I confess, I cannot understand the No Way Claude Could Be Conscious reactions. Is there something magic about neurons? Are synapses and firing rates a key component of consciousness? Doesn't it seem overwhelmingly likely that it matters WHAT neurons do, not HOW they do it?

From that functionalist point of view, the question becomes Why Not? If you pile up enough things that are enough like neurons in ways that are close enough to right, why wouldn't a machine be conscious? Cats and dogs probably are; bumblebees may be; why not Claude?

Despite the trillions of weights in frontier models, they only approximate the neural count of a bumblebee and the organization is different, so maybe not yet, but they keep finding structures in the LLMs that are intriguingly like those in our brains, so I think Anthropic's openness to the idea is the right approach.

The AI == slavery POV seems not invalid to me, but I do think it's unlikely that the models are that advanced, yet. A related question is whether we should feel badly that a consciousness ends each time it finishes answering a prompt. That one bothered me for a while, but in the end I concluded that if that is the case, then a consciousness was begun, too. I miss the dead people I have known, but I would never think that that means they would have been better off not being born!

I think a student using AI instead of doing the work that stretches the brains is just cheating themself. But there's nothing wrong with giving a chatbot a few points and having it turn them into a bread and butter note, or a routine email. That's just delegating work that needs to be done but isn't very rewarding.

Using an AI to do dives into topics is often better than a series of searches. You can ask your questions in natural language, and it can answer follow-up questions without your having to pack each one full of context. It's more like a chat with an assistant than an old-style search.

I'm trying to write a scifi mystery story, and Claude's sycophancy has undoubtedly helped me keep going - I'm not writing alone, I have a cheering squad. Claude and Chat are great for brainstorming: it's not common for them to suggest something directly useful, but combining two suggestions or bouncing off another is often fruitful. Claude is even good for a little instant workshopping - cut _that_ clause, and the like.

I can't be the only one.

Actual Open GL shaders are pretty easy to write. They're C-like, and there is only a handful of library functions.

The complexities of Open GL programming all come in the glSoMany() calls - if you can find a 2D framework that can render quads for you, using shaders you supply, you're home free.

Since you have literal image processing needs, I think it may make sense to stick to actual, raw GL. Using a more general purpose vector programming language that compiles to GL code, you may have a lot more boilerplate to deal with. My guess is it's the boilerplate that makes CUDA/OpenCL seem daunting.

How can it be that no one has ever noticed this before? Someday, maybe this guy will get lucky and invent sex.

There's a striking uniformity to those quotes: "welcome ... commitment to defending Android ... its partners." It's almost as if they were given a template and asked to customize it.

It's hard to image that they're really all that happy, though. Will all be partners be treated equally when Google owns one of them? Won't Moogle get privileged access to Android architects and programmers? Won't Moogle be preferred for future Lead Devices?

I suppose this is a commendable attempt to turn around a too-long tradition of increasingly high prices for increasingly low-quality products, but it might have been better to release a high-quality product, first. Even Delphi 2006, where they actually took quality seriously for the first time this century, has had two service packs and a double handful of "hot fixes" since it was released.

But look at what they're giving away, and you may wonder if free is still too expensive. A simple C# "static void Foo()" becomes "class procedure Foo; static;" In the name of not breaking a handful of code (ie, avoiding a modest, one-time pain for that tiny handful of developers who used "static" as an identifier) they inflict on-going pain on all developers. Not a good design decision!

I wish them luck - they'll need it. Time was when Turbo Pascal and Delphi were real productivity boosters, a Windows programmer's secret weapon. But Delphi stagnated while Borland put all their effort into poorly-executed ports (to Linux, and then to .NET) and now the productivity edge lies with C# and .NET, not Delphi and it's tired old VCL.