ChefboyOG

That "(according to the media)" bit of hand-waving is doing a lot of work.

I'm not sure I understand what you're saying. Google's Smart Reply, along with most (I think all?) of Google autocomplete features, uses a neural network:

https://research.google/pubs/pub45189/

Are you saying that in general statistical modeling is not the same thing as truly "understanding" a concept? Your original comment seemed to suggest that there wasn't utility in this kind of model--which I disagree with--but if you are more generally saying that this is not the same as human intelligence, I think that authors would probably agree with you.

By and large, no.

That's not to say that those sites are not generated programmatically--without a doubt, most of them are--but not by a cutting edge transformer model. The fact is, generating words has never been the bottleneck for blackhat SEO types. Generally, those sites are generating their content through some kind of scraping, or in rarer cases, paying pennies for nonsense articles. The page itself is structured for search (targeted H1s, metadata, etc.) and some kind of private blog network is used to create a pyramid of backlinks.

Have you ever used any sort of autocomplete?

I think this gets at what the actual value prop (not the marketed one) of chatbots really is: scale.

There will not, in the foreseeable future, be a model/ensemble/pipeline capable of conversing at the level of a human representative. Thinking of them as "humans, but cheaper" is destined to fail.

There are, however, use-cases where communicating with a huge group of individuals on a one-on-one level is useful. For example, years ago a company called Mainstay (then AdmitHub) built a chatbot for Georgia State designed to reduce "summer melt," a phenomenon in which intending freshmen drop off between graduating high school and enrolling in classes. For this use case, getting human-level conversation wasn't as critical as effective performance on tasks like question answering. Their bot, Pounce, was credited with a ~20% decrease in summer melt. They have a proper study here: https://mainstay.com/case-study/how-georgia-state-university...

This is an example, in my opinion, of the sweet-spot where intelligent, individualized, but not-near-human-level communication is needed, at a scale that goes beyond what you could reasonably do manually without an enormous dedicated staff.

Now, I don't know how numerous those opportunities are, or if they come anywhere close to the level of hype and funding chatbot companies have received. I just think that in a vacuum, there is a set of problems where they have value.

Quanta is one of the most consistently high quality publications operating today, for my money. Their Joy of X podcast with Steven Strogatz is highly recommended for anyone interested in math.

As a funny (to me at least) piece of anecdata:

I've never worked at a startup that wasn't either in the middle of or planning a UI redesign.

One of the other benefits of this kind of approach is that it allows the two sides to inform each other without hindering each other. Each function gets the proper space to "breathe" in a sense.

For example (and this draws entirely from my own anecdotal experience), I find that in the course of marketing, I often get the best feedback/inspiration from users/community members, and this is very helpful in deciding what to implement next. However, I find that this pseudo-research process is only effective if I take enough time to process the information in whole. In other words, if a user has an extremely compelling idea, it can be tempting to implement it right away. If I resist and take my time, considering the idea in the context of the dozen other points of feedback I've collected, I can often distill the jumble of ideas into a stronger, singular feature. If I jump in right away, I'll inevitably end up in a state of feature creep.

These kinds of notes never worked for me, but I've found the Feynman technique (inspired by Feynman) to be a very effective strategy for me that similarly derives its benefits from forcing me to synthesize and distill information.

For those interested, it's basically just teaching concept (writing an article, talking to someone, etc.), usually after having studied it/taken notes. The idea is that teaching a concept, even to an imaginary audience, forces you to learn it on a fundamental level:

https://blog.doist.com/feynman-technique/

...they're also great?

The parent comment was just that it would be cool to bring Hubble back eventually, and that Starship seems promising towards that end. It's not a diminishment of all other space-focused organizations.

I think this highlights the best path forward. Large-but-not-Manhattan cities should eliminate cars in the densest areas of their downtown, where it makes sense. Jersey City did this in their downtown area, and it seems to be going well.

As people who want to live without cars move to the downtown zone, the area adapts to meet their needs (in terms of the types of stores etc). If the policy proves popular, the city has the option of spreading the "car-less" zone as the dense urban core grows. Similar to suburban sprawl, but inverted, I suppose.

The "undo" functionality, in my experience, is not amazing because it provides an enhanced eraser. It's amazing because it allows you step backwards through your free-hand thoughts.

As an example, I routinely use my tablet like a whiteboard for working out bits of math. The ability to bring back a value I'd previously erased to see exactly where I miscalculated is extremely useful. Similarly, I find this sort of thing to be really helpful when whiteboarding my way through a particularly thorny programming problem at work.

I had a similar setup at a previous job. Basically, the journey went like:

1. I already used VSCode as my daily driver. 2. Backend team all used IDEA for their Java codebase/had some workflows designed with it in mind. 3. I familiarized myself with IDEA for working on the backend, but used VSCode for the rest of the codebase.

I agree with this, but I think there's an additional point here. I'm from (I'm assuming based on context) a generation just past yours. I know many programmers older than me who had childhood experiences in which figuring out how to build their own games or random software was extremely formative. I think part of what made that so formative was that the gap between "professional software" and what they could build was extremely minimal, if existent.

For me, I can distinctly remember deciding to tinker with building games when I was very young. I was completely overwhelmed, however, by the jump from "basic Batch programs" to "OpenGL and C++." Obviously, that's an enormous jump, and I could have simply dabbled with Flash or whatever, but the point is that the software I enjoyed playing with was years and years of dedicated learning away from anything I could build or tinker with.

The web, however, was not. It was still nascent, there weren't mature, robust technologies for it, and I could build all the things I saw online completely on my own with maybe a dedicated weekend. This was huge for me, and kicked off my lifelong involvement in computing.

I guess the point that I'm getting at is that technology is so good now that the gap between what an average child can consume and what they can produce is enormous. When I tried teaching my nephew basic scripting, their first questions were "How do I make it do X" where X is some involved bit of graphical interactivity. That level of development simply is what software is to them now.

My hope is that this is cyclical. That there is always another frontier cropping up where the gap between "state of the art" and "hobby project" is incredibly small. I think that is naturally where you'll see kids getting into development.

The "unknowingly" caveat makes that one a bit difficult to answer.

I don't know of people dying from a belief that they were cursed, but we do certainly know that placebos can have very real, measurable neurobiological effects. For example, there many studies looking at the body's response to placebo painkillers. The brain often releases neurotransmitters that bind to opioid receptors as if an actual painkiller had been taken: https://www.nature.com/articles/535S14a

The original comment was about placebos bringing "powerful relief," which there is a great deal of scientific research to support. As far as I'm aware, there is no serious medical researcher advancing the hypothesis that stage 4 cancer or type 1 diabetes should or can be treated via placebos.

It feels like people have been predicting the end of the "hype cycle" for the last decade.

I agree that this particular application seems... questionable, to say the least, but finance is probably the last place where data scientists need to work very hard to justify their efforts. Statistical modeling has been a core part investing for a very long time, and ML is just a subset of statistical modeling.

It's literally the last section of the page, just before the citations, and it's only a few paragraphs + two tables to show the model's performance on industry standard benchmarks.

Also, complete anecdata on my part, but I've found developers who've gotten used to working on Macs find Pop OS to be a smooth transition. At least, in my circle of friends/coworkers.

There are also some pretty cool open source projects dedicated to tracking this kind of thing:

https://www.comet.ml/site/introducing-codecarbon-an-open-sou...

Mankind made it a long way without antiseptics, antibiotics, or electricity. I'm not sure what your point is.

Alternatively, we delete comments by whichever user's screen name reminds us most of the name "Chuck"

I dunno. If AWS went away suddenly, or if Google Search/the G-Suite suddenly stopped existing, the internet as we know it would need some time to recover.

Even if this is the case, why does it matter?

If the explicit goal is to create a human intellect, then sure, there's a really interesting conversation there—one that is happening constantly in the DL/AI research community, in which virtually no one believes that we're close to AGI or that current deep learning is going to achieve it.

But that's explicitly not the goal that 99.9% of neural networks are designed with. Their traditional use case is where they excel: programmatically approximating functions that are exceedingly hard to approximate manually.

This includes but is not limited to image recognition, speech synthesis, recommendation (including search), fraud detection, ETA prediction, even medicinal chemistry.

There is something deeply ironic about this author's righteous indignation over "intellectual dishonesty."

Looking at the first chart offered, comparing Kentucky and Tennessee, the chart interestingly begins in December of 2020. Both states have a drop in cases beginning around January 2021, and as the cases seem to bottom out, both roll back mask mandates—albeit with Tennessee doing so first. Here's what's wildly dishonest about this:

1. Extend that chart back a year, and you'll see that Tennessee—whose governor was opposed to any statewide mandate—spent most of July/August with a 7 day average hovering around 2,000 new cases. Kentucky, where the "cowardly" governor did mandate masks, was hovering around 600. Tennessee's population is 50% larger, and yet, they were experiencing more than 3x the case count.

2. In January, when all of these graphs seem to drop, the vaccine was rolling out. These charts, if anything, do a better job of demonstrating the efficacy of vaccinations than they do of showing the ineffectiveness of masks.

There certainly are tasks where a 2% failure rate is fine, but even more importantly, where we see ML/DL having the biggest impact today is in regards to complex system-type problems, where there is often a less-than-discrete notion of failure.

Looking at apps we use every day, almost all of them owe some core feature to ML/DL. ETA prediction, translation, search, spam filtering, speech synthesis, autocomplete, recommendation engines, fraud detection—and that's not even touching the world of computer vision behind nearly every popular photo app.

A key understanding gap in the general public's knowledge of ML is that people think AI === Skynet, and they've therefore been lied to about the field's progress and impact, when in reality, they probably interface with a dozen pieces of technology that are built on top of recent breakthroughs in ML/DL.

Similarly, I remember taking a handful of courses there many years ago. They were very high quality relative to what was available then.

I wonder if their particular niche of online education is tougher nowadays. There seems to be a wealth of online platforms tailored towards the "professional skills" edge of the market—if I ever need a course on migrating to Azure using only a TI-86 calculator while respecting HIPAA, I'm sure Pluralsight has a course—but when I think of the more general "learn to code" style courses, I don't think of Treehouse anymore.

In particular, having watched multiple family members/friends transition into software development (coming from no real background in code) over the last couple years, I've noticed they swing between two extremes:

1. Completely free resources, like FreeCodeCamp, CodeAcademy's free plan, or App Academy Open. 2. Going all-in on an immersive bootcamp, typically with some kind of job placement assistance program at the end.

I wonder if more middle-of-the-road premium options like Treehouse are losing marketshare to this. Though obviously, this is big time anecdata.

Does CNN know about this?

Texas is a big place, and as far as I'm aware, doesn't make it a policy to train individuals to be vigilantes.

If your anecdote proves anything, it is the relative safety of that liberal area of PA. The fact that a manager of a Walgreens was comfortable confronting a criminal without real consequences speaks to the level of danger they were in.

If you've spent time in any rougher areas of Texas (I have), you'd agree that physically confronting a criminal over petty theft, as a store manager, carries a huge risk of violent escalation—something you'd probably want to avoid in a city with a higher murder rate.

...what? The average Texan, as it turns out, isn't literally Wyatt Earp. It would be shocking if a random Best Buy manager stopped a shoplifter with a firearm—not to mention probably in violation of Best Buy's policies for handling shoplifting.

The average cashier's ability to administer lethal force is probably not a major influence on the prevalence of shoplifting in Texas.