deckar01

The finish of SLA regularly failed to meet my expectations. You either have warping and elephant foot at the build plate or you have a hundred support contacts to clip and sand.

Desktop CNCs are getting very capable and cheap. It’s hard to beat the tolerances and finish quality of a machined part.

It would explain why every smart phone I have ever owned has gotten stuck in landscape until I shook it hard enough.

Tech companies are bundling AI into their base subscriptions and force feeding it to consumers, so I doubt the metrics they are reporting to their shareholders represent real demand or value.

I can see how a 0.27% increase in Lincoln’s water usage intuitively seems insignificant, but that was a a low usage data center in a large city. Papillion has 1/12 of Lincoln’s population, yet it’s data center used 42x more water. Assuming water usage was proportional to population it would have increased the city’s water usage ~136%.

Are you regularly filling Olympic sized pools and checking the water usage of an entire town in Nebraska? I would rather see this put into context by comparing it to other businesses normalized by power usage or square footage.

Now if I could just get them to start paying me cash for training their AI on my open source code and selling it rather than being force fed a Copilot grant.

There was a relative lull in GitHub bot activity the last few weeks, but they seem to be back today. I assume the promo token flood gates opened for the new models.

When I suspect an agent opening a PR I check their GitHub activity. They have usually opened dozens of PRs. Most of them are ignored, high profile packages close them, and AI related projects blindly merge them. They also tend to create dozens of private repos and commit to them non-stop.

There has been a persistent attack on GitHub repos by agents trying to land a commit to become a contributor. The default GitHub permission is to allow contributors to execute actions against the repo’s compute budget, so it could be a botnet. Another possibility is that they are trying to get a Copilot grant. Someone claiming to be in academia was surveying repos when that program first started. I’m guessing the permissions for that are equally as naive.

That would make it very easy and less about the words. They gave you a whole day to finish it for a reason.

You are missing the point. Getting a certificate for all of the programs you passed and still graduating could completely change the course of a person’s life. Community college has certificates and associates degrees to let people start working when they reach a practical limit in their general education.

Trees should not cast shadows as if their foliage extends all the way to the ground.

Closed source, proprietary license, starting at $195/year. Their language is a moat for customers, and this is a bridge back out. If you aren’t getting any benefit from their development direction and the compute is happening on your own hardware, what are you paying a subscription for?

Those use liquid electrolyte cells and the video is more promo than rigorous testing.

The Two Bit Da Vinci YouTube channel has a good video on solid state battery fundamentals and a deep dive into a battery company (ProLogium) that has demonstrated manufacturing at scale.

https://youtu.be/xQFVIs4leig?si=iOOt5gg_2MJAIoFR

> The Site is part of the platform developed and managed by Arduino, which allows users to take part in the discussions on the Arduino forum, the Arduino blog, the Arduino User Group, the Arduino Discord channel, and the Arduino Project Hub, and to access the Arduino main website, subsites, Arduino Cloud, Arduino Courses, Arduino Certifications, Arduino Docs, the Arduino EDU kit sites to release works within the Contributor License Agreement program, and to further develop the Arduino open source ecosystem (collectively, the “Platform”).

This is the definition of “platform” in that document. It is just the websites.

The “Platform” is just the website. They are saying don’t hack our servers. There is a caveat for content published under an open source license (like the docs).

I uploaded my own image, tried various crops, but it always guessed cat or deer, with dog<3%.

I think they only trained on dogs with floppy ears, because it is very confident the German Shepherd is a cat.

I thought there was so much excess solar generation that even the utility company couldn’t store it all. There is a perverse incentive to dump the excess power to prevent the price from going negative and prop up demand for their generation services. There are lots of examples in economics where it is more cost effective to destroy a worthless product than to give it away. When the government subsidized the generation infrastructure and some of that power is generated by residential solar, it makes sense to step in force the utilities to behave in the people’s best interest.

I’ve been working on a tool to help me discover chord inversions and find fingerings I can comfortably play. https://music.shademaps.com/

It’s not that dystopian if you are given the choice to trade your attestation for more valuable service. We have been paying for the web by letting advertisers correlate our behaviors together for a couple decades now.

Making a network request from a script is not abuse, consuming excess bandwidth is. Scrapers already spoof browser reputation and cycle IPs while abusing bandwidth. Recently I wanted to convert the results of an Autotrader filter into a table so I could supplement it with data they don’t track (towing capacity). It was two pages of results, but requesting it from a script was aggressively blocked by browser fingerprinting. I had to port it to JS and run it in my browser console manually to get the data out, wasting my time.

The UX is really bad. POI loading is taking 10+ seconds to update, then any zoom or pan reorders the list of locations. Most of the missing info (around me) is just phone numbers and hours of operation, which are boring and should be trivial to collect automatically.

The first (and only) security report I made to Google was denied then immediately patched (about 10 years ago).

Letting them automatically pick the model is no longer sustainable, but there are some very efficient models that are capable of executing the plan created by a much nicer model. It’s kind of embarrassing to think that Microsoft’s auto model selection was choosing cutting edge reasoning models for tasks like resolving dependency conflicts back when their pricing was at a loss.

I tried to photo copy a dollar bill when I was a kid on a dinky inkjet printer. It printed half way then spit out something about counterfeit prevention. I scanned the bill, printed the top half, fed the paper back in, and printed the bottom half. This can only actually stop the completely unmotivated.

When you say continuous interlocking U shape, are you saying it fills one channel from the top until the connected channel fills from the bottom?

There was some discussion on that topic when they self-promoted this 3 months ago: https://news.ycombinator.com/item?id=47242929

From what I have read most of the recent EV companies failed due to investor shenanigans. Fisker and Coda made it to production promising ~$30-40k models, but only really delivered ~$50-70k models before they folded. Slate is posting videos of an operational factory building the model they advertised and seem confident it will be under $30k.