nkozyra
My hunch is that software is pretty much going to end the way we think of it today.
Hardware, too, but that will take longer. If you think big enough, the idea of waiting for a new TV or phone or gaming system when you can just have your agent take your description, spec and order the hardware and have it fabbed for you in days ... is kind of quaint.
The broad societal disruption will be pretty devastating. But looking at software as the microcosm really lets you dream (/nightmare) big.
To me the question is not will we be able to get anything we want how we want it, but more who will have money to buy this?
Yeah we'll still need to know a sequence to flip bits in the right order, but it isn't code you'll recognize.
The AI assistance you need is "rewrite this in something else"
Robotaxi is trying to enter an already crowded market and isn't exactly flourishing there. I'm still seeing every one in Austin with a human behind a steering wheel.
As for the Semi, yeah, if it actually comes to fruition. But like all things Elon ... let's wait and see.
IIRC the spec itself says it was built for C applications.
mod-perl was a godsend, of course.
But it could be the case!
What I suspect happened with Deloitte specifically is they found a way to appear to expand globally without taking on a lot of the risk. More like brand franchising.
Which is all to say that I wouldn't doubt Deloitte Cyprus (or other branches) have shady stuff going on, but that this structure wasn't designed specifically for that purpose.
"The member firms are primarily organised on an individual country or regional basis and each operates within the legal and regulatory framework of its particular jurisdiction(s). They are separate and independent firms that have come together to practise under a common brand and shared methodologies, client service standards and other professional protocols and guidelines."
https://www.deloitte.com/ce/en/about/governance/network-bran...
Sure. This is only going to be anecdotal and sparse examples, but this became a thing post-Covid, primarily as a "natural alternative" to vaccination.
Not just random people on random forums, Joe Rogan several times espoused the value of D along with other usual suspects:
"Rogan said he started feeling ill on Aug. 28 and tested positive the next day. He added he began taking his regimen the same day, which apparently included ivermectin, IV drips of vitamins D and C, nicotinamide adenine dinucleotide (NAD) that helps repair cells, and monoclonal antibodies. He said he felt fine the next day."
"Man Vomits For Months After Taking Vitamin D at Almost 400x Daily Recommended Dose"
https://www.news-medical.net/news/20220706/e28098Hypervitami...
https://www.sciencealert.com/man-overdoses-vomits-for-months...
Also unsurprisingly happened more often around this time with infants:
https://www.frontiersin.org/journals/pediatrics/articles/10....
Unless, as mentioned, it had some traction as a cure all around the time of COVID.
This happens periodically and is again happening with vitamin C (which, whatever, getting to any dangerous amount might be physically impossible).
It became something of a panacea nutrient, recently around COVID.
This doesn't mean you can just pound it, but for the most part it just becomes waste product.
They're generally intended to run flashed firmware and be single-purpose. They usually come bundled with peripheral components and typically have breakouts to add more (pots, sensors, etc).
They're great for doing one thing (though some support firmware switching), and you can build the 'software' for them pretty easily using Python or C-like variants.
So for hobbyists they're perfect. You want a quick LED christmas display you can toggle between Rudolph and Santa from a quick network call? You want a humidity sensor in your bedroom? You want to create a networked security camera? These are the types of use cases for MCUs.
If you know what you're doing, a loaded gun without a safety is fine, too. But you have the option, why not pick the unloaded one with a safety mechanism?
Sure, but have you seen the Typesafe.ai site itself? I think this is meant as a homage.
Because terms of service require both parties to accept!
If I hold a contract outside your window and scream "you accept this contract" it doesn't bind you simply by looking out the window.
From the intro blog
"Hallucination and type-safety are intrinsically related"
I'm not entirely sure why we're conflating type safety with, I guess, value or output safety.
"Would you say a linear classifier hallucinates?"
No, but it can be (and often is) mathematically correct and functionally incorrect. It doesn't help to say "a linear classifier can't hallucinate" when you get even 99% accuracy. That's 100% a semantic play, and it doesn't help when the picture of a dog is labeled cat and the response is "yeah but that's not a hallucination, only stupid LLMs do that"
Even non-frontier models are absurdly good at this in a broad sense.
Which would make it hard to judge "a model that will never produce unreliable outputs in the first place" against something that is already really, really good and exceptional in domain-specific areas with the tiniest amount of elbow grease.
Speed and cost look good though (for now)!
But terms of service are contracts which you sign by agreeing.
Courts have looked somewhat favorably on "I didn't read that" defenses, but accepting the agreement is a legal contract between you and the service provider.
If it's auth-gated, it needs to be, well, authenticated.
The person who provides that, and the subsequent automated scraping, is the party, and would be the recipient of a civil suit.
It's just a different kind of gate, though.