As an Arab person myself (and not getting into the Palestine occupation debate). It's sad that a non-Arab speaking nation is creating LLMs to better understand Arabic but I am not aware of any Arab nation doing the same effort.
I would say that it's sad that LLMs are being used for mass surveillance, and such applications are things no government should be proud of deploying.
And it's things like this that are ripe to come to other countries. An interesting book that goes into this concept is "The Palestine Laboratory." Sad times indeed.
Why would LLM training allow arabic native speakers to "better understand Arabic"?
This isn't a tool to better understand arabic in some innocent sense, it's a compensation for israelis refusing to learn arabic and having trouble gathering intelligence from linguistic sources. When you're actively involved in the genocide of a people you'll also have problems finding informants among them, it's much easier under a less intense apartheid regime.
" will AI be a bicycle that we control, or an unstoppable train to destinations unknown? To put it in the same terms as the ad, will human will and initiative be flattened, or expanded?"
Interesting line of thinking here. I've never considered this but an observation from my usage of chat AIs at least is an increased willingness to defer my thought-organising process to the AI by just jotting down some random ideas and asking it to make it coherent. I guess on a personal level I'm flattening my own mind?
I'm in two minds about it - in the strictest literal sense, it's true - but it also rhymes with Plato complaining about writing degrading the ability of scholars to memorize things.
Actually, Plato is not wrong. I'm a big pen and paper user, and write everything down. While it allows me to turn back six months, I can't turn two weeks back without my notebook in some cases.
Brain has a strange ability to understand that information is stored elsewhere. Write it, and you forget it. Take photos in a concert, and memories become fainter. Talk about something important, you start to forget it part by part.
AI is something way bigger. It's a dumb system which mimics us without the essence of the human or personal style, yet does mundane tasks without questioning. So you trade your own fine tuned, honed, polished ability to some GPUs which runs a software which is tuned for the masses. Like trading your beaten tungsten tool with a shiny iron one. Looks nice but way inferior. The ability to do these mundane tasks is the foundation for not doing the same tasks in a mundane way, or doing more complex tasks built on these mundane tasks. You rig your own foundations with explosives. Not wise.
I'd not do that. I don't use any AI systems in any of my tools.
> Take photos in a concert, and memories become fainter.
I don't know about short- or medium-term, but long-term the photos can remain objective anchors for memories that reduce the memory drift for everything all around them. When you have a photo of an event, even if you didn't take it yourself, it locks in many facts that individually weren't important enough to memorize but do constrain other possible facts and thus keep memory more accurate.
When I got into photography I took high-effort high-quality photos everywhere I went. There are so many minor events I wouldn't remember at all if not for the photos, and even for events I do remember, I likely would have forgotten that certain people even attended.
Long term, I can be completely objective about the times, places, people, conditions, etc. of a great many events in my life, and every time I refresh and reinforce memories it's anchored in those objective details.
There's a less objective angle to this that I still acknowledge and enjoy. When you take a good photo of a good moment, and look back on that as representative of the event, it has a way of making the entire event look that positive. You scroll back through a timeline of photos like this, it can make entire years of your life look as good as you want them to.
That's part of why I think there's a big difference between scrolling a backup of all your phone photos vs hand-curating albums (regardless of the device that took them). When you choose what you'll see in future you influence how you're going to feel about it, and that's an under-appreciated mechanism for investing in your future headspace.
Note: I'll take short snips from your quotes to keep this comment tidy.
> don't know about short- or medium-term, but long-term the photos can...
You're absolutely right, however I said concerts for a special reason.
> When I got into photography I took high-effort high-quality photos everywhere I went...
This is also what I do. When walking around in a city, or taking photos in a vacation, etc. You can do this. Because when taking photos in a relatively serene environment (when compared to a concert, esp. an open air, festival one), you can internalize whole event before taking that photo, so that photo becomes an anchor for that event. In a concert, everything is so fast. Trying to concentrate to take a good photo makes you ignore a large chunk which prevents you forming that emotion and memory.
> Long term, I can be completely objective about the times, places, people, conditions, etc. ...
True, but as I said, you had time to internalize that event before taking that shot. This is what it creates the anchor and reinforcing effect. You can't reinforce a memory you didn't form.
> There's a less objective angle to this that I still acknowledge and enjoy. ...
Photos bring joy, but not always. There are many positive photos which makes me feel bitter, or even sad.
> That's part of why I think there's a big difference between scrolling a backup of all your phone photos vs hand-curating albums ...
I can relate to that, but I also take a different approach. If I have the time and feel for it, I challenge myself with "1 scene, 1 shot", regardless of the camera I have with me. This allows you to ingest the scene you're in, take a purposeful look and decide which photo to take, which forces you to create a relationship with your surroundings. When you take that photo and revisit it in the future, that frame will make you remember the whole area where you searched for that one shot, and will bring tons of memories and visions back.
If I don't decide to or can't do that, I curate albums as you do, yes.
I still lug around an actual notebook to all meetings and take notes by hand. I approximately never look at those notes later, because I remember everything important that I wrote.
I've tried every variant of taking notes electronically and the result is I don't remember any of it. Counterbalancing, it sure is easier to grep a text file than search on paper pages in my notebook.
But, having the info in my head is still orders of magnitude faster than grep, so taking notes on paper wins for me.
When you're writing about a subject or meeting notes, sure. When you're writing a 30-item to do list, your brain will say "Nope, everything is on that notebook, refer to that".
As I noted in my original comment and elsewhere, I'm a heavy pen and paper user. I finish approximately two notebooks a year, plus I do my thinking/research on their dedicated notebooks.
Personal knowledge base and other digital notes are just distillations of these notebooks in a synced/searchable form, and I'm slowly making these public, as I have time.
Putting that there's no guarantees for correctness for AI models aside, you miss the opportunity to read the docs and learn (or at least be aware of) the whole capabilities which may help you in the future.
You can bookmark the relevant doc page, and return to that whenever you need it. Instead you now have a buddy which you can bug and get answers in general, so you slowly wither your ability to do research, read more complex docs, and learn with collateral information (i.e. you learn something, and be aware of other features, so you slowly learn and internalize the whole subject).
Instead you ask, and get a nibble of information which is harder to connect to other bigger corpus you might have. You probably saved time this instance, but if you read the docs, you'll progressively spend less time on the subject, saving you tons of time down the road, plus you'll sharpen your ability to read docs and be faster at searching, reading and understanding them.
These arguments seem like they'd apply equally well to googling for the documentation to a specific function and then using the answer from the preview snippet (or clicking the link, reading exactly what you need, and then closing the tab without reading any further). I'm not saying that means they're wrong, but it's hard not to feel like it would be hyperbole to call that "rigging your foundations with explosives".
It's a matter of habit. I personally don't search for answers on the internet to begin with. I use Zeal/Dash to store docs of the tools I use locally, and directly read reference docs for what I need. If I need information from a very specific page, I always bookmark or take note of the specific page, and return to that automatically. I also always read beyond what I need to see whether I'm missing something or holding something wrong. If that's I use very frequently, I further document what I do, and how it works in my personal knowledge base.
In surface asking ChatGPT is not different using StackOverflow for everything, but I can argue that StackOverflow bears the same dangers, unless the answer given is comprehensive and written with collateral information in mind.
However, having a personal assistant on tap which can answer (or hallucinate) anything or everything will definitely make you lazy.
F#, Jupyter, Python, R, MySQL, PostgresQL, ZSH (awk, sed, fill in the rest), VS Code, Excel, Word, git, and probably more.
This weekend I continued to work my way through Learn You a Haskell For Great Good, taking time to go over the functional programming concepts in both Haskell and F#. In addition I read most of Data Science at the Command Line, where I was introduced to ggplot2, so I then worked my through the online ggplot2 book.
I'm not very interested in memorizing the complete syntax, standard library, and common third-party libraries for the myriad of tools that I've listed above. I don't really see the point of learning how to read through the MySQL documentation.
Here's what I want to have in my brain: Most of Townes Van Zandt, Willie Nelson, and the Grateful Dead's catalog of songs so I can play them on acoustic guitar without a songsheet as I like to make eye contact with the audience.
So yeah, thank you for telling me that my approach is slowly withering my abilities.
I'll put this as nice as I can: You are very judgmental.
Since you seem a bit touchy about the feedback you specifically requested, I'll take a tangent:
How are you finding ChatGPT for functional stuff? I found it to be unusably bad, unable to transform trivial programs. Have you found it helpful for Haskell or F#?
FWIW, it takes practice to get good at using ChatGPT. You have to direct it in certain ways and break up problems into bite sized chunks to do complex things like, eg, assist in writing a language server for a custom DSL using FParsec. If you don’t have a higher level understanding of the task and don’t break things into smaller tasks it won’t work very well!
> I'll put this as nice as I can: You are very judgmental.
Thanks for your direct and honest view. No hard feelings here.
Let me tell you. I know C, C++, Go, Java, bash, Eclipse IDE, git, Docker, Saltstack, Terraform, OpenStack, Kubernetes, some MATLAB and probably some other tools I forgot that I know, and I manage a big fleet of servers while I'm writing this.
I don't "remember" the syntax of anything. I somehow internalized them. I don't think about them. If I make a mistake, my text editor (which is NOT VSCode) politely tells me about it.
I also used to remember double bass parts of Bizet, Beethoven, symphonies of local composers, plus tons of songs, because while I had the sheets in front of me, I had to listen tubas & percussion to make sure that I'm in sync with them and watch the conductor to double-check the metronome in my head and get the tone cues if he's not happy with our tone. To make sure that our 100 person orchestra was playing at its peak performance I had to make sure that I know every kink and chicane of the traffic for our specific arrangement.
In these days I remember tango songs' traffic because I have to plan my figures 2-3 sentences ahead while dancing in a crowded hall.
Oh, I sometimes play a couple of songs in my bass guitar if I have time from other activities.
So, yeah, thank you for telling me that I'm judgemental.
I'll put this as nice as I can: The choice is yours, but you're underestimating your abilities. Plus, people who like to read docs and write code the hardcore way are not dorks.
You're not giving up your ability to reason, you're just reasoning at a higher level. Think of AI like an employee, CEOs don't lose their reasoning ability because they have employees to make their directives into reality.
Depends what the alternative is. If on average people would lose that fleeting thought and never dig into it or research it more it can be a huge net positive to write them down, even disorganized and having a history of what you thought about. And actually debate the thought with the AI and get an outside perspective.
On the other hand if on average people would in fact dig into them, you might lose the ability to organize your thoughts yourself over time, but I always think that sort of fear is overblown.
I have found that writing everything down eroded my ability to remember a couple of important tasks for the day, so I started to exercise that part of the brain to get the ability back.
Any capability of the brain not exercised will certainly wither. While brain consumes the most energy, it tries to minimize it at the same time. Using AI extensively to accomplish mundane tasks will rob you of the ability in the long run.
> an increased willingness to defer my thought-organising process to the AI by just jotting down some random ideas and asking it to make it coherent
...and then accepting that outcoming as sufficient, and moving on, right? Is the alternative that without the AI, you would spend more time researching and thinking about your ideas, pursuing (or stumbling upon) related ideas, and maybe ending up with a more broad understanding at the end of the endeavor?
If so, working with a chatbot is good to extent that you value economy. That is a worthy outcome in many circumstances. But the "inflated" alternative sounds valuable in its way too.
From a broad social/behavioral perspective, I wonder the extent to which the "inflated" way will die off (or perhaps adapt and become something new?) given these new tools.
> > jotting down some random ideas and asking it to make it coherent
Isn't this is more or less the same thing as spending a bunch of time going to a bunch of webpages from Google search, except it takes way less time, arguably leaving _more_ time for the critical thinking and subsequent research you espouse?
> ...and then accepting that outcoming as sufficient, and moving on, right?
Much like you can either immediately trust what you read on the Internet, or apply some critical thinking and dig into it more, you can do the same with whatever an AI returns to you.
AI can be an accelerator for mundane tasks and help you get to the high-value work quicker, or it can be a lazy shortcut that lets you do less.
You can't separate Will from attention, attention spans or human initiative was already shattered way before AI, which is now the best hope for more people to leverage time and control to expand their minds. I'm sure that centaur-mind will be the norm soon, if it really is as good as it promises to be it won't be needed for long.
This goes beyond flattening. Given that only a handful of companies have the financial means to do AI at scale, and that those financial means are provided mostly by ads, I think that allowing these systems to essentially "think" for people like this will be unbelievably destructive for society long-term.
I think sometimes we underestimate how impactful 10 minutes can be. Imagine sitting in a call, waiting for folks who are late to join the meeting, or that's five 2 minute video ads on YouTube - Google is taking 1% of your day in ads.
I think it increases the perceived value of 10 minutes, especially when looking at the visual representation.
Once you account for your own non-negotiable slots it may well be 5 or even 10 percent of your (available) day.
I don't get the same effect when looking at an agenda planner as the typical day to day items are not present. There seems to be time to fill on calendars, the opposite here.
The first browser to index (locally preferably) my browsing history and help me extract/surface useful information from previous pages I have visited will win me over. So far they all seem to be offering the same kind of AI that will crawl the current page and extract information from it.
I guess this can be done with an extension that would work on all browsers. Like parse each page, save compressed text that supports fast decompressing (maybe even images), build tf/idf + cosine distance for verbantim searches & save in db and maybe add some ai to enhance the search. The next steps would be custom ai searches like "show me pages related to x", it'll quickly decompress the pages & apply the search for each but there are different approaches, like maybe saving history on a server and apply some ai model with huge context or do an initial filtering with tf/idf and apply ai after that.
i’d want a feature to only do this for pages i mark as interesting (so, a bookmarked page i guess). i may not care to recall the match summary of as milan vs napoli.
i have rafts of bookmarked pages, but i don’t know which of the 45 postgres-related pages contains a helpful sentence or phrase about indexing that i think i remember reading a few months ago.
it could be done with a boolean flag. Like the addon would work for all pages, but you would be able to filter the results by using some checkbox "Only favorites" or something like that
Fully local (or private cloud sync) I would go as high as $2 a month. Cloud based solution, which is the only solution I expect, not interested just yet.
$0, unless maybe someone finds a way to not pillage the data of the service and add a lot of actual value, and doesn't squat on this ultimately simple thing that should be a small piece of a much larger thing.
open source AI and browser/OS companies are going to provide an answer soon anyway.
- extreme corporate profits
- better wages
- US jobs
- lower prices
You can’t optimize both for extreme shareholder returns and extreme wealth at the top and also pay first world good wages/living standards and have lower prices. Unfortunately, in this race, shareholder returns is often still the last thing compromised on, it seems.
That's not actually pick three. It's that if you have uncompetitive markets (and therefore high rents/inefficiency), you end up with worse wages, loss of US jobs and high prices. Because then the price companies can charge is no longer tied to their labor cost, so they can cut labor costs without lowering prices.
This is clearly what's happened and needs fixing, but even after that, you still have a trade off between high-paying US jobs and low consumer prices.
But the answer to that one isn't as obvious because it isn't linear. In a competitive market, competition with labor in other countries might cause you to get paid $1000/year less, but lower your cost of living by $2000/year.
> can’t optimize both for extreme shareholder returns and extreme wealth at the top and also pay first world good wages/living standards and have lower prices
Ignoring new entrants, sure. In a competitive market, shareholder returns entice new entrants. Flatten those and you lose that edge.
For many markets, the competition caveat is missing: this is something we can improve with policy. But pushing down profits for labor's sake is a false economy; it leaves the industry less resilient in a global context. Put another way, a solid repeatable business plan is finding a market leader suppressing shareholder returns and exploiting what they're missing from a separate jurisdiction.
I understand that, but if better wages lead to increase costs, it should be a zero sum game in the grand scheme of things. But, at least, you are not dependent on China for making stuff.
Can you state the price of the "packages" in the landing page?
Also, does it always give me 8 bullet points from each video? It would make sense for videos that have chapters to give summaries to each chapter instead.
Yeah, sorry. I'm changing prices every 2 days here, and I want to try subscriptions instead of packages as well. I'll put prices on the site in 1-2 weeks!
Currently the prices are:
$8.6 for 20 summaries ($0.43 each)
$19.8 for 60 summaries ($0.33 each)
$48.6 for 180 summaries ($0.27 each)
> Also, does it always give me 8 bullet points from each video? It would make sense for videos that have chapters to give summaries to each chapter instead.
Yes, it's now a fixed number of 8 parts. I tested different numbers, it seemed to me a universally convenient simple solution for videos of any length. But yes, I will take chapters into account when splitting, right now I ignore that information.
At the same time, the product isn't as useful as if when there are no chapters at all.
Calibre is awesome to use on desktop. But as someone who wants to sync his Calibre library with other devices (iphone + ipad), it gets really frustrating.
Do you want to sync the whole library with the iphone or just search and download from the iphone? Calibre can be used to just transfer a few books by installing the calibre app on the phone and select books on the computer to transfer. There are also solutions for searching and downloading books from the phone but I haven't tried them much, found it much easier to use the computer to do that.
I can't fit my library on the phone so haven't even thought of syncing the whole thing.
I want to sync the whole thing. Apple Books ineptly forces you to manually copy all of the books themselves to each device, but then syncs collections that you've set up to organize them.
Have you tried Resilio Sync?[0] I'm using it on Android but seems to exist for iOS [1] too. I'm not syncing the library though, might want to make sure you only run the sync when Calibre isn't running to prevent data loss.
It does, but there is an ongoing bug [0] - which I suffer myself - that is always removing the local version of the book. I grew tired of always having to redownload books after a few days.
I wish that were true; but I just checked the computer I'm typing this on, and it's missing books from my other one (which is logged in with the same iCloud account).
Enable the Calibre web server, then use KyBook to connect to it and download the books you want while on the same network as the Calibre server. You can set up a public URL if you want to access them remotely
I use dropbox for syncing. I point my calibre library to the dropbox folder, and in my ipad I open the dropbox app and "share" the book I want to read to the Books app.
The calibre manual warns against using dropbox on the main library folder, users have lost their complete libraray doing that. Might want to at least pause dropbox while running calibre.
I have using calibre with dropbox for 10+ years now. The warning was given out mostly for the internet speeds of the time where the time for your edits to be synced took very long. I had a problem only once in 10 years even then because of dropbox revert file ability was easily able to revert.
The only caveat is I do not use calibre server so I usually start calibre do what I want and exit. If you use the server and keep calibre on all the time then Dropbox might not work as well for you though just restarting calibre after any library edits should be enough.
Awesome is disbutable. The amount of awesome parts is equal to the amount of aweful parts. Calibre is really a prime example of product which shines despite the manifold deep flaws. And this makes it really problematic, as it seems to dominate it's space so strong, that no many even care to work on an alternatives. Very problematic for the health of the ebook-space.
It is easy to criticize, but it takes a toll to fix books in ebook processing.
I had to do some minor ePub work and I was like „WHAT, that’s the standard? This can’t be right“ but yeah the formats are trash and complicated.
To be dedicated over a decade to fix and develop software is hard.
It's incredibly cumbersome to set up for even the most basic, common tasks that everyone wants to do. He's forced to release constant updates because he refuses to separate the news processing code from the main application code (the processing "recipes" should be separately distributed) and worse, there's no auto-updater, so keeping the application up to date is a pain in the ass. The UI is outdated, hostile to users in general, and not low/impaired-vision friendly (six icons in the standard main toolbar all look virtually identical save for very minor differences, and Goval loves different shades of grey, reducing contrast.)
Last but not least: it's almost exclusively developed by one person (a problem by itself for such a large and widely used project) who is infamous for being at best abrasive and at worst an asshole - and not a particularly good programmer. The commit history is an absolute mess to try and navigate because he seems to have "save" and "commit" confused.
Edit: the very link itself is a perfect example of how Goyal seems to have zero awareness or care for others. Why do I have to click to expand text items in release notes? And he says that some plugins are no longer compatible because of the switch to Qt6. Which ones?
Being an asshole doesn't have to be about what you're doing, it can be about how you're doing it. In fact, that's normally how I see it used -- referring to an abrasive person who's difficult to interact with.
(I'm not commenting on the specifics of Goyal at all; I don't know about their situation.)
I don't know how to put this in a formula, but my observations about my family's wealth disappearing is the following:
Great grandpa had tons of land, dies and leaves it to his children who then left it to their children when they died to split as heritage (so my parent and uncles and their cousins) which they ended up selling and splitting it between around 60 heirs. They bought things that are not really an investment like cars, new furniture, vacations...So all the wealth from my great grandpa has vanished because his great children like to spend money rather than invest.
I wonder how wealthy your great grandparent was. Cars aren't relevant purchases for the types of wealth I'd consider under examination for this type of article.
It's tough because a multi millionaire family certainly has wealth worth discussing. It's just such a different ballgame from a family worth hundreds of millions of dollars.
I understand this is all relative. I come from an African country but even there around 250 acres of both residential and agricultural land is significant money. I'd say only about 20 acres are left where the owners are one of the ancestors.
My dream is to one day repurchase some of the lands and have financially sounder children that could take care of it ;)
It's a good point. I can I imagine a family having the local purchasing power (homes, a night on the town, etc) of a hundred millionaire family but that wealth diminishing quite quickly when used to consume luxury goods marketed to very expensive countries.
