Tried this hardware's demo to write code, and I can tell you it changes the paradigm of how you use it. Virtually, insanely instantaneous, way fast enough to be truly conversational.

Frontier models are nice and all, but they're way too slow. Interacting with them is like running underwater. Developing software with the current models is painful and deliberate. It breaks your flow, slows down iteration. Maybe that's part of why devs are feeling somewhat demoralized when they switch to AI. That's super-important for productivity and loving your work as we've learned from having instant-turnaround IDEs vs. submitting a job to a mainframe to get a compile error back 10 minutes later.

That delay and expense forces you to treat existing AI-generated code as valuable. You rescue existing code the AI spit out previously, instead of iterating fresh code fast, telling the AI "OK, with everything we've learned together in the last 10 minutes, make a brand new version." (In case you hadn't noticed, AI's much better at fresh code than modifying old.)

2 words: "Bic pen" and all that suggests. 3-year h/w upgrade cycle isn't law. If a like-priced chip can do the job of 100 GPUs, you can afford to change it every month if you like and bank the power savings.

Besides, this things's so freaking fast that sticking a decent-sized RAG bundle into the context won't kill performance. There's your update mechanism.

Seems to be for real.Tried it on substantial prompts that couldn't possibly be pre-cached, 100ms response. Translated 1500 words to French in 50ms, and it's a good translation. Their demo's just an 8B model for now, but it's not hard to see how this can scale up. If nothing else, it can have a real-time conversation with no delay. That alone guarantees adoption in all sorts of applications where it interfaces with people. Methinks AMD got itself a bargain.

Speaking of getting paid:

So, it's a business. How much does GM get paid for using their customers' property vs. how much do customers get paid? Can anybody shed some light on this?

Talk about ROI: Looks like GM's only investment is a software update which probably includes other fixes they needed to do anyway!

I'm a careful coder to the point of paranoia, as you'd expect for someone who came up writing life-critical software.

Yet, I keep falling for subtly-imperfect "helpful" AI suggestions for low-level, supposedly-simple code. Because humans are inherently susceptible to being led astray, lazy at some level. Because it's hard to un-see things that seem to make sense.

Let's face it: even in rich countries, price always matters. Even for the military.

Why launch 20 missiles @ $1M when you can launch 200 @ 0.1M? Sure some missiles will fail, but nobody's defenses can counter the remaining 150+ at once. This is why they have automatic guns: you point (as opposed to aim) & spray bullets. One of them eventually connects. I think the US army doctrine is on the order of 100 shots for a single hit.

They can do that because bullets don't cost $1M each. I can afford the $20 Harbor Freight tool to do a one-time job the right way because it's not the $120 pro equivalent. It'll break after I use it 5 times, but that's OK, more where that came from. And often as not the $20 tool is pretty good and does last; the $120 tool is nowhere near 6x better.

Oh, and let's not forget that even those super-pricey weapons still have duds, and anybody with an iPhone or a Toyota will tell you that civilian technology can be pretty damned reliable.

So why isn't the US military doing the equivalent of buying in bulk at Harbor Freight when appropriate, following its own doctrine of "overwhelming" force? Why does every single little thing have to be so high-grade and expensive? An Air Force officer recently was quoted "quantity has its own form of quality." I realize that carrying all that extra quantity is an issue, but these days we have machines and logistics systems for all that, even robot pack mules for the last mile.

You can have both text and icons like on most keyboards. Icons-only on keyboards, devices, on-screen, etc is a usability cop-out, often done to ease internationalization or save space.

I've very rarely ever seen an icon that's clear and obvious to all, not matter how obvious its designer and focus group (i.e. designer's friends) think it is. And there's the issue that there are a lot of confusingly-similar icons floating around.

Yes, people learn, but what about all the new people coming into your device, app, etc? Their brains are already pretty full... like with other people's icons which may look the same but do something different. "Lemme see... what does this square with lines sticking out of it do?" (Turns out it that it turns on a light... circle with lines coming out of it activates the self-destruct feature!)

Butterflies are one thing, but this means anybody with a few $ can invisibly track anybody else.

https://mbi-berlin.de/research...

SpaceX has a structural advantage that NASA can't ever match: they're allowed to fail. How many rockets have they blown up, etc?

Imagine all the congressional investigations, firings, political backlash if NASA did the same! NASA can't run, has to walk very carefully to make sure it doesn't stumble.

Ain't it funny how things never really change? The advertising revenue game is an oldy but goody that just inevitably keeps coming back: the entire WWW, Google, Amazon, Netflix, Slashdot...

Advertising will inevitably a major income line item once OpenAI opens it up. The thing about multi-$billion line items is that they get a lot of scrutiny. If OpenAI tries some cute "be nice to the user" policy that fails to absolutely maximize profit, Brinkman will be told in no uncertain terms to fix that **fast** if he wants to keep his job.

He might want to fight that but, he's not a known marketing genius who can make long-term strategic marketing arguments stick. If there's anything he's learned in the last few years, it's to pick his battles carefully. And when you're making $T commitments, well...

Leaks are the least of the bad things that pointers can do, e.g. reusing deleted memory, exceeding bounds, flat-out corrupt memory access, etc. _CrtDumpMemoryLeaks is useful but very limited. Address sanitizers are an improvement, but right I'm working on what's clearly a memory access problem and yet the address sanitizer is happy.

Seems to me that DNG and RAW are separate issues: RAW is a bunch of good 'ol bare-metal, hardware-dependent formats. I'd hate to let Adobe dictate how my hardware works. That said, there's nothing stopping cameras from emitting DNGs as a secondary derived format same way they emit JPGs.

"commuting -- but so what?"
Dimon and senior leadership could show some actual leadership by example: start making a 2x/day 1-hour+ everyday commute himself -- and not by helicopter, limo, apartment in town, other unaffordable means.

I understand his time is super-valuable, but a above all a CEO's first job is leadership. That is why they pay you the big bucks, Jamie. Leadership by example well might be worthy of your time and effort. 'Coz if commuting's not worth your while...
Something for other hard-ass CEOs to consider too.

...are bits that you cut & paste. Even if you think you "carefully reviewed" it, it's many times more bug-prone than fresh, real code. I've been a freak about reviewing code ever since I worked on life-critical systems, but have to admit it's just too easy to get misled down the garden path by pasted-in code.