> So: is this just something wacky with my algorithm?
No, it's not. Once Meta identifies you as male, you will get almost exclusively thirst trap posts no matter what you do. It started about two years ago.
Some other interesting points: A woman posted on reddit recently saying she noticed her son's feed was filled with this stuff, so she created her own instagram account, identified as a man, and had the same feed. No matter what she did she couldn't fix it. She asked other women about this, and they all said their partner's feeds were the same.
This is not a problem for women. At least not one I've ever talked to or read about on the internet.
Another point: I tried very hard to fix this at one point. I went through instagram and hit like on nothing but pottery and parenting videos. For about a week I had a feed that looked like my wife's -- pottery and parenting. And then it reverted.
I got a whole bunch of thirst traps again.
It doesn't bother me anymore, I just tune it out and scroll past it because my feed still has the parenting and pottery too, and my friend's updates, which is what I'm there for.
But it would be good for more people to learn about this so they don't get angry when they see their male-identified partners/friends feeds.
It's odd to celebrate having the key sanctions unwound.
Before this ruling:
1. Apple were prohibited from charging any fee for external/referral purchases. Now this is once again allowed and the district court will work with all parties to develop a reasonable commission.
2. External links were permitted to dominate over IAP options. Now they must have equal size, prominence and quantity.
3. Apple were prevented from showing any kind of exit screen, that is now restored (but it can't be a scare screen).
4. Apple were barred from preventing certain developers/app classes from using external links (such as those enrolled in the News or Video Partner Programs) those are now reversed and Apple can once again prevent them.
Epic/Tim Sweeney are trying to spin these recent losses as a win. It's the old marketing playbook of hoping no one reads the fine print.
here’s a basic impl that i trained on tinystories to decent effect: https://gist.github.com/nikki93/f7eae83095f30374d7a3006fd5af... (i used claude code a lot to help with the above bc a new field for me) (i did this with C and mlx before but ultimately gave into the python lol)
but overall it boils down to:
- tokenize the text
- embed tokens (map each to a vector) with a simple NN
- apply positional info so each token also encodes where it is
- do the attention. this bit is key and also very interesting to me. there are three neural networks: Q, K, V – that are applied to each token. you then generate a new sequence of embeddings where each position has the Vs of all tokens added up weighted by the Q of that position dot’d with the K of the other position. the new embeddings are /added/ to the previous layer (adding like this is called ‘residual’)
- also do another NN pass without attention, again adding the output (residual)
there’s actually multiple ‘heads’ each with a different Q, K, V – their outputs are added together before that second NN pass
there’s some normalization at each stage to keep the numbers reasonable and from blowing up
you repeat the attention + forward blocks many times, then the last embedding in the final layer output is what you can sample based on
i was surprised by how quickly this just starts to generate coherent grammar etc. having the training loop also do a generation step to show example output at each stage of training was helpful to see how the output qualitatively improves over time, and it’s kind of cool to “watch” it learn.
this doesn’t cover MoE, sparse vs dense attention and also the whole thing about RL on top of these (whether for human feedback or for doing “search with backtracking and sparse reward”) – i haven’t coded those up yet just kinda read about them…
now the thing is – this is a setup for it to learn some processes spread among the weights that do what it does – but what those processes are seems still very unknown. “mechanistic interpretability” is the space that’s meant to work on that, been looking into it lately.
I'm guessing because this is basically an AI for Dummies overview, while half of HN is deep in the weeds with AI already. Nothing wrong with the talk! Except his focus on "do everything" agents already feels a bit stale as the move seems to be going in the direction of limited agents with a much stronger focus on orchestration of tools and context.
Here are the achievements of the Wrights with the 1903 Flyer:
1. First 3-axis flight controls
2. First propellor theory that was twice the efficiency of other airscrews
3. First aircraft engine that had twice the power/weight of other engines
4. First design that used a wind tunnel to get an efficient wing shape
5. First directed research and development program to identify the problems and solve them one by one, with the results culminating in the 1903 Flyer
6. Properly documented everything with photographs, notebooks and witnesses
7. The Flyer is hanging in a museum today, and exacting replicas have been built and flown exhibiting the same documented flight characteristics as the Flyer.
If you look at other contenders, they were all lacking these points. For example, with the Wright propellor, engine, and airfoil their craft had an enormous advantage over other designs that were trial and error.
All modern aircraft can trace their lineage back to the 1903 Flyer, and no other claimant. The others were all developmental dead ends.
P.S. About the catapult thing - are airplanes launched from aircraft carriers not airplanes? Besides, the 1903 Flyer did not use a catapult.
Usually we keep the mainstream news out of here but felt I had to post this one because I've never seen anything like it and was curious to hear other people's perspective here.
To me it feels like a setup to look good in front of their audience without any proper desire to engage in true diplomacy.
Self-congratulatory, self-righteous prigs are all over the place within human society.
When people complain about them, the substantive content of their complaint is the context in which they issue it. For example pg is complaining about the prigs who nag everyone about transgender acceptance, but not the prigs who nag everyone to reject and abuse transgender people.
Matters of speech, manners, and decorum are convenient ways to launder the advocacy of a certain set of values. All you have to do is accuse your enemies of violation when they advocate, and stay silent when your allies apply the same tactics.
In order to consistently navigate politics, one needs to start with one's own values. That's why I posted my comment above. The core issue for me is whether transgender people can show up in their preferred gender. Not whether other people are annoying jerks when they talk about that question. There are plenty of annoying jerks on both sides of any value question, if one has the open eyes to see them.
I have a hypothesis about that, based on prior battles with depression. A characteristic of many people who struggle with depression is revisiting things over and over in your mind, looking for what is often a non-existent solution. This is a very, very useful trait in a software engineer. It helps us think of solutions to hard problems, to see options that are not initially obvious.
But it also enables mental illness when you encounter problems that cannot be solved that way -- for me, it was divorce. But life is full of intractable problems.
People like you are the reason I personally don't want to be in an office.
> asking quick questions and getting an instant answer. vs chat where I may not get an answer for hours.
Basically like having a toddler around. Sorry for the negative take here but how it's written it feels like you don't necessarily realize that some people do need deep work, and prioritize your quick satisfaction to other's focus. Or I might be projecting, but I simply can't stand incessantly being asked questions, I view it as a type of interruption.
> running ideas by others. This doesn't happen on chat for me, and VC is too scheduled.
Yes, because who needs order in their time when somebody just wants to ask the group from their opinion, without even bothering to write it down. Also, whoever is sick that day and missing from the office, that's on them, whatever discussion happened, no evidence of it, so sucks to be them I guess.
> I don't feel connected to me team at all working from home
As a European who has been living in the US for a decade, yeah, you're pretty spot on. Americans are a scared people, probably the most scared I've ever seen. Afraid of the gov't, the neighbors, and random people they don't even know. I've gotten a ton of hate as a foreigner and I am not surprised a hateful, greedy and selfish population like here is afraid somebody will take them out with an improvised device.
> When the Drink button was pressed it made an instant but highly detailed examination of the subject's taste buds, a spectroscopic analysis of the subject's metabolism and then sent tiny experimental signals down the neural pathways to the taste centers of the subject's brain to see what was likely to go down well. However, no one knew quite why it did this because it invariably delivered a cupful of liquid that was almost, but not quite, entirely unlike tea.
Apparently one of the things that happens when a material transitions into a superconducting state is it stretches (by an infinitesimal amount) in the conducting direction. This stretching is not like usual stretching where there’s a change in the actual structure of the material. Rather, it is the bonds between each atom that stretch - pulling them into a slightly different state, one that happens to let electrons flow completely freely (ie superconducting). This “infinitesimal stretching” is usually caused by magnetic alignment, which is induced by super cold temperatures.
But there are a few odd materials, like iron selenide, that don’t seem to show any kind of magnetic alignment despite being able to become superconducting. That’s where the absolute geniuses behind this paper come in. They took a thread of iron selenide and stuck it to a strip of titanium, then physically stretched the titanium. This induced that “infinitesimal stretching” in the iron selenide sample. By examining the sample with X-rays while artificially inducing the stretch, they could detect the mechanism that would usually be the cause of the stretch. Essentially it is sort of like manually spinning the wheels on a car and watching the engine cycle, in order to understand how vehicles work.
Very clever stuff, and it seems like a real advance in understanding superconductivity. The article doesn’t really go into it, but this does suggest there might be a third axis (“super precise tension”), alongside the usual two of “super low temperature” and “super high pressure”, that we can search along to find new superconductors.
Yes, they are called "Insecure overachievers". They are the bread and butter of investment banks, management consulting firms, big law firms, etc.
They are also the reason that partnerships still thrive. A couple of years ago I was talking with this (big law) partner at a dinner, and we came in on the topic around partnership - and he just said it straight: The competition / rat race towards partnership is 100% skewed to the firms advantage. They get dozens of highly motivated, highly competent professionals competing against each others to prove their worth. They'll be on call 24/7/365 to show that they're worthy of partnership. Even if they become partners, most won't be rainmakers - and many would have made the same amount of money if they struck out on their own, opening a small boutique firm.
I'm quoted in this blog post so I figured I'd respond. I'm a former member of the React team but I haven't worked on it in a long time.
Largely I agree with everything in this article on a factual basis but I disagree on the framing of the trade-offs. Two points in particular:
1. Before open sourcing React we extensively measured performance on real world applications like mobile search and desktop ads management flows. While there are pathological cases where you can notice the overhead of the virtual DOM diff it's pretty rare that it meaningfully affects the user experience in practice, especially when compared to the overhead of downloading static resources like JS and rendering antipatterns like layout thrash. And in the situations where it is noticeable there are escape hatches to fix it (as mentioned in the article).
2. I disagree with the last sentence strongly. I would argue that Svelte makes huge sacrifices in expressive power, tooling and predictability in order to gain performance that is often not noticeable or is easily achieved with React's memoization features. React was always about enabling front-end engineers to take advantage of software engineering best practices so they could level up velocity and quality. I think Svelte's use of a constrained custom DSL is a big step backwards in that respect so while I appreciate the engineering behind it, it's not a technology I am interested in using.
Even though I disagree on these points I think it is a well-written article and is a pretty fair criticism of React outside of those two points, and I could see reasonable people disagreeing on these trade-offs.
I practiced Zen Buddhism for many years and left my sangha due to many of the points that Sasha brought up:
Many of the teachers and students I knew were not rising above their neuroses. Many of them were masking their life problems with the Buddhist aesthetic as opposed to really working with them. We would cycle through the same concerns repeatedly without any progress. I started to figure out that the process was to drop the issue and disengage with it. The problem is that does not work outside of a sheltered monastic community because you need to face your problems constantly in the real world.
I agree that many modern Buddhist schools stray away from the Buddha’s original teachings (as we know it from the Pali Cannon). Many branches won’t even really teach what Buddha said; only interpretations from later traditions. We did not talk about Buddha much in Zen practice at all. Much more time was given to Dōgen, and the Chinese masters than to Buddha. In Zen everyone is a Buddha so Siddhartha Gautama (O.G. Buddha) gets marginalized. There is also a pantheon of Buddhas which dilutes things even more. Buddhism has a pretty straight forward thesis (Four Noble Truths), but it has become esoteric after centuries of appropriation and reinterpretation.
Modern Western Buddhism pushes meditation above all of the other practices. We spent more time meditating than anything else, which was different than how the early Buddhists and even how most Buddhists in Asia practice. This leads to people thinking that all they need to do is sit and not change anything about their lives and it will magically work out. In fact in Japanese Zen Dōgen essentially states that sitting with the correct posture (zazen) is enlightened practice itself. This enlightenment is transitory, so one could imagine that the longer you sit zazen the more time you get to stay in this enlightened state. You can see how this could become an obsession. This in practice leads to a lack of engagement which would have you thinking you are actually putting in the work, but you are just eschewing reality.
Buddhism has a rich tradition of debating and challenging teachers. In fact the Pali Cannon is full of these debates. However, these days if you bring up a question or objection to some teachers they don’t really engage with you. In Zen you can cover up inconsistencies with esoteric vocabulary and wave it away. Just sit and it will be okay.
Buddha in the Pali Cannon was actually more human than we give him credit for. He made mistakes and learned from them (even after nirvana). He got old and died. He scolded his monks for breaking monastic rules. The Buddha represented in the Pali Cannon can be raw at times which goes against the ideal Buddha archetype.
Imagine you're a municipality like San Francisco. You control all of the zoning regulations, and many of the permitting requirements. You have a AA+ credit rating, and even in today's higher rate environment, you can borrow billions of dollars at 6%.
You already own oodles of land. You can build to whatever height you want within reason, because you control the zoning, and you even control many of the ordinances that allow citizens to block development (though certainly not all, like CEQA and NEPA). So you have lots of lands to build on, and what you build is largely in your control.
Construction costs in San Francisco are sky-high, at $440 sq ft. But people will happily pay you $40/sq foot per year for housing, probably for 75 years.
How is this not the easiest decision in the world?
Create a housing development agency, become a permanent developer, and landlord. Never stop building. Put proper incentives in place, so that employees at the agency can partake in the profits, incentivizing them to be efficient. Never stop doing this.
You might not be great at this at first, but fifty years later you will be.
This kind of criticism gets tiresome. Someone has to own these companies and pay the journalists. If they're private, then they're "owned" and suspect. If they're publicly traded, then they're beholden to "Wall Street". If they're publicly funded, they're "propaganda".
Please criticize the work, not the incentives[1]. In this case it seems like a pretty darn good piece of public interest journalism. Do you disagree?
[1] Example: Elon Musk banning journalists out of personal pique is a terrible thing, not because "Billionaires shouldn't own Twitter" (even though many people believe that to be true) but because censorship, especially censorship of inconvenient journalism, is evil.
Parent of three ranging from tween to college age.
We have never had limits on anything. Screen time/content unrestricted. Probably dozens of times, maybe hundreds, we have had "the talk" about how there's a lot of weird shit in the world, internet included, and it's better to know about that stuff and how to deal with it than to create a temporary secure enclave at home where it doesn't exist.
We especially didn't want to pawn off our responsibility as parents onto their future 18-year-old college-bound selves to learn how to deal with excesses. Since the moment they could point, they've been given the opportunity to make mistakes with excess. They've fallen thousands of times, we've picked them up thousands of times, and now they're pretty good at not falling.
You can guess that I love CI/CD, stable trunk, etc., in my day job. My attitude as an engineer/parent is you'll never make bugs/threats go away. So make sure you and your team/family are experts at dealing with them while they're fresh and small.
I have ADD. Diagnosed medicated 30 years ago as a child. Slack distraction isn't my specific achilles heel, but I think I get it.
What works for me is breaking up workdays into 3-4 parts. I physically move, rearrange my desk, restart my laptop, take breaks, close files and such to punctuate day parts and keep them distinct. I need a lot more than a timer to make this work.
Once I have the day split into distinct parts, I can deal with 2hr blocks at a time, rather than a whole workday. Whatever your specific difficulties are, I find they're more tractable within a 2hr block. Maybe you allocate 30m to slack and then close it.
I tend to "lay out my tools." Open only the files and apps that I need for the current block, print out my notes, stuff like that. Imagine laying out tools and parts before starting to assemble a piece of IKEA furniture. Then you put that stuff away, grab some other stuff, and start to work on fixing the sink.
The details are emergent. At least for me, this method is useful in the abstract. Whether the issue is slack, procrastination, big getting in the way of small or the reverse... I seem to find solutions much more readily within these smaller blocks of time.
I don't know if what works for me will work for you, but if it does... the way it works is that solutions to problems like your slack problem are easier to find when you just need to solve them for the next couple of hours.
I find I spend a lot of mental effort trying for big solutions to big problems, and besides medication, very few big solutions stick. Small solutions otoh.. seem to emerge once the framing is correct.
Also, not everyone is cut out for highly collaborative work. I'm not. I need decent size blocks of work that I can take and own. I'm just not reliable enough to have others depend on my work on a "by lunchtime" schedule. That doesn't mean I can't collaborate. I have managed teams and worked in people roles. I just can't work and have a conversation simultaneously. Instead, I will allocate a whole 2 hr block of time to pair programming-like modes.
If your solutions are good and you have confidence in them, workplaces tend to be flexible enough to contain them.
3 years is deceptive - in an exponentially growing company, the majority of employees are fresh, pushing the average tenure low. The relevant number is "length of tenure at time of severance", which I believe skews much higher. Any company seeing 30% churn is having a bad, bad time.
EDIT 2: After playing around with this repo, I've found:
- It offers better UX for interacting with Stable Diffusion, and seems to be a promising project.
- Running txt2img.py from lstein's repo seems to run about 30% faster than OP's. Not sure if that's a coincidence, or if they've included extra optimisations.
- I couldn't get the web UI to work. It kept throwing the "leaked semaphor objects" error someone else reported (even when rendering at 64x64).
- Sometimes it rendered images just as a black canvas, other times it worked. This is apparently a known issue and a fix is being tested.
I've reached the limits of my knowledge on this, but will following closely as new PRs are merged in over the coming days. Exciting!
3D audio on the PC was deliberately killed by Creative.
They sued Aureal into bankruptcy, bought it in the court auction, and the day the sale closed they nuked the support website and took the drivers offline.
They used similar scummy tactics to decapitate any other competitors. Then they considered their reverb based spatial audio solution sufficient, and promptly sat on their heels doing zero innovation while collecting a rent.
And them as chip technology improved, a basic "Soundblaster 64" chip became so cheap that motherboard manufactures started bundling it in as a selling point (which made a ton of sense for non gaming PC users btw). Additionally MS stepped in and provided some software spatial functionality within DirectX, as processors had improved to the point where dedicated hardware for it wasn't necessary.
Back then I worked in gamedev, and I briefly considered going into competition with MILES et all with a 3D audio library after the Aureal fiasco, after I stumbled on some interesting papers doing Fresnel Zone Tracing variations as low overhead spatial audio, but ultimately wasn't serious about it vs other options at the time.
It's not risk aversion. Hacker News doesn't avoid intelligent discussion about politics, rather this community has proven time and time again that it isn't even capable of it, especially where Trump is concerned, and people are pre-emptively flagging what's likely to be a very high noise, low signal thread.
This is a community whose education on computer science comes from MIT and Stanford, but whose education on politics comes from 4chan and the Unabomber manifesto.
I mean, I would love to see intelligent political discussion on Hacker News, politics is one of the great intellectual pursuits. But it just isn't going to happen here.
I'd seen a decent modular 3d printed tower for growing things like lettuce and it turns out Hoocho is using a newer and more advanced style of the same idea:
If fritzing gets #1 spot on HN I would like to wholeheartedly recommend https://www.kicad.org/ as well.
IMHO much saner workflow and good UX in the latest version.I did a simple PCB and got it manufactured on https://jlcpcb.com/
for a couple bucks. Quite a rewarding experience.
I would also suggest that anyone who cares about computers gets to know basic electronics. How do transistors work, what is a bus (e.g I2C or SPI) and how is it all connected? (Drumroll ... a PCB of course). There is a ton of tutorials on youtube that do it end to end. E.g Phils Lab https://www.youtube.com/watch?v=rLDqQ2L_mUQ for a PCB design. Or Ben Eaters excellent beginner tutorial "Hello World from Scratch" https://www.youtube.com/watch?v=LnzuMJLZRdU
I don't believe there is a shortcut for "hacking" languages. I'm a diplomat, and currently learning my fourth language to the C1+ level.
When we learn languages, it's a full time job. It was 9 months to learn Mandarin to a B1, 6 months to a C1+ in Spanish, and I'm currently at a B1 in Estonian after 5 months.
There are several things I think are crucial after years of full time study (note: this assumes you're going for professional fluency, not just touring around the country where interactions are largely scripted and predictable):
* There is no substitute for production - you must speak the language with a native speaker (not an app) and talk about topics that are relevant to the kind of scenarios you anticipate. We spend the first several months discussing current events in target language - at first scripted, then later free form. This builds vocabulary and helps fluency. This is quickly expanded to discussing current events in depth and participating in mock debates.
* Give mini presentations - target 3-5 minutes of talking about a relevant topic with little prep time. The difference between intermediate and advanced is the ability to move from discussing only facts to making a coherent argument. Native speakers will often not be able to follow your train of thought without learning to connect cause and effect using structures appropriate for your language.
* Interview native speakers - prepare 2-3 questions about a particular topic and check your comprehension by translating their answers to English. This obviously helps build your comprehension, but also helps to learn to "automate" comprehension while you are thinking about something else. If you can take notes in English while a native speaker is talking at normal speed (and achieving 90%+ accuracy), it will make it easier for you to participate in normal speed conversations.
* Read target language news - this is critical for expanding vocabulary and learning colocations - knowing what verbs are used in particular contexts (e.g., do they say "I talked with X" or "I talked to X". Do they say country X shot, launched, or threw a rocket?)
Bottom line - language learning is not just about the number of hours you put in. The quality and type of practice you do matters a lot. You aren't going to be fluent via Duolingo alone. You need to put in the time using structured practice with native speakers to really learn anything.
Until you've learned to notice automated judgment and unlearned automatically running with it, it'll take more analysis to answer the question.
You may have genuine needs being denied by the activities. You may have unacknowledged trauma patterns matching with some aspect of the activities. You may have socially conditioned judgments moving too fast for you to currently notice.
Do you want a simple practice in noticing and unlearning such judgments?
I used to read books all the time. I spent most of my waking hours as a teenager reading books. I’m 35 now and can barely hold attention on anything, I read maybe 3 books a year if I’m lucky. Even on Hacker News I almost exclusively read and comment on the comments, not the actual article unless I see someone says “have you read the article?”
I don’t really know why. I have 8+ hours a day screentime on my iPhone, which isn’t even including my Mac and work laptop which I might browse HN on. I feel sort of lobotomized but most people I know are still way dumber, I’m muddling through my career and not getting negative attention. I’m not even busy. I’m just lost and exist mostly moment to moment, and am lucky enough to have developed enough skill to write code before all this happened to me.
> (Ironically, a check that was bad because of the fee taken for 'a')
This is a real big problem in these kinds of situations.
To give you an idea of how this happens: when I was in college, I worked at a pizza shop. It was Thursday. I got paid Friday. I thought I had $20 in my bank account, but I didn't realize that day was the day that my World of Warcraft subscription was up, and that was $15. During the course of the day, I bought lunch, a snack, and some pencils or something. All that total was less than the $20, it was something like $10 for lunch, $2 for the snack, and $5 for the pencils. So I was good.
I wans't good. They applied the charges like this:
Still an expensive mistake on my part, but it's more subtle than 'made a bad choice.' I was able to convince the bank to refund two of the fees so I was only out $40, but that extra $70 was a _massive_ amount of money to me at the time. It basically meant that I could only exactly squeak by for the next month, since that was basically all of my discretionary income. You can construct a similar situation which would have forced me to then take out a payday loan to make my bills, had I made one or two more mistakes in that same timeframe.
No, it's not. Once Meta identifies you as male, you will get almost exclusively thirst trap posts no matter what you do. It started about two years ago.
Some other interesting points: A woman posted on reddit recently saying she noticed her son's feed was filled with this stuff, so she created her own instagram account, identified as a man, and had the same feed. No matter what she did she couldn't fix it. She asked other women about this, and they all said their partner's feeds were the same.
This is not a problem for women. At least not one I've ever talked to or read about on the internet.
Another point: I tried very hard to fix this at one point. I went through instagram and hit like on nothing but pottery and parenting videos. For about a week I had a feed that looked like my wife's -- pottery and parenting. And then it reverted.
I got a whole bunch of thirst traps again.
It doesn't bother me anymore, I just tune it out and scroll past it because my feed still has the parenting and pottery too, and my friend's updates, which is what I'm there for.
But it would be good for more people to learn about this so they don't get angry when they see their male-identified partners/friends feeds.