I can attest to this. I work in Midtown Manhattan. You'd think walking around meant getting distracted by the all the activity around you that you'd forget about the problem you're trying to solve.
But I've found that distraction is the catalyst. Creativity for me comes when I focus on something else for a while, not grinding on the same problem with unwavering focus.
People often say Mandarin and Cantonese are like Spanish and Portuguese, but that undersells how different they are.
Your example of Spanish and French is more accurate -- same language family, but different grammar and vocabulary.
I offer German and Dutch as another example pair -- same language family as well, but different enough that no one will say "oh they're just different dialects". Dutch is an example of what happens when a Germanic language (Low Franconian) gets it's own state.
Category theory is rarely useful by itself, but it can be a mental scaffold when designing things like query languages. Microsoft's LINQ dsl within C# used category theory ideas to ensure consistency. That said, the applicability surface area in practice is typically quite limited in my experience. It's like formal methods -- elegant in practice, but a good problem fit is often rare. It's like writing a LEAN proof for your web app -- rarely needed, but if your web app needs a high degree of correctness, then indispensable.
This is John D Cook's take:
Category theory can be very useful, but you don’t use it the same way you use other kinds of math. You can apply optimization theory, for example, by noticing that a problem has a certain form, and therefore a certain algorithm will converge to a solution. Applications of category theory are usually more subtle. You’re not likely to quote some theorem from category theory that finishes off a problem the way the selecting an optimization algorithm does.
I had been skeptical of applications of category theory, and to some extent I still am. Many reported applications of category theory aren’t that applied, and they’re not so much applications as post hoc glosses. At the same time, I’ve seen real applications of categories, such as the design of LINQ mentioned above. I’ve been a part of projects where we used category theory to guide mathematical modeling and software development. Category theory can spot inconsistencies and errors similar to the way dimensional analysis does in engineering, or type checking in software development. It can help you ask the right questions. It can guide you to including the right things, and leaving the right things out. [1]
If you lived in a high place (Denver), you will find it different from a flat lowland (Chicago).
Also in Rio, how high you live can be a marker depending on which part of town you are. Favelas are on hills, whereas wealthy people in Zona Sul live down the hill closer to the beaches.
It's not that altitude isn't important, it's that it's basically determined by latitude and longitude (and time). Cultures don't exist directly above or below each other, especially not at the national scale being discussed here, and even at the micro scale, differences within a single high rise are presumably minor compared to the same distance laterally.
DuckDB is both a standalone and a component. This effort is actually very coherent and brings it back into a familiar usage model — that of a traditional client server RDBMS.
RDBMS have always been multi-user concurrent systems. DuckDB is a very fast local engine that has a multitude of use cases because it is a embeddable in other systems.
It’s like saying what does SQLite wanna be? It’s in your phones, your browser, your desktop apps, iot devices and people have extended it in different directions. The only difference here is that this is first party not third party. But to me it’s a very legible move.
But why though? DuckDB can still be used as a local query engine — I still use it as that. I haven’t touched any of the DuckLake stuff and the duckdb cli and Python library are still my bread and butter. They can add new use cases, but it doesn’t affect the core engine.
Is the concern that the duckdb messaging is now diluted by it having all these extra features? That you can’t sell it to friends as “this thing” like you can a one use tool like curl? I get that, but I also feel that duckdb is so much bigger than a “do one thing and do it well” tool.
It’s an engine that drives the modern data tool stack. Duckdb’s team has been prescient in that it has made many tasteful bets on what users want —- the ability to interop with pandas and polars, addition of geospatial, the plug-in infra. They’re all optional but when you neeed these things, they’re so useful. They’ve also clued me into what the broader data world is thinking about (I didn’t know about sketches and hilbert, but those are so useful in probailistic large scale queries and in geospatial queries). And they exist in larger database systems like Redshift too.
So far duckdb’s bets have been tasteful, and mostly ignorable if you don’t happen to use them.
It sounds like this replaces the PCA reconstruction function with a quadratic.
The normal PCA encoding:
1) Given a mean-center-scaled X matrix, get the latent variable matrix T with X = T * P’ + e, where P = loadings and e = residuals. The P is your model, so for a new vector xnew, you can calculate tnew = xnew * P (because P’ * P = I).
This is the encoder —- nothing changes here. The original matrix is dimensionally reduced with residuals e discarded. This is why PCA is lossy.
The decoder is where things diverge
The usual PCA decoder reconstructs a given latent variable t_any by using the trained P loadings, like thus x_reconstructed = t_any * P’. This reconstructed data lies on a linear hyperplane, so if the original data did not lie on the hyperplane, reconstruction errors are potetially high.
In your proposal, instead of a linear decoder, you train a quadratic decoder (essentially a classic ridge regression using a quadratic) on the original X. So for your reconstruction, you have x_reconstructed = poly(t_new).
This achieves lower reconstruction error in-sample (naturally, because quadratic is higher order than linear), but your poly function is trained on a particular corpus. Which means that when you’re in-distribution within that corpus, you’re good but when you’re not, you can be very wrong in biased ways that PCA’s linear reconstruction is not.
SO this is not a better technique than PCA in a general sense. It’s a better reconstruction machine when your data is mostly in-sample. It’s a kind of computationally cheap “specialization” on a particular distribution of data, which can be useful if you’re mostly in-distribution but introduces new risks when out-of-distribution.
Whereas PCA just drops the residual and makes modest claims, a quadratic decoder is trying to predict the residual and on out-of-sample data, it can be wrong in biased ways that PCA is not. In other words, it can hallucinate.
But if on a large enough training corpus, chances are we’re going to be in-distribution most of the time, so maybe this could generalize well.
Right now Alexa+ and Gemini are objectively better.
The best is ChatGPT voice mode. It understands non English words and accents amazingly well, and even though the LLM model isn’t the full fledged one, I can have deep conversations with it for an hour without it missing a beat.
Speech to text should work. I regularly have to manually edit the transcribed input. The more special words the more frequent. Completely disregards the context of the current input, for example, on Hacker news might involve special technical and IT vocabulary.
Any of the LLM-based ones should pull this* off - so that's to say.. none of the popular commercially available ones, yet?
Alexa+ does, but I don't use it for anything except kitchen timers and home automation triggers, so I can't speak to how well it works in a longer conversation.
Zoom's meeting notes excels at this, Google Meet is terrible at it. Meet mishears our company name about 90% of the time; various attendee names are a coin toss.
* "this" being: context consideration in speech-to-text/transcription.
The iOS equivalent would be Shortcuts, which, while not as powerful as Tasker depending on the context, is an official Apple feature that most apps support. Claude and ChatGPT both have various Shortcuts hooks, including voice conversation.
The experience of having to tell Siri to "Ask ChatGPT <about something>" really sucks, though. It doesn't consistently do it, the handoff frequently just stalls out and you never get a response, the transcription that gets passed to ChatGPT is low quality, etc.
And though I have the feature enabled that should cause it to ask ChatGPT about things it can't answer, that works even less frequently.
But even if all of these things were true, the stuff on your phone you would expect to be exposed to the model as available tool calls, are not. So their efficacy is very limited.
Oh I was just thinking creating a shortcut that you'd tap on your Home Screen/control shade (whatever it's called) to activate ChatGPT, or wire up to the action button. I forgot you can have Siri do the "ask ChatGPT xyz" thing – I agree, that integration sucks.
I'd definitely do the former. I don't even think this is specific to ChatGPT or Claude's apps.
There seems to be something about how intents get triggered by Shortcuts on iOS that feels flaky to me. Whenever some app suggests a shortcut (most recently Starbucks promoted a shortcut that orders your "usual"), the success rate when I tap it is <50%.
It's possible it's uniquely worse on my device, since I haven't done a "clean install" (vs letting the device upgrade flow copy over) in like a decade. But I'm also not up for dealing with the pain of setting up from scratch just to find out it's bad on a fresh profile, either.
Alexa+ has been a massive downgrade for me. It's extremely laggy and constantly misunderstands me, whereas the old one never did. "Set a timer for 20 minutes" used to be instant and just work, I did this the other day and it took 10 seconds to respond and set a timer for 10 minutes.
Same here. I can see why LLM-driven voice assistants makes sense to product people in the abstract, but introducing non-deterministic behavior into a device I primarily use to help with timekeeping and control lights is nothing but a regression.
I concur that the ChatGPT voice mode is excellent. I can't even think of anything to knock it for other than for whatever reason it never 'hears' my kids, but that's probably because it's not intended to be used in multi-participant chats?
But for one-on-one, it is a really outstanding experience. Especially since they tamped down the way over-the-top humanisms.
My preference, however, is for a voice-control UX just like I get with my Amazon Echo and "classic" Alexa like I have been for the past 10 years I've been using it: I think I can best describe it as a "voice-driven command-line" just like your OS' CLI shell, which makes its interactions predictable, even if it means I need to "know" what commands are valid in a given context. We all need predictability and reliability when it comes to my home-automation integrations.
...but computer interaction with a LLM / transformer-driven / "AI agent" is anything but predictable. When Amazon opted everyone into Alexa+ I agreed to give it a go and see if it really made things better or not - and it did not. I opted-out of Alexa+ and went back to something actually reliable.
Here's a question: I don't understand the gap between these LLM powered voice agents vs CLI coding agents, the latter of which are obviously useful and quite resourceful at getting something done when asked in plain English.
Seems like an agent given 20-30 tool calls like "read_sms" "matter_command", and "send_email" would be able to work out what to do for things like "set the house to 72° and text Laura that I did it."
> Seems like an agent given 20-30 tool calls like "read_sms" "matter_command", and "send_email" would be able to work out what to do for things like "set the house to 72° and text Laura that I did it."
Incidentally, a major headline in the news this past week was about a coding-agent that wiped its company's entire system, including backups; which the company's staffers were confident was utterly impossible (as it didn't have any access to that system), and yet somehow, it did[1] (the TL;DR is the agent randomly came across an unprotected God-tier admin API-key/token saved to a personal text-file in a filesystem it had read-access to). If an agent can do that with only read-only access to a company's routine/everyday storage area then there's no way I'm giving it the ability to deactivate my house's fire-alarms and security-cameras via Google Home/Matter/Thread/HomeKit/X10/OhFfsNotAnotherCloudBasedAutomationScheme.
If you are really worried about that, the agent already has that access since itll go find that key anyways.
the HN thread about that case was much more of a "why are you putting your prod keys in random text files" and "the sota in prompt engineering is that putting DONT FUCKING DO THE BAD THING" makes the agent more desperate to get stuff done
putting limits at the harness level would do just fine. one LLM call, one tool call per voice message.
Siri's one job I care about is doing exactly what I want while I'm driving. I need it to check my text messages, take dictation, start phone calls and deal with music. I don't need to have conversations with it, I need deterministic responses to known commands.
Whenever I see one of these comments, it's always from someone that tried it at the start and then gave up because of a bad experience. And many times there are more people commenting back that this was essentially the 1.0 version and that the current 2.0 version is much better. So as someone that uses none of these products (old voice assistants vs. ai ones) it's really hard to evaluate if any of these anecdotes mean anything.
You could have tried Alexa+ at the start when it was shitty compared to plain Alexa, and maybe it's better now. But equally none of the people that comment that it is "amazing" in its current iteration qualify their statements with their experiences comparing and contrasting the old version vs. the new version making them seem either unqualified to make statements based on how much "better" it is than the old version or at worse they are shills (paid or not). The best take is that they are comparing (e.g.) day-one Alexa+ vs. the current Alexa+ without a comparison to the original Alexa.
... which is to say that it really feels like there are no clear conclusions that could be drawn from all of this.
No matter how good the LLM features are, I just want to turn my lights on and off and check the time. A perfect LLM could maybe perform on par with a simple deterministic command system for these tasks, but not better. All an LLM does is introduce the possibility that a command that worked fine yesterday will randomly not work
Also, one of my first interactions with this Alexa+ thing was “how long is it until 8:45am”, one of only a few commands I use it for to work out how much sleep I’m getting, and it proceeded to ask me what the current time was… I immediately turned it off after that
> All an LLM does is introduce the possibility that a command that worked fine yesterday will randomly not work
Aren't hallucinations part of GenAI? I would assume that "AI" voice recognition doesn't have that baked in, but I'm not working in either of those spaces so maybe I'm missing the details. So many things are being looped into the "AI" umbrella that would have just been called machine learning or pattern recognition a decade ago (e.g. "facial recognition" vs "AI" at a time when "AI" also means chatbots like ChatGPT).
The point is Amazon is adding an “Alexa+” mode that uses LLMs. The plain voice recognition + keyword matching or however the old version works is more reliable (I assume, I didn’t use the new mode much because it immediately failed at what I wanted)
> that tried it at the start and then gave up because of a bad experience
I've had enough bad experiences with products that never got better, or just got worse (Exhibit A: Windows 11). Like most primates, I am capable of learning, and I've learned that once a consumer product/service goes bad there's little hope of a turn-around. I accept that you're telling me that it's gotten better, but of the people I know IRL who also use an Echo, none of them have told me that Alexa+ is worth trying, let alone committing to.
Yes, it's on me for not giving Alexa+ a second chance, but I'm not willing to give Alexa+ a second chance because, as a technology product/service customer, I just don't feel respected by the industry I work for (...lol); if Amazon, Microsoft, Google, et al won't respect me, why should I venture outside my comfort-zone for... what benefit, exactly?
> I accept that you're telling me that it's gotten better,
I'm not telling you this. I'm basically saying that with Alexa/Alexa+ and with Google's Gemini vs Goole Now(?) I've seen many posts like this. Where someone complains about the AI version, but then there are other posts that come in and claim how much better it is. Even for things like Claude Code you get people complaining about how many mistakes it makes, and then people coming in and saying that it's because they are "doing it wrong". Either "Claude has improved by 10x in the last 6 months. It's so amazing! If you used it a year or so ago it doesn't even compare!" or "You aren't using the most expensive tier of Claude which increases context and thinking abilities that are hobbled in the cheaper versions!"
I never really see a comparison on the same level and it sounds like people talking past each other or some people having legitimate complaints and then others coming in to shill for a product.
I'm not in anyway implying that "You should totally try this out now that they fixed everything" or anything of the sort. I even stated that I don't use any of these tools, and I was commenting as something more akin to an "outsider."
I don't run Windows 11 so I haven't taken a look, but I speculate it's because it contains a bunch of ML blobs for Windows Photo's image-classification and photo subject/contents keyword search.
On Windows 10, the Photos app package is about ~140MB on my computer. A good chunk of that is because the package includes a lot of dependencies - including platform deps that I'd expect would be part of the UWP runtime in the OS - kinda like how since the introduction of Swift/UIKit/etc in iOS the IPA packages all bundle their platform dependencies, even though they're demonstrably redundant, because UIKit isn't an OS-provided framework anymore... I'm not up-to-date in the iOS dev scene so I'm unsure why Apple went with that approach.
I'm not an Alexa user myself but I have watched my wife interact with it for around 5years now.
The new Alexa powered by an LLM is objectively better that previous Alexa in a few ways. This much was apparently from day one and has only gotten smoother.
1. It can reliably execute direct or vague-ish commands "play X movie in app Y" or "play x show" and can infer X movie is only available in app Z so use that.
> It can reliably execute direct or vague-ish commands "play X movie in app Y" or "play x show" and can infer X movie is only available in app Z so use that.
...how does that work, exactly? (or rather: what's the context here?); there's no possible way for an Alexa+-powered Amazon Echo to control my AppleTV or interface with VLC on my desktop.
It's not the early 2000s where just messing around and wasting time on this stuff is cool in itself. None of that time wasted turned into much long term apps that stuck with me. Maybe a banking app and a trail running app.
I ruined multiple dinners with timers that didn't work (with a time/labor cost).
I had to get out of bed in the freezing to turn the lights out. It's easy to hit the lights when I go to bed but annoying having the tool fail and getting back out.
Music stuff didn't work well because I used Youtube Music not Spotify.
Those were my 3 use cases for Google voice, and it failed them all enough I just stopped using it all together. Who cares if it works today if in another month they just change something and break it again? They've shown it's not a tool to use for tool things, it's a 'gee wow' thing. I don't need to be impressed. I need not burnt food.
Alexa+ is terrible compared to Alexa. It's so bad that I've dusted off my v1 echos cuz they're too old to run Alexa+. Complete shit show that is.
I do like Gemini better than Assistant, even though it's not quite there yet. But that's just a matter of time because they actually designed it from the ground up to be a drop in replacement for Assistant.
I read this article 10 years ago by a guy named Ricky Yean who went to Stanford as an economically disadvantaged admit and couldn’t shake his poverty mindset and it cost him when he was running a startup.
Why “few successful startup founders grew up desperately poor”
Poverty mindset is maladaptive because it teaches you only money is worth anything, so you hoard it. But in truth time is also worth a lot and sometimes it’s wise to use money to buy time.
> Poverty mindset is maladaptive because it teaches you only money is worth anything, so you hoard it. But in truth time is also worth a lot and sometimes it’s wise to use money to buy time.
This is something I've observed in overly frugal family. Stuff/money is worth too much to them.
You can't gift them anything "nice" because they will put it on the shelf and never open it (don't want to damage it).
Gifting them consumables with an expiration date also doesn't work as they'll "save it" until the expiration date lapses and then eat expired food.
Taking them out is weird because they'll insist on taking leftovers home off every other persons plate at the table, including stuff they don't normally eat.
They'll fill up 2 bedrooms in their home with 40 year old cheap clothes & furniture that is worth so little we'll need to pay someone to haul it away. They won't donate it because they think the people who receive it will waste it. So they'll pay money to ship some of the 40 year old cheap clothing to poorer family back home who it doesn't fit and could just buy cheap clothes there with the money they paid for shipping.
Owning multiple 30+ year old cars until the mechanic literally refuses to work on them anymore telling you they are too rusted out to repair or drive.
This from people who are wealth enough to own multiple properties, have retired early, have government pensions, etc.
I grew up poor. Everyone I knew was poor or barely getting by. It took me a lifetime to realize over-frugality can also be self-destructive.
My extended family members use to trade stories of working dangerous jobs or getting mugged and laugh it off. That was until my aunt got shot in the head working the late shift at a convenience store. She was in the news and the community and the hospital paid all her medical bills; but she was never the same after that.
Growing up gambling was a mysterious force that ran through my family. They worked hard. Like their bodies were physically wrecked and they saved and saved. They would often spend so much time at work or at their side hustles, they would neglect their children who would teach themselves how to cook because no adult was at home. And yet, they would blow their entire savings in a weekend in Las Vegas. Or sometimes came back even in debt.
Now, I would say that frugality is not a moral imperative. Being poor is not a virtue. Sometimes fate deals it to you. But no job is worth wrecking your health. And it is not a sin to spend a little on something that makes life a bit sweeter.
But I've found that distraction is the catalyst. Creativity for me comes when I focus on something else for a while, not grinding on the same problem with unwavering focus.
reply