Hacker Newsnew | past | comments | ask | show | jobs | submit | quibono's commentslogin

If I'm not mistaken there's also the author of Starlette / FastAPI

Well there's Make a Lisp [0]. And Lispy [1] and Lispy2 and Lisp.py [3].

[0] https://github.com/kanaka/mal [1] https://norvig.com/lispy.html [2] https://norvig.com/lispy2.html [3] https://khamidou.com/compilers/lisp.py


The only reason I have a dual-boot Windows setup now is for MSFS2024 and DCS. They will probably be the last to ever get full-Linux support (if ever).


DCS World and MSFS2020 play on Linux just fine. MSFS2020 is flawless; DCS World has a few visual glitches with smoke tessellation and MFD textures that need to be converted before working, but besides that it's great. I fly modded and first-party modules on multiple maps, hours at a time, zero complaints.


That’s interesting… have you tried MSFS2024? That’s my main driver for now; god it would be amazing if I could run it on Linux. I’ll definitely need to try DCS on Arch! By the way - what module would you recommend (apart from F-18)?


Haven't tried MSFS2024, but ProtonDB suggests that it works okay. The 2020 version is definitely set-and-forget once you log into your Microsoft account.

Module-wise - once you have the F/A-18, a lot of the other modules feel like wasted money. I can wholeheartedly recommend the AJS37 Viggen when it's on sale, for $30 it's got a lot of flexibility and surprisingly simple systems for a 1970s jet. It's very easy to memorize the cold start process in the Viggen too, which makes it a joy to get off the tarmac. The JF-17 is also a barrel of fun, and much easier to fly/fight in than the F-16 if you don't have a HOTAS and don't need JHMCS. Besides that, the only other modules I can fully recommend is the Syria map, Persian Gulf map, and Flaming Cliffs for all the modded planes it unlocks. Not personally a fan of the Sinai map right now, and the Apache is very difficult to fly without a copilot/gunner in the second seat. Black Shark 3 is fun, but one of the most goddamn complex modules in the game. Even with translated cockpit labels, you'll be permanently conjoined to a YouTube guide telling you how to do the most basic flight maneuvers.


It's been a while since I took a Coursera course but I LOVED it at the beginning. Between Machine Learning, the (numerical) optimisation courses and NAND-To-Tetris (even for the platform alone) it had so many great courses to pick from.


I did Andrew Ng's old Machine Learning, Obarsky's Scala course, the Ng's Deep Learning specialization, Nand to Tetris part 1 and a small Data Science course which wasn't very good. I think my very first course was "Model Thinking" course, but I never took the exam there.

I also tried the sequel to the Scala course at one point, and the Cryptography course, but I dropped out from those after finding out they were a bit too hard - I spent way more time on the coursework than I'd intended.

But I can't say I like the direction it's taken in recent years.


Odersky ;)

"Model Thinking" was great!

And I really liked the gamification course by Kevin Werbach (The topic was still hot back then) - something I used extensively at my start up.


Didn’t the gamification course have one of the relatively few well done peer assessments? The course was good, but it’s interesting now that gamification features completely turn me off now on any platform or program attempting to motivate me toward a specific end, regardless of whether that goal is in my interest or the interest of someone else trying to make money.


Yes the peer assessments really did work.

And yeah: Gamification became shallow real fast. Even the Gamification techniques in games! I think the reason is that everyone focused on adopting the easy part (checkboxes, achievements, levels) while skipping the real core (player types, intrinsic motivation)... But the course even warned about this level of shallow implementation.

(Btw: A few months later I enrolled in a course on educational psychology on coursera that was supposed to showcase some SOTA techniques... They canceled it because they could not work out the details. I think academia is often just not good at pulling things off.)


Whoops, Obarsky was the Amiga synth guy, yeah, I haven't taken any courses with him. Although I might consider it.


The model thinking course was interesting but it should have had a follow up that was much more than a freshman survey course treatment of each model.

Reading online it seems like most people got the impression that it was establishing that all models are essentially useless. Instead it was showing that each of these models were an extremely efficient way to understand some dynamic situations, but that it’s still absurd to focus on only one model when trying to understand the world.


Agreed about Odersky, the Scala course and the Scala Functional Programming course were solid (the latter a bit less so, a blemish was its insistence on Akka, but the concepts were interesting).

There was also a very interesting introduction to Programming Languages (by Dan... something? He was from the University of Washington I think) which covered multiple paradigms and had interesting things to say about the ML family.


Yes, Programming Languages by Dan Grossman! The course was later split up into three parts, with each part focusing on a particular language/paradigm: SML, Racket, and Ruby. Definitely one of the higher quality offerings on Coursera.


Yes, that's the one. I forgot that I discovered and played with Racket because of this course. Really cool language (and course).


The Odersky Scala course was pretty cool. It gave me a basic understanding of monoids and monads that has stuck with me. Robert Sedgewick's courses were very good. Learned a lot about graphs.


I'll have to look at the Scala course, thanks!


+1 for NAND-to-Tetris. I combined it with a visual logic simulator so I could actually see the structures beyond the VHDL. I would love to go back and do Part 2.


I have the book, but I was unaware of the course. Was the book necessary for the course? Also, do you think the book is sufficient without the course?


The course has all the materials you need. The book itself is sufficient, but I think the course adds a bit more detail to the explanations.

Any courses you would particularly recommend? I always found that Udemy's vast catalogue made it hard to actually pick a course.


Yes! I have had unfettered access to it via a couple employers and the Illusion of Choice is real. The best thing they could do (for users like me, not sure if this is true for the majority) would be to go back to being a curator of quality and not a marketplace for anyone to make a course.


Being a big marketplace is their big differentiator from sites like masterclass

But at the same time I agree that they aren't doing enough to surface the high quality courses


>But at the same time I agree that they aren't doing enough to surface the high > quality courses

They have forced the creators to agree on being scrapped by AI or otherwise not showing up on their own top search. Ironically this has sealed their fate, and most top creators decided to move their content somewhere else.


There was one course I did gor mongoose, muber I think it's called. I really liked it as a student because it's all very bite-sized and you could stop/start whenever. They do recaps at the beginning.

Compare that to a 6 hr video on YouTube, next day you already forgot what the timestamp was about.


I recommend anything by jonas schmedtmann for js/ts/react to work colleagues.


Am I to believe that those 700K+ downloads are organic traffic? Who's listening to all this stuff?


HN sends tens of thousands of views to AI-farmed articles about why AI is good or why AI is bad. These articles get upvoted to the front page literally every day. They don't say anything interesting, but many of us just like having our existing beliefs recited back to us.

So to answer your question, I think we all do, it's just that different audiences have different sets of topics for which they let their guard down.

There is a huge market for content that makes you feel smart without requiring thinking and makes you busy without requiring work. I'm not not saying it's inherently bad. I'm listening to music on my daily commute and it's the same thing: just enjoyable filler so that you can do something other than getting angry at other drivers. The internet just weaponized the formula, and now AI is the equivalent of nuclear weapons I guess.


How is a listen to a podcast counted?

If someone listens to a couple of minutes of a 30 minute slopfest and nopes away, is that counted as a listen?

Your example of HN sending views to shit is interesting, because I presume a lot of people sometimes click on a link expecting something insightful and is greeted by bullshit. A view is counted, but no meaningful interaction happened.


As I understand Spotify et al may do something a bit more sophisticated, but the traditional model for podcast analytics purely tracks downloads, which could very well be your client auto-downloading a subscribed episode you never play. I don't think anyone actually has visibility of "listens". And the traditional model for ad sales is a creator (or an agent on their behalf) emailing a brand "Hey, we make this podcast which gets X monthly downloads, want to buy an ad read?" I think they usually point to iTunes store rankings to somewhat support these claims but again, iTunes just tracks downloads. (Obviously, this is all rife for fraud.)


In some ways it doesn't really matter because once you've got enough data points you'll know what percentage of views result in an ad click (or whatever) and then you can figure out how many views you need to hit your revenue targets.


No, but to misinform people you have two main strategies: limiting through tailored scarcity and dilute in extra-generic overabundance. Don’t get it wrong: both can be combined and even can sometime overlap.

It doesn’t matter if no one is listening. Equally saturating all channels, metrics and indicator is enough to create hindrance so preventing relevant information to spread in meaningful time.

Attention is all you need, so distraction is all that will be given.


Also, fracturing audiences to infinity.


I listened to a podcast a while back (human authored I'm pretty sure) about low-quality gutter level streamer content and how popular it is, speaking of personalities like asmongold and a vast number of even worse imitators.

This content is made by humans but is pointless grindingly stupid filler spiced with a dash of obviously performative offensiveness. You're basically listening to a complete loser (or someone LARPing as one) telling you about their boogers and then being racist and then playing video games for 6 hours.

But it's wildly popular. Millions of people stream this kind of shit for hours every day.

There's a lot of people out there who just want to numb their brains, and there seems to be no floor. You can just keep making it dumber. The stuff people stream (and doom scroll) on the Internet makes 1980s daytime soaps look like high art from a lost golden age.

So it's not at all surprising that millions of people listen to low-quality un-curated AI slop podcasts.

I actually unsubbed from the podcast I heard. Meta discussion of crap like this isn't much better than the content itself. Keep driving. Do not look at the car accident.

I had kind of an epiphany like that in the last year. The Information Age means information is free. It costs $0 and is produced to infinity. That means you are not missing anything. Your attention is actually 100% yours, and if you choose to ignore the car wreck that's fine. There are infinity car wrecks. There are infinity everything. Keep driving.


The problem is I want to live in the "correct information age" - that qualifier is hard to find. I suspect that correct will cost money. Unfortunately I don't know how to pay for it. Many of the major publishers are also using AI with questionable fact checking. Where I most need correct information is my local small town news, and there isn't even a newspaper anymore. (there is the nearby big city newspaper, but they don't cover my local issues well)


The correct Information Age is the one you get when you let your prefrontal cortex play DJ and pick your media based on what will enrich your life or teach you something.

If you let your brain stem drive you’ll spend your life scrolling political rage bait and slop.

Whether the slop is made by humans or machines doesn’t much matter. I kind of think the AI thing is a red herring, though AI does make it possible to make a lot of slop. So maybe AI is the thing that forces the issue.


One of the real costs of the end game attention economy is that when your "car" crashes, noone is going to stop to help. When the market you engage in gets swallowed up, everyone will buy the swill that outcompetes you on perceived surface level value. Communities get fractured. Organizations that used to be community pillars (church) become self serving. All these things create a positive feedback loop of intellectual degradation.


"No one ever went broke underestimating the intelligence of the American people."

--H. L. Mencken (or at least attributed so.)


The vast majority of people tuning into those kinds of slop streams are not really active listeners. It's more akin to turning on the radio while you work/clean/perform some other task that doesn't require strict focus or attention, with the added benefit that you can personally interact with the streamer (chat) when you have attention to spare. But I'd wager most viewers never directly interact or even pay much active attention to the stream at all.

I wouldn't be surprised if the same dynamic is playing out with these AI slop podcasts.


By McHealy's logic, we ought not be concerned about that. After all, it's low-stakes content.


My podcast app downloads way more podcast episodes than I actually listen to.


I occasionally put on a (human-made) podcast for the word-sounds rather than the content. I can imagine others do the same without caring whether it is human-made.


If the sonic quality of a human voice is what you're seeking, then I imagine a generated voice will still be less appealing, no?


It's reminding me of twitter. I occasionally open it, half of the content I see is AI garbage (and by that I mean, poor quality AI generated stuff that is obviously AI), and 95% of the replies are bots responses, which aren't even AI based (most of the time not it's garbage, unrelated text)


They have 4k podcasts releasing 3k+ episodes per week, so that's ~250 listening per episode. That's an achievable number with organic traffic.

https://www.inceptionpoint.ai


Other bots?

Dead Internet Theory.

AI produced, AI downloaded. No humans in the loop.


There was one of those "memes" a few years ago that is just a screenshot of someone's Twitter post that was essentially:

"My wife is a teacher, she used AI to help create an assignment, all the kids used AI to complete it, and now she's using AI to grade it. Nobody learned anything, nobody really did anything. What's happening?"


Being the devil's advocate, it sounds like no one involved see any value in that exchange, therefore they don't care.

In that sense AI slop is a symptom, not a disease. But perhaps also a catalyst.

I really wonder if there is a sort of silver lining here, and in the long term low value activities will be filtered out of society. Though that borders on the AI maximalist view which I don't fully agree with.

Of course the glaring question is what value even is.


"silver lining"

agree. if internet is so filled with slop. people will move away from it, start to read books, walk, hang out with each other again?


not when all the world capital is pushed into that bubble while eventually eroding the freedom to do all these things

can't move away from internet when you can't earn money without it, and all the services require you to participate

can't read books if no one is going to be publishing those, after they get out-competed by cheap endless slop

no point in walking when cities are built for cars and businesses, and public spaces continue to dwindle and be defunded

can't hang out when you're too tired trying to survive


There are absolutely people "trying to survive" but nobody reading HN falls into that category.


Or maybe Ms. McHealy was simply lying.


I know what you're getting at but those examples are reaching


Just waiting for Gemma 4, DeepSeek 4 now. Then the only thing I'll be able to complain about is the completely different API to interface with (unless they FINALLY move to full OpenAI support).


Amazon Bedrock Mantle provides OpenAI compatible API endpoints for model inference, powered by Mantle, a distributed inference engine for large-scale machine learning model serving. These endpoints allow you to use familiar OpenAI SDKs and tools with Amazon Bedrock models, enabling you to migrate existing applications with minimal code changes—simply update your base URL and API key.

https://docs.aws.amazon.com/bedrock/latest/userguide/bedrock...


Just use litellm or any-llm if you want to convert to OpenAI format. They did used to have bespoke per model interfaces that were way worse


Nice, I’ll definitely check this out. You might want to look at optimising the PDF, it’s sitting at 40MB right now.


Thanks! I think most of the weight comes from the PNG diagrams, but I don’t actually know: I’ll put it on my to-do list to investigate, maybe there are some easy wins here.


The ~750 pngs are about 15 megabytes. The 77 photographic images are 22 megabytes.

Somewhat a death of 1000 cuts.

I used https://www.xpdfreader.com/pdfimages-man.html to extract them and take a look.


Thanks for that. At first glance, it looks like that is a good place to be, regarding images. I will still investigate when I find time.


Probably the main thing you could look at is what pixel density you want in your images.

For example, there's a 1 megabyte image of a tanker trailer that is displayed at about 1.5 x 3 inches, you could get rid of 3/4 of those pixels (going from ~400 ppi to ~200 ppi) and not really change the quality of the image for a casual reader.


that’s a really good place to start, thank you!


One week later, a follow-up. I went through the largest JPG files and was able to shave 7 MB off the final PDF file size, with no noticeable drop in quality, as far as I can tell: https://framagit.org/olivier/thermodynamique/-/commit/063038...

As for the PNG files, I gave up after spending a whole day on them. I was able to compress them losslessly to save 6 MB, but this did not carry over into the PDF, where they are not embedded but re-encoded. The best way to save space here would be to switch to SVG, but this is too much work and last time I tried 12 years ago it did not work well. Alternatively, removing transparency would help (and anyway the source SVG is given for each file). But that’s not very future-proof and somewhere I know there’s a teacher who just is going to right-click-save a diagram and will get a nice crisp transparent-background PNG they can instantly reuse. Verdict: at this point, I am calling it good enough. :-)


Claude was able to optimize the hell out of one of mine, might be worth a try


Does it matter?


I love this -- I'll have to do something like that for my site. I always liked the big initials on the start of a paragraph. Though it feels a bit more prose-applicable than for non-fiction writing.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: