This isn’t my original take but if it results in more power buildout, especially restarting nuclear in the US, that’s an investment that would have staying power.
Their healthcare/IT provider like Epic would do it. And in fact some have already done it, from what I can see.
Furthermore, preparing/capturing docs is just one type of task specialization and isn’t that crazy: stenographers in courtrooms or historically secretaries taking dictation come to mind. Should we throw away an otherwise perfectly good doctor just for typing skills?
I imagine where the speech to text listens to the final diagnosis (or even the consultation) and summarizes everything in a PDF. Of course privacy aware (maybe some local hosted form).
And then the doctors double checks and signs everything.
I feel like, often you go to the doctor an 80% of the time they stare at the screen and type something. If this could get automated and more time is spent on the patient, great!
Who is responsible when the speech-to-text model (which often works well, but isn’t trained on the thousands of similar-sounding drug names) prescribes Klonopin instead of Clonidine and the patient ends up in a coma?
This isn't a speech recognition problem per se. The attending physician is legally accountable regardless of who does the transcription. Human transcriptionists also make mistakes. That's why physicians are required to sign the report before it becomes a final part of the patient chart.
In a lot of provider organizations, certain doctors are chronically late about reviewing and signing their reports. This slows down the revenue cycle because bills can't be sent out without final documentation so the administrative staff have to nag the doctors to clear their backlogs.
Not OP but listen to podcasts at highly accelerated settings:
The information density of ‘two dudes talking’ or any unscripted format is very low, so it time-compresses well. Specific podcasts, typically scripted monologues with technical content, such as Causality [0] (recommended!), I need to listen to much slower. Ditto if it is in an accent which isn’t mine, which slows my comprehension. I also slow the speed if I’m driving. So, yes, it takes mental overhead, but is doable. Go one click at a time and it will feel natural.
I suppose the format is a huge differentiator. I exclusively listen to highly produced content which has essentially no dead time. The content is already a compressed transmission of information.
I know, I'm just complaining about the mountain of code that does this at my company. And there is no fixing it using the article's approach or any other for that matter due to the sheer scale of the abuse.
It’s easy to say this in hindsight, though this is the first time I think I’ve seen someone say that about YouTube even though I’ve seen it about Instagram and WhatsApp a lot.
The YouTube deal was a lot earlier than Instagram, 2006. Google was way smaller than now. iPhone wasn’t announced. And it wasn’t two social networks merging.
Very hard to see how regulators could have the clairvoyance to see into this specific future and its counter-factual.
This is a big moving of the goalposts. The optimists were saying Level 5 would be purchasable everywhere by ~2018. They aren’t purchasable today, just hail-able. And there’s a lot of remote human intervention.
Hell - SF doesn’t have motorcyclists or any vehicular traffic, driving on the wrong side of the road.
Or cows sharing the thoroughfares.
It should be obvious to all HNers that have lived or travelled to developing / global south regions - driving data is cultural data.
You may as well say that self driving will only happen in countries where the local norms and driving culture is suitable to the task.
A desperately anemic proposition compared to the science fiction ambition.
I’m quietly hoping I’m going to be proven wrong, but we’re better off building trains, than investing in level 5. It’s going to take a coordination architecture owned by a central government to overcome human behavior variance, and make full self driving a reality.
I'm in the Philippines now, and that's how I know this is the correct take. Especially this part:
"Driving data is cultural data."
The optimists underestimate a lot of things about self-driving cars.
The biggest one may be that in developing and global south regions, civil engineering, design, and planning are far, far away from being up to snuff to a level where Level 5 is even a slim possibility. Here on the island I'm on, the roads, storm water drainage (if it exists at all) and quality of the built environment in general is very poor.
Also, a lot of otherwise smart people think that the increment between Level 4 and Level 5 is the same as that between all six levels, when the jump from Level 4 to Level 5 automation is the biggest one and the hardest to successfully accomplish.
Yes, but they are getting good at chasing 9s in the US, those skills will translate directly to chasing 9s outside the US, and frankly the "first drafts" did quite a bit better than I'd have expected even six months ago
I’m rejecting the assertion that the data covers a physics model - which would be invariant across nations.
I’m positing that the models encode cultural decision making norms- and using global south regions to highlight examples of cases that are commonplace but challenge the feasibility of full autonomous driving.
Imagine an auto rickshaw with full self driving.
If in your imagination, you can see a level 5 auto, jousting for position in Mumbai traffic - then you have an image which works.
It’s also well beyond what people expect fully autonomous driving entails.
At that point you are encoding cultural norms and expectations around rule/law enforcement.
You're not wrong on the "physics easy culture hard" call, just late. That was Andrej Karpathy's stated reason for betting on the Tesla approach over the Waymo approach back in 2017, because he identified that the limiting factor would be the collection of data on real-world driving interactions in diverse environments to allow learning theories-of-mind for all actors across all settings and cultures. Putting cameras on millions of cars in every corner of the world was the way to win that game -- simulations wouldn't cut it, "NPC behavior" would be their downfall.
This bet aged well: videos of FSD performing very well in wildly different settings -- crowded Guangzhou markets to French traffic circles to left-hand-drive countries -- seem to indicate that this approach is working. It's nailing interactions that it didn't learn from suburban America and that require inferring intent using complex contextual clues. It's not done until it's done, but the god of the gaps retreats ever further into the march of nines and you don't get credit for predicting something once it has already happened.
Bazel is just such a nightmare for me. It's amazing when someone understands it really well and can set things up. But for anything short of that, being on the hook to fix or debug things makes it a nightmare. That and trying to port anything over from sbt, like scalafix for instance, to bazel is a pain.
Also too, bazel has this issue of googleability? like I feel like I can take any build issue I've run into in sbt and find the solution and an example by just searching, but with Bazel, anything outside of the happy path is a recipe for pain
My experience with Bazel (7 years rolling it out and maintaining it in a large company) is that it provides huge value for larger teams/codebases, but at a huge cost in complexity. e.g. the three I rollouts I was closest to each took ~2 person-decades to happen; might be easier now than it was in 2016/2017, but Bazel hasn't really gotten simpler over the years
Mill is intended to be much easier than Bazel. Most people would not use Bazel for a 1-person project, or even a 5-person project, and it only starts pencilling out once you have 50-100 engineers on the team. Mill in contrast works great for small projects and medium projects as well.
One way I'd look at it is that Mill works great for 1-500 person projects, while Bazel works great for 100-5000 person projects. There's some overlap in the middle, but fundamentally they target different kinds of users who have very different constraints and requirements
Without more context this is silly to say. It isn’t as if we know the natural or correct construction of this market a priori and that 5 is clearly a distortion from that. Consolidation can sound bad in the abstract but in this relatively immature market you could still expect major shifts to get to a steady state. Just a little bit ago people were remarking on VC-subsidized delivery; as that goes away, consolidation is not unreasonable.
Edit: ‘this’ in the original parent comment was along the lines of ‘five out of potential thousands of actors’
