rsp1984

- For services that you never signed up for (there are certain institutions in Germany that can bill you for stuff you neither want nor need but they have the full force of the government behind them): IHK, GEZ (Rundfunk), Verwaltungsberufsgenossenschaft (in case white collar workers hurt themselves by typing on the keyboard...). If you refuse to pay, they can literally put you in jail.

- For services whose necessity is entirely generated by the government: 1k EUR for the Notary to read you some boilerplate text. 500 EUR per month for bookkeeping services (govt creates all these arcane rules to make sure you never want to do it yourself). Plus many legal bills because German business and employment law is a minefield and you don't want to blow yourself up.

- For tax on imagined revenue that you didn't even realise yet. Yes, the govt. literally makes things up and says "that's probably going to be your revenue this year and you owe us tax on that".

If you're a big company with large predictable revenue and departments of people to deal with this, I guess it's tolerable, but there is no doubt in my mind whatsoever why the vast majority of successful startups exist outside of Germany.

You're making it sound like it was a switch that I could simply turn on/off as a web dev. The reality is that most sites are a complete clusterf*k of 3rd party components, dependencies, backend services etc.

Doing an audit of which of these components are compliant with tracking / cookie laws is just not a realistic ask. Hence why devs decide to just tack on a cookie banner and call it a day.

Why are they still in business? Because the law prevents more efficient solutions. Why are laws made in such a fashion? Because said industries offer lucrative deals to lawmakers if they give in to their lobbying.

The corruption in the western world is deep and pervasive. It's just one level removed from the public perception s.t. the price we all pay for it is hidden.

[1] https://gotit.pub

The cosine interpolation simply ensures that the interpolated-function derivative is 0 at the control points. If anywhere in the sequence there was a point that is lower / higher than both its predecessors the downsides of this technique would become obvious.

Personally I see zero differences in the images on that page and unless the author has some really super-human vision abilities (possible! but unlikely) my guess is he doesn't either. WebP looks perfectly fine to me.

Running non-trivial ML workloads on the edge has been on my wishlist for years and it sounds like Apple has just the thing.

It's useful either as a preprocessing step that you feed into further NN processing or -- super old school -- you are modeling your problem by hand instead of letting a NN figure it out (for better or worse).

The Qt Creator bug tracker [1] is bursting at the seams but nobody seems to really care. 5+ years old, very reproducible, very annoying bugs are being ignored by the 100s.

Here's an idea: can we have a feature freeze for like 1 or 2 years and only solve bugs / tech debt? There's no new feature that you could put into Qt Creator that I'd prefer over bugfixes. None.

Eike, or others from the Qt team, if you're reading this, please give it a thought. You have a fantastic IDE and it's painful seeing it degrade so much. Thank you.

[1] https://bugreports.qt.io/projects/QTCREATORBUG/issues/

Because if so, why not just use one repo per app/library/tool? Sure, if you have a cluster of things that all depend on each other, or a cluster of things that typically is needed in bulk, by all means, put those in a single repo.

But putting literally all your code in a single repo is not a very sane technical choice, is it?

Thanks for the clarification of the uploading / compilation step.

Keep in mind that the venerable (and enormously successful!) gradient descent method does not model dependencies between variables either and manages to find solutions too. It just has to iterate a bit on it -- actually not unlike the method presented here.

1. Simple Gradient Descent is already linear in time w.r.t. # of parameters (but it is an iterative method). It seems to be missing from Table I too. This method requires waiting for equilibration, so could it be seen as another form of iterative method? If so, wouldn't the proper comparison be against known O(n) first-order iterative methods like GD as opposed to exact methods (O(n^3)) or pseudo-2nd-order iterative methods like CG (O(n^2))?

2. In 2.B.3 the paper says "Note that this includes a compilation step that scales as O(d2)". I think this needs some clarification. Is this saying that, in order to run this on actual hardware, there's a compilation step involved that is O(n^2)? Of course O notation says nothing about linear factors but wouldn't that contradict what's stated in Table I?

Not so much.

> If it's with not being able to use a local IDE - that is something that's currently possible by bundling your Svelte components into vanilla JS and importing them into your Primo blocks & passing along data as fields - but I'd give it a couple weeks before using it in production until we smooth it out.

Yes. As a developer I don't want to deal with WYSIWYG drag-n-drop interfaces. I want the freedom to develop using my already established workflows. What you suggest sounds like development for the Primo interface, i.e. Primo dictates the high-level structure and I can develop components for it. What I'd like is at the other end of the spectrum: I develop the entire website in Svelte and make it compatible with Primo by adhering to some constraints.

> But what you've described is essentially how I do my client projects - I build it all with code (usually reusing blocks from other projects) and hand off a site that literally any of my clients can edit on day one with minimal training.

Sounds like we're on the same page then :)

What is is: a visual website builder where I can customise blocks using an online editor.

What I'd like: an online interface that clients can use to add / change text, or make other small adjustments, for Svelte websites that I build offline using my own tools. The idea would be that, as long as I build the website according to certain interface standards, it is readable by Primo and the client can modify using the visual interface. For any bigger modifications the client would come back to me and I have all the speed and freedom of my offline workflow.

Some might suggest that what I need is a headless CMS, but I have tried those and they are all over-engineered and giving me a big headache just for setup and maintenance.

I think he meant development overhead, not performance overhead.

Source?

As an outsider it seems that RE agents bring exactly nothing to the table except FUD around the transaction process which they then exploit for commissions so everybody, except the agents, are worse off.

More than happy to be convinced otherwise.

And if that wasn't enough the Open Source world is releasing new models almost weekly now, for free.

Anthropic is putting on a big show to convince gullible investors that there's money to be made with foundational models. There's not. I expect a big chunk of the raised money to go out the door in secondary sales and inflated compensations. Great if you're working at Anthropic. Not great for investors.

Err on the side of doing things where you'll face competitors. Inexperienced founders usually give competitors more credit than they deserve. Whether you succeed depends far more on you than on your competitors. So better a good idea with competitors than a bad one without.

[1] http://paulgraham.com/startupideas.html

> If building a valuable product, talking to users, and focusing on facts sounds good to you, you should consider applying to YC.

Also being perfectly honest, it does sound very appealing, but I'm afraid it unfortunately just isn't what makes startups successful in 2023. Speaking from personal experience and carrying the battle scars.

I sometimes get the feeling, also by watching their public content on YT, that a lot of YC advice, especially w.r.t. what to prioritize, is from an era of startups about 10-15 years ago.

Back then there were about 10x fewer SaaS startups (maybe that's a conservative estimate) and if you had a decent product that solved a problem for your customers and you were just plugging away at it, that was enough to get noticed and grow.

Nowadays every worthwhile niche seems to have 10-15 companies in it in no time, all of them doing more or less the same thing. Even if your company differentiates in a very special and useful way, getting that across to your potential customers is a totally different game now. You basically need 1-2 full time staff doing nothing but content creation and social media outreach marketing. Also locking in (micro-) influencers in your niche is key. Once they start promoting your competitors it's game over.

From the content that I see publicly on YT etc. it seems to me that YC thinks such things are basically a waste of time. And back in the late 2000s and early 2010s that was largely true. If you are targeting a narrow hyper-niche it is probably still true, however it's just not the reality for 99% of today's startups, even YC ones.

> This subtlety led to more than one case where I didn't understand why a component didn't update. ... In the end I found it was difficult to determine reliably when to reach for the $ label. I'd use it in one scenario and it seemed to work like I expect, then throw it at another scenario and it didn't work like I expect.

OK, that's like you're used to using an old donkey for transport and then someone hands you a brand new car and you start complaining about it not having seat heaters.

The $-based reactiveness of Svelte is great! Just don't expect that every possible side case is covered 100%. Same as with just about any other technology.

For most of the technical text drafting ChatGPT proved to be overchallenged, however it was a phenomenal help answering dozens of questions that I had about specific wordings, goals, processes, things to avoid and more. The type of information I would have searched on Google for hours with uncertain chances of success. Sure, there's a chance that ChatGPT just made it up but most of the answers made complete sense in my view.

I also used it to rephrase some boilerplate from other patents that needed to go in there but which I didn't want to copy verbatim. It did well in most cases but failed in about one or two.

But overall still blown away by it and pretty sure we'll see rapid progress from here.

Here's a scan of several km (!) of length all done with Dot3D on an iPad [1]. You can also explore the point cloud on the web.

Also, digging a little deeper, here's a study comparing the accuracy and loop-closure capabilities of several 3D scanning apps. It's spot on. [2]

[1] https://twitter.com/netch87/status/1632392369143435264

[2] https://rafaels-blog.weebly.com/blog/why-loop-closure-is-cru...