News   Jun 17, 2024
 231     0 
News   Jun 17, 2024
 266     0 
News   Jun 17, 2024
 409     0 

PM Justin Trudeau's Canada

A lot of junior lawyer work is already being shrunk by AI capabilities.

In my lifetime, we've moved from four people in an airplane cockpit to two with automation, while airplanes have actually gotten bigger, more capable and gained range. Now there's debates in aviation about going to single pilot. And of course, there's drones, where one pilot can control several aircraft. So even while the number of passengers flying has grown exponentially, the number of jobs for pilots has been linear. And that's not even real AI. It's just high end automation.

White collar is just more noticeable to more people now that there's a ton of white collar jobs. Even at the lower end, I think of what AI and LLMs could do for call centres for example. Like all the long waits on the phone to talk to CRA or Service Canada. Just imagine how much both service could be reduced and jobs cut. I expect a future government with a dislike of the Public Service is going to be pushing hard to deploy technologies like this.

But in general, the more expensive the work, the more motivated developers will be to find some combination of AI and robotics to replace that worker. And if we're being honest about ourselves, virtually everybody who works in an office can think of routine tasks or processes they do that could be automated.
 
It's more about tasks than jobs. But it's the tasks that are expensive and high frequency that will be the most actively automated. Expensive but rare may not be worth validating that an AI solution works well.
 
And translators too, already beat down by Google Translate will be made irrelevant by AI.

That's not even that impactful. How many actual translators are there?

By comparison, I think of engineering projects I've been involved in, with everything from conceptualization to final design to permitting and I can actually imagine AI cutting more than half the people involved and finishing these kinds of designs in days instead of weeks to months.
 
It's more about tasks than jobs. But it's the tasks that are expensive and high frequency that will be the most actively automated. Expensive but rare may not be worth validating that an AI solution works well.

Eventually jobs. But it starts with tasks and processes. And automating those automatically increases productivity which means there's simply no need to hire more, even we workload increases.

But definitely jobs that simply involved a ton of data entry and analysis will definitely be reduced first. I think of inventory and supply chain workers (buyers, analysts, managers) who work at major chains. So much of that can absolutely be automated over the next few years with a lot of the technology we have today. And if any chain is slow to do it, they'll get eaten by the competition who will be ahead.
 
And translators too, already beat down by Google Translate will be made irrelevant by AI.
Google translate is far from being ready to replace translators if you want current natural language. Especially in places like Quebec where you'll have disagreements on how something should be translated.
 
AI is definitely at a peak of the hype cycle. That doesn't discount the fact that advancements are continuing to be fairly rapid. I don't think the claim that we have reached a plateau or diminishing returns in the latest breakthrough with transformers/LLMs is supported by the evidence. The fact that the cost of inference has declined by 12x in a matter of months and has much further to fall yet through purpose-built hardware is in some ways more meaningful that advances in model size. Additional inference resources can extract more value out of existing models. It continues to be a target rich environment. Of course, most of the startups are BS and will be crushed by base models becoming more capable. It's a bit like how pets.com etc. all got eaten by Amazon.

Also not sure if people are aware of the huge advancements in things like protein folding, drug discovery (AI helped identify a novel antibiotic) , materials science, etc. enabled by AI. Big things are coming on that front.

Content generation is an example of BS jobs that are being automated away. Think copywriting... lots of Amazon listings found with ChatGPT errors...

Those advancements are still fundementally limited by the type if 'AI' this is, since it's not actually intelligent in any way. There is no path here to AGI as some are hoping for. And since you mentioned cost, yeah, the big tech companies are eating it for now, but that doesn't last forever either, no matter how low they can go, training uses a hell of a lot of compute and electricity.

In terms of applications in science, etc, yes, this is what AI can be helpful in, but those systems act in addition to humans, not replacing them, so they won't be adopted widely outside educational contexts.

In terms of some low-level content generation jobs going away, in the short term this is likely as part of this cycle. But the content generated by AI won't be as novel as produced by humans, and of course, prone to some fun errors, so the pendulum will swing back. See Air Canada chatbot as just one example of this, but, there are, and will be, many more. And yes, humans are prone to errors as well, that's why 'AI' works as a great assistant to review work, detect errors, etc, but not generating new things. This means less productivity, since work will take longer, and businesses won't like that.
 


The truck driver who caused the deadly Humboldt Broncos bus crash is set to be deported to India following a decision from a federal immigration official on Friday.

Jaskirat Singh Sidhu’s lawyer did not contest the decision, noting the official had “limited discretion”, because all that is required to deport Sidhu is proof he is not a Canadian citizen and has committed a serious crime.

A permanent resident is someone who has been given permanent resident status in Canada but is not yet a Canadian citizen.

Sidhu became a permanent resident a month before the crash that left 16 people dead and 13 others injured.
 
Google translate is far from being ready to replace translators if you want current natural language. Especially in places like Quebec where you'll have disagreements on how something should be translated.
My team uses it all the time for our website and labels. We drop in the Google translate and then send the draft to our one of our francophone staffers to double check and recommend revisions. Normally we get maybe four words or six technical terms wrong, which isn’t bad when it’s hundreds of words.
 


The truck driver who caused the deadly Humboldt Broncos bus crash is set to be deported to India following a decision from a federal immigration official on Friday.

Jaskirat Singh Sidhu’s lawyer did not contest the decision, noting the official had “limited discretion”, because all that is required to deport Sidhu is proof he is not a Canadian citizen and has committed a serious crime.

A permanent resident is someone who has been given permanent resident status in Canada but is not yet a Canadian citizen.

Sidhu became a permanent resident a month before the crash that left 16 people dead and 13 others injured.
Good. But now we must be fair, so any non-citizen who kills anyone with their car should also be deported. Otherwise we’re just picking on the Indian.
 
Google translate is far from being ready to replace translators if you want current natural language. Especially in places like Quebec where you'll have disagreements on how something should be translated.
Google Translate is still 'old school' machine learning, not based on more recent LLM translation. LLM translation still gets it wrong, but is much better/more natural. Open AI recently was demoing real time translation with ChatGPT 4o.
 
Those advancements are still fundementally limited by the type if 'AI' this is, since it's not actually intelligent in any way. There is no path here to AGI as some are hoping for. And since you mentioned cost, yeah, the big tech companies are eating it for now, but that doesn't last forever either, no matter how low they can go, training uses a hell of a lot of compute and electricity.

In terms of applications in science, etc, yes, this is what AI can be helpful in, but those systems act in addition to humans, not replacing them, so they won't be adopted widely outside educational contexts.

In terms of some low-level content generation jobs going away, in the short term this is likely as part of this cycle. But the content generated by AI won't be as novel as produced by humans, and of course, prone to some fun errors, so the pendulum will swing back. See Air Canada chatbot as just one example of this, but, there are, and will be, many more. And yes, humans are prone to errors as well, that's why 'AI' works as a great assistant to review work, detect errors, etc, but not generating new things. This means less productivity, since work will take longer, and businesses won't like that.
Not much to say beyond I think this is the wrong take. AI is as bad as it will ever be going forward, and we haven't even seen a lot of the products based on the current AI tools come to market yet. Even a fair amount of code can be generated by LLMs, and just reviewed by a developer. The productivity gains are non-trivial, and not limited to listicle generation.
 

and

https://www.thestar.com/politics/fe...cle_4eacb967-62e2-5ca5-a234-943d5e05132c.html

Glad Trudeau has said it like it is; Netanyahu is doing immense harm to Israel and very happy that he did not try to obfuscate about the ICJ actions today. Just as with elections (pay attention US!!), you cannot only follow Court Orders if you like them!
 

Back
Top