relatively low output voltage--that would eliminate a lot of the CE test load. However, a cursory glance at the product photos suggests the power supply sits within the base of the lamp. Maybe the product developer can shed some more light on this. ;-)

Cells specialize based on the presence of certain chemical gradients, for example. These types of 'signals' that guide specialization already become apparent in a very early stage of embryonic development. I don't think that's part of the realm of epigenetics.

I can’t imagine anyone in my environment becoming really angry over this. Some of them would have to find new time sinks though—-hopefully ones that actually benefit them instead of turning them into commodities.

Don’t agree with this take; software doesn’t need repairs when you drop your phone, when it is submerged, when it is charged/discharged daily for a few years, and so on. Freedom to use your hardware however you like is more of an ideological discussion, whereas repairs are simply necessary due to the unavoidable wear-and-tear.

If we assume the indicated 400 W load with a total processing time of 20 s per user query, and 20 mio queries per day (‘daily users’), shouldn’t the daily consumption be closer to 44 kWh? (20 mio x 0.4 kW x 20 s / 3600 s/h) Other than that, the energy use for LLMs is obviously substantial. I seem to remember that the operating costs should be fairly evenly split between hardware (i.e., GPU) depreciation and energy costs.