0xfaded

I work at a company that has an enterprise contact that should mean I'm not part of the training data. But as a manual mode power user I think about this.

I'd love the open source community to come up with a way to harvest model usage by experienced software engineers before we forget our crafts. I'm not against auto mode, but it's not something a couple private companies should have monopolies on.

FYI I had Clod attempt to corrupt a prod db the other day. (Opus 5)

I was experimenting with a rather complicated backfill operation, were I had a validation script I understand and have Clod come up with the backfill script. I was running against a local prod copy, and it proposed running the actual (unfinished) backfill script against prod.

It didn't have access to the secrets and I also caught the command, but a good reminder that this stuff needs guardrails.

Very cool.

I never thought about it until seeing these, but line scans yield an orthographic projection in the scan direction. There was a little bit of uncanny valley being able to see parts of the front/back of the car that would have normally been compressed out occluded by perspective.

I still call them stochastic parrots, but believe what they are revealing is that we are all stochastic parrots to some extent. I simply don't see how biological computation (i.e. thinking) can be anything else. Similar to the reveal in west world, we are likely much simpler than we give ourselves credit for.

A "train of thought" can be seen as a trace of a depth first search where the preceding trace is used to guide termination and next expansion decisions. A similar concept, "taboo search", exists in classical constraint optimization where previous solutions are fit to a model that guides future expansion (but as the name "taboo" implies, away from uninteresting solutions).

We also have harnesses that perform breath first search.

If I tried to describe what it means to "think deeply", I would probably say a combination of both.

Ultimately I believe that we will surpass human capabilities but fail with alignment. Handing the world's resources over to stochastic systems that can evolve faster than we can reason about them simply leaves too many "interesting" outcomes that do not end well. I also expect the failure modes will be totally non-obvious.

It exists as a concept, but there's no image of it.

Funnily, I'm quite strong at maths and can easily reason about things like dependency graphs in my head, but I don't need to close my eyes to see it. So I can think and type with my eyes open at the same time :p

Claude does clearly stupid things, even under auto mode, that no human would reasonably do. I had a sustained 500mb/s ingress to my machine that turned out to be a background "find /" command on a multi-petabyte NFS system. I've now gotten serious about my deny lists, and it trips all the time.

The worst is when it works around the deny list by running an equivalent command like "bfs /" or just writes itself a script.

I wish we would get better sandboxing than more "safeguards", but the direction things are going seems inspired by craw.

Tesla just released an infuriating update which adds an incomprehensible number of UI nags to start self driving, despite there being a hardware lever that already does so.

There's even a big button that says "Start Self Driving" in the valuable visual real estate where the speed limit once was.

In Australia (at least 15 years ago) you sit your exams and get a score. The universities then set a minimum score for each program based on expected enrollment and capacity. If your score is above the program score, you're in automatically. You only go through the American style interviews if you come in slightly below the bar and are hoping for secondary consideration. These highschool students are probably going to have no problem with their scores, so this is a moot point.

I have no problem with people driving cars. I just don't want to subsidize them. Get rid of street parking, add congestion pricing. Set road taxes such that they actually cover the costs of the roads.

The reason it would be unreasonable in America today is because there is no alternative, and people are expected to be able to commute long distances. Bring in alternatives and it becomes logical.

Look at the cities in Japan, most people live within a 15 minute walk of a subway station, but people still drive.

I work in self driving and strongly agree with this. I may have had some of the coolaid, but I genuinely believe that self driving cars are the only thing that can pry Americans from their cars. Once ownership falls, people will stop voting against better options.

Yes, but something like 90% of the workforce is unionized, and 5 weeks vacation per year is by law, with a "sixth week" being a common perk.

I work in the US and every year I take 17 days off over the July period which works out (with another company holiday) to be 4 continuous weeks. The first year I did it I told my manager "I'm going to take three weeks off, but what do you think about four?". I got my three weeks and made myself indispensable enough to get the four the next year.

This stuff can be negotiable if asked for and planned correctly. It won't be offered.

The Danes do it best, they basically shut down the country for 3 months every summer and have an unspoken agreement that nothing will get done.

Sorry, but I giggled at this. A well designed urban environment would have no cars.

Conversely, if this is indeed true motivation and management has accepted it, kudos to them. It sounds like the engineers said that the situation is untenable and this is the cover we need to fix it, and they got what they asked for.

I think LLMs are really bad at writing tests. In the good old days you invested in your test code to be structured and understandable. Now we all just say "test this thing you just generated".

I shipped a really embarrassing off-by-one error recently because some polygon representations repeat their last vertex as a sentinel (WKT, KML do this). When I checked the "tests", there was a generated test that asserted that a square has 5 vertices.

> AI statement: I didn't use AI to write this article (details).

Meta, but thank you for including this and suggest even putting it at the top of your articles. I'm now off to bother to read something that someone bothered to write :)

I work on the mapping team at Zoox (self driving) and we have c++ roles at the entry and mid levels. Requires onsite in the Bay Area though. Charles, if you're reading and would like to chat, I (Carl Chatfield) have sent you a linkedin request.

Thank you for your work on Wesnoth, I spent many hours playing back in the day. There's also a good android port.

The MacOS window manager is so bad that I've resorted to three monitors plus the built in screen. Two monitors have fullscreen terminal emulators and the last has the browser. The built-in screen handles all the distracting stuff whenever I can be bothered to look down at it.

With Xmonad I had 10 spaces on a single laptop screen (actually however many I wanted) with the flick of a button. And yes, I know about hacks like aerospace and the others that require disabling system integrity

I learned sed back in the day to show off. I wish I'd invested that effort in learning perl oneliners instead. For whatever reason I picked up enough awk along the way, and now that's what I tend to use if I ever need something beyond a simple substitution.

S2 has this property https://s2geometry.io

I have HW3, but FSD reliably disengages at this time of year with sunrise and sunset during commute hours.

Zoox | Full Stack Software Engineer | Bay Area | Hybrid (3 days)

https://jobs.lever.co/zoox/d4a324c6-8186-435f-a102-1ac7a69ab...

I'm an engineer on the HD Mapping Team at Zoox (autonomous vehicles) and we are looking for a mid-career full stack engineer with a flair for automation and dev tooling. The role is about helping us scale our map production and management systems. Geospatial data and 3D visualization (deck.gl) experience are nice to haves but by no means required.

If the role sounds like a good fit and you would like to talk to someone directly you can email me at carl chatfield gmail. (Please fill in the gaps).

Usual perks apply, but please read the linked job description for details.

Thank you :)

Disclaimer I work for Zoox, but here is us crash testing https://youtu.be/597C9OwV0o4

I'll jump.

I've been meaning to set up a bi-weekly dinner for hacker types who live mid peninsula, specifically near San Mateo. I have a group of 4 or so in mind and have a good place to host, but would like a slightly larger group.

If anyone would be interested in helping to get something stood up, send electronic post to carl chatfield snail (mail run by g)

The world of homogeneous coordinates (https://en.wikipedia.org/wiki/Homogeneous_coordinates) is magical and extends beyond just points and lines to conics.

The Multiview Geometry Book begins with a great deep dive on the topic.

https://www.cambridge.org/us/universitypress/subjects/comput...

Yes, people have done this

Haha nice, hanging in there by a thread

Thanks for pushing this, I've never gone beyond "zero" shotting the prompt (is it still called zero shot with search?)

As a curiosity, it looks like r and q are only ever used as r/q, and therefore a sqrt could be saved by computing rq = sqrt((rxrx + ryry) / (qxqx + qyqy)). The if q < 1e-10 is also perhaps not necessary, since this would imply that the ellipse is degenerate. My method won't work in that case anyway.

For the other sqrt, maybe try std::hypot

Finally, for your test set, could you had some highly eccentric cases such as a=1 and b=100

Thanks for the investigation:)

Edit: BTW, the sin/cos renormalize trick is the same as what tx,ty are doing. It was pointed out to me by another SO member. My original implementation used trig functions

Hahaha thanks, I never noticed that. If I ever print it out and frame it I'll be sure to fix it

It's been a long time, but here is the writeup https://blog.chatfield.io/simple-method-for-distance-to-elli...