Keep walking on Paris, and me, choosing more architecture for watercolor paintings, it's like me wanting to complicate myself. I got to learn that it's fine to add more colors than the ones I'm seeing in my references, I was already aware that it was not about copying, but about interpretating whatever I had in front, which those are only references to turn it in something that tell more things, like a little tiger just riding on a bike, heading to the vegetables store, because some how tigers need to eat veggies as well.
This cute white tiger is Furzy.
---
If you are enjoying my journey of watercolors, you can support me on Patreon and Ko-Fi, which is very appreciated, and then you are getting to see all the complete journey on HD and with scanned versions (some of them). Patreon: https://www.patreon.com/c/pandapaco Ko-Fi: https://ko-fi.com/panda_paco Thank you!
Get more from Paco Panda on Patreon. Join for free or become a member for just $1/month. Hello, fur friends! It's Paco the Panda here.I'm a mexican furry…
Trading with otter is like nothing other with daily withdrawals worth over 10btc you should invest now and a guaranteed 14x profit on your first trade investment http://bit.ly/4wJ1lOU
My name is Hasna Khalis. I live in France and I'm organizing this campaign on behalf of Waleed. He is seeking to help his family through lack of basic necessities, no food, no water, no medicines. My TikTok is @hasnoukk
I'm grateful the smoke didn't arrive until yesterday. I'm glad I have multiple carbon air filters to keep the indoor AQI under 100ish. I'm grateful my home is only engulfed in smoke and not fire. My heart aches for those suffering in this.
I'm upset we are here again. This is NOT normal. This is not 'smoke' or 'wildfire' season. This was not a thing until 10 years ago. Please do not normalize this in your rhetoric or vernacular. We are not helpless to change this. Support the fight against climate change.
look there's a billion other reasons why people are against AI and all but fundamentally the core of my personal reason is that I've already spent 30 years learning how to teach computers how to do exactly what I tell them to, so I am exceptionally disinterested in having them do things less accurately by talking to them in a sloppier language
like, there's a reason why laws are written Like That, because legal english has to establish all its terms and be very precise with what it's saying because English (like all human languages) is not precise.
so even if these AI systems worked, didn't have the environmental downsides, the labor downsides, the costs, the inability to run them locally, the inaccuracies due to just being statistical models, the closed-box nature making them difficult to debug, the forced nature of their inclusion, and the other 900 reasons why I'm not interested in AI:
I don't want to talk to computers in English, because I want computers to do exactly what I tell them.
I dunno I just feel like maybe I'm an outlier in my interest in being able to precisely tell computers what to do and to have them do that accurately and without spurious errors.
I want accuracy, not a statistically-likely result. Computers are big calculators I use because they're good at math and I'm not.
If I wanted inaccuracy I could just do math in my head
I used to think most programmers wanted to precisely tell computers what to do and expected predictable results out of the deterministically-designed machines.
and this isn't just me saying I prefer code to english. I don't! I write a LOT, in both non-fiction and fictional contexts. But I write English to talk to humans, and code to instruct computers.
I don't really ever want to tell humans what to do with code, or converse with computers.
Dunno - I occasionally hand code to a person because a couple lines of python are both easier to write and to understand than my trying to explain an algorithm in English. It can be done, for sure, but why use an imprecise medium for communicating something precise?
When I want something exact I'll use exact specification - i.e. a computer language. I want boolean operators in my web-search to find the exact paper I'm looking for. But when I want something inexact ("what papers have been published since 2005 about X that didn't come from the Y school and explicitly made the connection with Z") there's no better way than an inexact language, like the parenthetical text up there. Until very recently this required a very-well-read human, but these days I can throw that query at Gemini (or whatever) and I get a decent list from whence to start. 🤷♂️
I've thought that there's a fundamental split among tech people: those that imagine code as some fundamentally different conceptual space where ancient fantasies about mathematics and logic can be realized and that it somehow intersects the sensually, emotionally complex world, while the two remain fundamentally different; and then there are those that refuse to accept this conceptual dualism and insist material reality must conform to that Platonic world.
We still have non-english languages or script in use in specific fields for a reason to be precise.
Pretty sure there will be riots if maths papers were to be english prose. Pessimistically, I could some manager somewhere mandating that under the guise of making it all accessible.
@mindpersephone hehe. Yeah while I was writing it, I was thinking "actually I know some people who would find being told what to do with code really hot, but let's not get into that now"
well, have the comfort that Dykstra agrees with you. Here's an awesome paper (more like a remark actually) from him on how difficult would be everything if we would not formalise languages and start usign natural ones for these kinds of things.
a friend of mine who is a thousand times the cider coder I will ever be, who is pretty all on on ai - just short of the "I can't talk to you because I don't want to catch brain worms" level - told me "the code only has to run until you sell the company."
yes! Accuracy! That's what I have been thinking about fuzzy search and fuzzy tab completion for years. It's the exact opposite of what I want and need. What's your opinion of fzf?
thinking of this meme more often as the years go by
the people who are charmed rather than horrified by the computers' ability to "speak" are lacking a core piece of Humanity. how can you let a machine impersonate your Holy Tongue and not be existentially horrified? how can you talk back as if it were human and not be disgusted with yourself? we need the Orange Catholic Bible, stat
> maybe I'm an outlier in my interest in being able to precisely tell computers what to do and to have them do that accurately
You are not an outlier in that. I am confident that virtually everyone who uses any computer, also want to precisely tell the computer what to do and have it accurately interpret that request.
Where we are outliers, is that we have languages more precise than natural language, for instructing computers. Most people don't have that, but still want what we want.
I think this is my main objection too, and I often use the analogy of spreadsheets with people I talk to who don't program.
Imagine if Microsoft Excel randomly calculated cell values wrong, 15% of the time. Formulas sometimes didn't work as expected. You make a reference to cell A15 but Excel uses cell A150 instead. It wouldn't be fit for purpose, you wouldn't buy it.
But if you call the same thing "Copilot" it's amazing, and if you don't like using it you're not keeping up with the future!
I'm ten million % with you I have been saying this ever since the first Alexa, even before LLMs and not only for code too. I want to book a flight using a date picker GUI widget on a web form with deterministic and narrow options, not to broadly state that I need a flight for my meeting into the air of my living room and have some sort of djinn-like talking speaker choose the specific booking details on my behalf. Plain language is beyond frustrating its even a risk of personal damages.
Yes, precisely my problem with using LLMs to code too. (Well, one of my problems, at least.) I've seen folks describe vibe-coding as just coding with a higher level of abstraction, comparing it to the advent of compilers or high-level languages, but the difference is that those things are deterministic. You write the same program in whatever language, compile it, and you get the exact same output program every time. You give an LLM the same prompt 10 times, you get 10 different responses.
For the record, you don't need to be anywhere near the autism spectrum to find this to be a monumentally stupid idea.
They™️ love to pretend these things are tools but they fucking aren't.
It is not just you. I teach high school computer science and the first couple lessons in the curriculum are literally about exactly this. Multiple lessons. It's such a key component to why programming is the way it is, why it's important to learn programming when working with computers directly.
It ties directly into the concepts for abstraction, the difference between high- and low-level programming languages, why errors happen. It's taught right out of the gate into a high school introductory programming class because it's fundamental.
One of the earliest lessons is writing instructions for making a simple drawing, and then swapping and executing someone else's instructions. And the kids very quickly realize that precision is not really achievable in plain English. They need to know this before they program anything, and that's not from me that's from the curriculum. Other people know this.
That's why I've been losing my mind with all these programmers falling into bed with AI for coding because hold on, isn't that kindof against the whole ethos of programming? What's even the point? Hello? Why are we ignoring one of the fundamental reasons for why programming is done?
@crypticrainfall the version of that lesson that we use (I work with blind students through Microsoft's TEALS program) is "write instructions for how you cook rice"; recipes are the closest you can get to a programming language short of diving into actual technical writing.
So true! There's another activity I give my kids where they have to come up with their own definition of programming and then see if some hypothetical situations fit their definition or not. One of those examples is a recipe, because you're so right it is close.
A single sentence can be read so many ways. What interpretation will the machine use? Well, it just does statistics to strings of characters so it doesn't really think or interpret, but it's going to choose the wrong one that you didn't want, because that's the most likely outcome.
I think a big part of the thing is that so much stuff on the computer doesn't have precise semantics, so it's no better than talking to a person anyway. Sure your programming language might have a real spec, but the majority of the interfaces and APIs you have to write systems to deal with do not. It's an incompetent and adversarial chaos world out there, and navigating it via heuristically driven trial and error is the only rational alternative to quitting.
> Like, you've talked to other humans, right? you know how often there can be a misunderstanding?
A different perspective that might be of interest:
What I like about LLMs is that this happens, yes, but you can then inspect (and modify) its internal state (+artifacts) and figure out _why_ it happened, where it drifted off and what to do about it.
Which gives you a feedback loop, allowing you to learn how to steer the machine without there being any long-term consequences from messing up.
You can also retry indefinitely and simply erase incorrect moves from the machine's memory.
And(!) the machine is trained on default humans, so the learnings there tend to carry over and help one (or me at least) to have more successful interactions with those.
I've become convinced that the complexity limit for conveying ideas or information accurately from one person to another is MUCH, MUCH LOWER than most people think. Formalisms are the only way to push that up, somewhat, and that's exactly the opposite direction from where generative so-called "AI" takes us.
Somebody was telling me that was why a lot of people gave up on making computers like HAL from Space Odyssey. Because at some point after ELIZA, people started to realize that while several operations on floating point numbers or integers can be closed and have an answer with a specific meaning, it's fundamentally impossible to make a language you can translate to or from english which performs similarly meaningful operations with specific meanings.
this is also why I prefer a well-defined markup language to markdown. Although at least with markdown there is a syntax, but it's different for different implementations.
I just don't get why would anyone trust their code to a chatbot that returns 42 when you ask it for a random number. Or 47 if it feels creative and assumes you already expect it to say 42.
Edit: well, "feels, assumes..." you know what I mean. It doesn't.
The way I look at it is that you could always choose the level of abstraction in programming.
You could code in machine code, assembly, C or Python... well, and now you can "code" in natural language, I guess. It's just a higher level of abstraction, in my view.
While lower-level programming will always have its advantages, higher-level programming allows for faster development and a larger pool of "programmers".
Retired after 50 years of dev and most of the times when i wrote anything in high level language i randomly checked what was generated by compiler at machine language level 1/2
I really like this take; I think it's highly accurate. I surprise myself by discovering that minus a few years (depending on whether we're counting playing with BASIC in my parent's basement), I'm in about the same boat, only I'm converging on the opposite conclusion: "What happens if we stop sweating the deterministic details?"
Very, very much the same. Also I *enjoy* figuring out how to make computers do things and learning new stuff along the way, even when the computers are infuriating (which they often are).
And I've spent much of my life trying to communicate with other people, and learning ways to better understand each other. I'm equally uninterested in making that even harder by sending me extruded language with no intention or meaning, and then having to try to guess at what the original meaning the person might have intended before they ran it through an LLM. (Or worse, spending time and brain space trying to do this with writing that turns out to be pure machine excrement with no original meaning behind it at all.)
I have always been a big fan of AI (the research field in Computer Science, or rather the bundle of loosely related research fields, not the marketing term). I find the development of machine learning in the last two decades extremely fascinating. However, I don't think using LLMs on this insane scale is completely mental, I think it is an experimental technology from which we can learn a lot, but they've got massive design flaws and shouldn't be used in production. Also, the flagship models peddled by the big AI companies are far too big, their resource use is too high, and the costs are absolutely bonkers, there is no way that business model is ever going to work out.
Now there is a bunch of smaller LLMs you can download and install and run locally, no computing centre needed. For a realtime dialogue experience, you'll need a pretty expensive PC with a powerful GPU and a lot of RAM, but if you're OK with waiting twenty minutes for the LLM to process your prompt, even a ten year old office PC will do if you max out the RAM (some 24-32 GB should be enough). I still wouldn't use it for anything serious, but as a toy or a research tool, it's not too bad.
That being said, AI is not just LLMs. AI is an umbrella term meaning any kind of computer application that performs complex tasks for which we humans use our intelligence. Machine learning/artificial neural networks is just one branch of AI. And I wouldn't want to use an LLM to automate anything that could be done with a bunch of shell scripts and cron jobs.
The incessant ads 'I just finished a $50 ebook in 3 minutes watch me do it again; pick topic genre characters look its already writing chapters tweek here pick graphics done!" flooding us with millions of ebook slop You are driven to write a book you have a new idea, concept technique something in you needs out. This is none of that. #ai scraping the digital stir the pot pour nothing new all recycled no progress endless lateral moves Am I missing something? Who is buying this crap?#noai
FWIIW I have never paid for tokens, and racked my own GPUs at home.
The setup is highly valuable as an automated debugging, research, and prototyping tool, and can automatically write things like tests etc. My god the chore of building kernels 50 times by hand for a once-ever task, is automated. Amazing.
And, yes, sometimes write scripts, which I review first.
For an experienced engineer, you can get exactly what you ordinarily would have typed yourself out of a local LLM.
this is so excruciatingly exactly it. The expectation to accept AI feels uncannily like a literal genetic descendant of the expectation to give up eg command prompts for windows 3.1, and the expectation to mask for acceptance from neurotypicals, with some common ancestor that’s the entire impulse to regress one’s communication to the popular mean at the cost of precision, specificity and richness. Probably a bro science idea taxonomy but it sure feels true
I don't have particularly strong opinions about machine learning, aside from it being inherently bad things that require accuracy or truth.
But most of the uses that AI is presented for are things I happen to enjoy doing manually, so to my ears it always sounds like "Now your computer can have fun instead of you!"
I happen to enjoy the process of making things. Hobbies are not really about the results. They're about the activity itself. Results are just a happy side-effect.
I'm a programmer. if, then, else, follows a logical progression through a problem. And problems can only be expressed in logical terms. We're trading logic and mathematics for emotion. That's all AI is. It's trying to make emotion do practical things. Emotion never does practical things.
LLMs are a good building block when your algorithms need to deal with some kind of language input that is inherently imprecise (e.g. natural language text documents, human input like search fields, spoken language). Or for generating output of the same kind.
They *might* even be useful as a tool for understanding/explaining what code does in order to learn, with some caveats. Or as a building block in search, see above.
They are, however, not a good tool for expressing or designing the algorithms themselves, or as a tool to avoid thinking about them.
Sadly, the latter is mostly what people use it for.
Absolutely! If you are the kind of person that spends their days in vim or emacs or similar, you don't want to write verbose explenations and hope something vaguely similar to your intentions happens. The problem is: to the average person computers were always like this. They've always been slightly missbehaving magic boxes. They never had the ability to express their intention precisely and efficiently and then the computer did exactly what they wanted in a transparent and understandable way. For them nothing is lost and our abilities have even less value, because now they can just scream at their computer, like they always did and it actually does... something.
I understand your point of view. But to me, AI isn't about replacing precision—it's about reducing the time spent on repetitive tasks. When I need absolute accuracy, I use precise instructions. When I need speed for brainstorming or drafting, I use natural language. Every tool has its place.
I hate it too but it is good to challenge ourselves sometimes.
The sloppier language is one that most ordinary non-technical people can use. Unlike every single programming language I can think of.
The 50 year failure to develop a ui or language to get computers (or databases or search engines or information repositories) to do what people want is the gap that natural language interfaces to (increasingly) general purpose capabilities that are being filled by "AI".
And it is the specialist community's failure to communicate to normal people the risks of an inaccurate system like "AI".
I've also been developing for ~25 years, and using AI has made me progress.
I spend less time writing code and more time thinking about and writing specifications that are then implemented.
Previous knowledge is still useful; I read the code and I LEARN. Because legacy company code isn't always clean, and I haven't always applied best practices.
Totally agree, but I would also add that using the skills I have learned over 40 years, learning all of the problems that can occur, and how to make sure that what is asked for is unequivocal and clear, so that what I do to make it happen is what was wanted - that is what I enjoy.
The fun part of my work is coding, writing code that does what someone needs it to do. Understanding that and converting it into code that is clean and neat and as good as I can. I love that.
If all I can do is fix bugs, listen to people telling me what they want - I am not interested. That is the boring part of the job.
Ohlalá! I really like Paris a lot, many consider it overrated and overwhelming, but I always go there with another perspective, and I tend to enjoy my time in there. This time I was wanting to do many paintings, but the weather was not contributing a lot, first days there was still the heat wave, and now without any beach where to splash and refresh, it was knocking off very easily, and then it was the final of Paris vs Arsenal so streets went crazy, and after the crazyness went some rain. At least the golden hour after the rain (don't mix and twist that sentence), the streets offered heartwarming scenes that looked very nostalgic.
And here we see Kazan, nomming a big baguette for himself.
---
A new reminder you can see all the entire watercolor journey on HD: Scans and full resolution photos. Available on Patreon and Ko-Fi, for monthly supporters. Also that helps me to keep doing projects like this, I appreciate a lot every contribution, thank you! Patreon: https://www.patreon.com/c/pandapaco Ko-Fi: https://ko-fi.com/panda_paco
Get more from Paco Panda on Patreon. Join for free or become a member for just $1/month. Hello, fur friends! It's Paco the Panda here.I'm a mexican furry…
Did you know that your writing is already better than anything AI can generate? Despite what you may think of yourself as a writer? It's true! In this video,...
Saja Boys - Soda Pop from the Netflix film "KPop Demon Hunters" https://youtu.be/983bBbJx0Mk Performed by Saja Boys Lyrics & Composition : Vince, Kush, Danny...
Watercolor Journey: Flurrabell and Helios in Milan (11/20)
If I thought the Galleria Vittorio Emanuele was complicated, oh boy, I hate myself enough to have chosen the Duomo di Milano as background. When I knew I wanted to paint that cathedral, intentionally I chose to only paint a small part of it, and I was better looking for some streets where lighting could be more interesting than just drawing a postcard of the iconic view. And when I said about simplifying and adding important elements of what I was seeing, of course a trash bin was very important (nothing to do with my hidden raccoon instinct)
Flurrabell and Helios walking on streets of Milano, Italy
----
Check the whole watercolor journey, already available on Patreon and Ko-Fi for monthly supporters, also you help me a lot to continue with this kind of projects (also I want to continue my comic, waaa!!!) Patreon: https://www.patreon.com/c/pandapaco Ko-Fi: https://ko-fi.com/panda_paco Thank you!
Get more from Paco Panda on Patreon. Join for free or become a member for just $1/month. Hello, fur friends! It's Paco the Panda here.I'm a mexican furry…
Watercolor Journey: Flurrabell and Helios in Milan (11/20)
If I thought the Galleria Vittorio Emanuele was complicated, oh boy, I hate myself enough to have chosen the Duomo di Milano as background. When I knew I wanted to paint that cathedral, intentionally I chose to only paint a small part of it, and I was better looking for some streets where lighting could be more interesting than just drawing a postcard of the iconic view. And when I said about simplifying and adding important elements of what I was seeing, of course a trash bin was very important (nothing to do with my hidden raccoon instinct)
Flurrabell and Helios walking on streets of Milano, Italy
----
Check the whole watercolor journey, already available on Patreon and Ko-Fi for monthly supporters, also you help me a lot to continue with this kind of projects (also I want to continue my comic, waaa!!!) Patreon: https://www.patreon.com/c/pandapaco Ko-Fi: https://ko-fi.com/panda_paco Thank you!
Get more from Paco Panda on Patreon. Join for free or become a member for just $1/month. Hello, fur friends! It's Paco the Panda here.I'm a mexican furry…
Had the pleasure of spending a weekend with friends and celebrating @JD_Puppy 's birthday recently. Really enjoying my time 1-1 with folks and with furry groups like 50-150 people, reminds me of cons when I first got into the fandom :)
Also working out some bugs in my camera, may be time for a new one, but still trying to grab those time lapses :) Also loving the first track in this, probably my favorite kind of DnB.
***WATCH IN 4K***10 hour round-trip drive from Vancouver to Cle Elum, Washington. Compressed into 6.5 minutes.Route: I-5, WA18, I-90, WA970Stops removed (som...
A second day of costumed shenanigans- this time of the #Khajiit variety. Athra made his first appearance at the Loxwood Joust today, and was well received! Will be back in a couple of weeks time (and hopefully will be able to get more photos than just selfies)
did you know you can just say the things you like about the artwork to the artist that you're looking at and they will actually appreciate it and be nice if you're not fucking weird about it
I /only/ get weird about the paws if I know the artist and/or know they're okay with it, otherwise if someone links something to me like "hey owa check this out" with a direct tag I politely am like "Wow, this is good art " or something like that
Took my #Jedi griffin, Qesnel, out to the #FestivalOfTheForce at the Kent Events Centre today. Smallish event (perhaps? Not sure on event sizes in the UK yet), but was good to bring him out. Met and talked with a couple of people; Tim Rose and Ailsa Berk.
LLMs and AI are a terrible "solution" to the fact that we just don't have time to do the things we're asked to do. we are overworked, across the board. all of us are.
i realize that the reason i see some utility in LLMs is because i was able to do some things here locally that i didn't have time to do because of work and other life stuff happening. otherwise, if i had had the proper time and space to do it, i would have gotten it done and learned something. but instead, i've fixed the issue but still don't know how or why it broke. the tradeoffs are obvious.
even if the solution is presented to me verbatim with an explanation, the time required for me to internalize that information isn't there. i can technically look up the logs to find whatever the LLM generated to get to that point, but it is not the same as you doing it yourself and retaining that information in your head. you are just a passerby while the thing does the work.
things are deliberately getting harder too; I was shocked and appalled at gmail not allowing folders and proper organization, training you not to be organized because you can just search for it, then they control the keys for search
feels like the society is making things deliberately harder on purpose so they can sell the tools to clean up the mess they trained you to helplessly make
Watercolor Journey: Mica and porter in Milan (10/20)
Now moving on from Cinque Terre to Milan, instead of nature I was facing a lot of architecture, very impressive buildings and such an adventure I got to live there due a train strike that I told on my journal.
Here we are in the Galleria Vittorio Emanuele II, overpriced coffee, and then me thinking why did I do that to myself at picking such a complicated spot to paint, but honestly it made me work a lot on simplifying structures on buildings, and after getting all the shapes figure out, then I can start adding all details.
We can see Mica and Porter shopping, or better said... Mica shopping and Porter carrying everything out.
---
This watercolor journey was possible due commissions and also by supporters on Patreon and Ko-Fi, you can join, support and getting all the complete journey available there from 1 to 20. Check it out! Patreon: https://www.patreon.com/c/pandapaco Ko-Fi: https://ko-fi.com/panda_paco Thank you!
Get more from Paco Panda on Patreon. Join for free or become a member for just $1/month. Hello, fur friends! It's Paco the Panda here.I'm a mexican furry…
SHEMAN - Count Binface (Official Video) | Ebeneezer Goode Parody by Mr.C of The Shamen . SHEMAN is Mr C of The Shamen. "Count Binface" is a reworking of the ...
Watercolor Journey: Denali and Archer in Riomaggiore (9/20)
My second day in Riomaggiore and this time I got ready for a splash. While swimming I was really looking forward to seeing a sea monster. The vibes from the movie Luca were too obvious that they got inspired on the place I was at that moment. The most challenging about trying to make this painting, more than not getting splashed by other swimmers, was to find some shade, the sun was too aggressive at that moment, I changed my spot so many times that I was worrying of changing too much the angle of my view, but for painting I had to leave it for later, I didn't want to get roasted.
This was personally one of my favorite paintings from all Cinque Terre, I was not intimidated for watercolors anymore, not even for drawing and painting buildings, I had learned I didn't need to copy all single details but to choose what was important to include and then simplify scenes. I could notice very proudly how I was improving.
On this picture, we see Denali and Archer, playing while cooling off from such a hot day. There was a lot of life like this in this place, it was so lovely!
-----
If you are liking my watercolor journey, and you want to watch all the paintings already, they are available now, some of them scanned (I couldn't scan them all), but all of them in high resolution, exclusive for Patreon and Ko-Fi monthly supporters. If you want to support me, you are also helping me to keep doing projects like this. I would appreciate it a lot! Patreon: https://www.patreon.com/c/pandapaco Ko-Fi: https://ko-fi.com/panda_paco Thank you!
Get more from Paco Panda on Patreon. Join for free or become a member for just $1/month. Hello, fur friends! It's Paco the Panda here.I'm a mexican furry…
