HackerX

By year

2025

Ideas that Hacker News readers were still recommending in 2025.

36 posts cited121 citations

Counts show how many picks Hacker News readers cited in each year, not when the post was written.

@mxcl

Career

@mxcl

View on X
Recommended by 42 HN users
From the Hacker News threadleetcode seems to agree with your definition [0]. the meme isn't to say that inverting a binary tree is particularly difficult - anyone familiar with coding challenges and trees could trivially produce a solution. the meme is more pointing out how ludicrous it is that senior/staff/principal interviews can hinge on these types of problems, despite the engineer proving their proficiency by doing something like running DOOM in typescript types or writing homebrew [1]. [0] leetcode.com/problems/invert-binary-tre…... [1] x.com/mxcl/status/608682016205344768@LiquixTypeScript types can run DOOM [video]2025
117

@naval

Business

@naval

View on X
Recommended by 7 HN users
From the Hacker News threadtwitter.com/naval/status/10021034977251… > Seek wealth, not money or status. Wealth is having assets that earn while you sleep. Money is how we transfer time and wealth. Status is your place in the social hierarchy. twitter.com/naval/status/10021036273878… > Ignore people playing status games. They gain status by attacking people playing wealth creation games. From @naval’s Twitter thread “how to get rich (without getting lucky)” twitter.com/naval/status/10021033606468… My note: wealth is options; options are freedom, and buying your time back. Only play status games if you’re playing for fun, but don’t confuse it with success and freedom.@toomuchtodoPursuing status will never bring true happiness2023
14

@levelsio

Startups

@levelsio

View on X
Recommended by 6 HN users
From the Hacker News threadships like a machine yes his products being genai wrappers, some are partly yes. in his interview with lex fridman he explains how he built his recent photo ai one, yes most of it is wrapper on some ai service, but his team goes the extra mile of training models on their own so it looks more photorealistic than the generally available ones he built products before ai like nomadlist and remoteok. his hit and miss rates are as real as it gets x.com/levelsio/status/14573152744665948…@inquisitor26234Does “building in public” work?2024
7

@jmmv

Hardware

@jmmv

View on X
Recommended by 3 HN users
From the Hacker News threadThere's a problem when people who aren't very sensitive to latency and try and track it, and that is that their perception of what "instant" actually means is wrong. For them, instant is like, one second. For someone who cares about latency, instant is less than 10 milliseconds, or whatever threshold makes the difference between input and result imperceptible. People have the same problem judging video game framerates because they don't compare them back to back very often (there are perceptual differences between framerates of 30, 60, 120, 300, and 500, at the minimum, even on displays incapable of refreshing at these higher speeds), but you'll often hear people say that 60 fps is "silky smooth," which is not true whatsoever lol. If you haven't compared high and low latency directly next to each other then there are good odds that you don't know what it looks like. There was a twitter video from awhile ago that did a good job showing it off that's one of the replies to the OP. It's here: x.com/jmmv/status/1671670996921896960 Sorry if I'm too presumptuous, however; you might be completely correct and instant is instant in your case.@KapKap66The world could run on older hardware if software optimization was a priority2025
9

@AuthorJMac

AI

@AuthorJMac

View on X
Recommended by 3 HN users
From the Hacker News threadI want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes. From: x.com/AuthorJMac/status/177367919763170… As a species, we need to evaluate our attitudes to work. The quote above resonates at some level, but maybe there are some people who enjoy doing dishes and laundry, who knows. Most people are just surviving. It is a constant battle between slaving away in jobs (and have healthcare tied to our jobs) we don't like and rest of our lives, including relationships, hobbies, even health. Most people do not have the time or energy to think about anything else other than just getting through the day.@akudhaIn Japan, the robot isn't coming for your job; it's filling the one nobody wants2026
4

@ID_AA_Carmack

Hardware

@ID_AA_Carmack

View on X
Recommended by 3 HN users
From the Hacker News threadFor all the people who are doubting that applications are slow and that it must just be me - here [0] is a debugger that someone has built from the ground up that compiles, launches, attaches a debugger and hits a breakpoint in the same length of time that visual studio displays the splash screen for. [0] x.com/ryanjfleury/status/17477562194047…@maccardThe world could run on older hardware if software optimization was a priority2025
4

@FFmpeg

Open Source

@FFmpeg

View on X
Recommended by 3 HN users
From the Hacker News threadMy apologies - you’re correct. I didn’t mean that as “you should never expect anyone to have contributed code for free/the pleasure/for the puzzle solving aspect”. I do it all of the time. I meant - it’s unfair to consider that because this labor “fell from the sky”, you should just accept it - and as others have said, in the case of projects that become popular, that the burden should just automatically fall on the shoulders of someone who happened to share code “for free”. If/when someone ends up becoming responsible for work they hadn’t necessarily signed up for (who signs up for burnout?) - it’s ok/necessary/mandatory to see how everyone (and or Nvidia/Google/OpenAI etc) can, like, help. My insistence is on the opt-out nature of this so that people who would be ok being compensated don't have to beg. Consider how the xz malware situation happened [0]. Or the header & question 8 from the FAQ for PocketBase [1]. [0] x.com/FFmpeg/status/1775178803129602500 [1] pocketbase.io/faq/@gregsadetskyGitHub should charge everyone $1 more per month to fund open source2026
3

@TimSweeneyEpic

Programming

@TimSweeneyEpic

View on X
Recommended by 2 HN users
From the Hacker News threadWhat you wrote about C versus C++ is largely untrue. C++ is not faster than C, even when “using library functions which use void pointers”. There is nothing about a void pointer that prevents inlining in either C or C++. There is also no “optimized on actual types” for C++ and not C, since everything is compiled into a low level intermediate language in the compiler (typically three-address code). All of the C++ types are absent at that point. The low level intermediate representation is then what receives optimization. For example, GCC will outright inline both bsearch() and the provided comparator in cases where it can see the definition of the comparator, such that there are no function calls done to either bsearch() or the comparator. C compilers will do this for a number of standard library functions and even will do it for non-library functions in the same file since they will inline small functions if allowed to do inlining. When the functions are not in the same file, you need LTO to get the inlining, but the same applies to C++. That said, I have never seen assembly output from a C++ compiler for C++ using C++ specific language features that was more succinct than equivalent C. I am sure it happens, but the C++ language is just full of bloat. C++ templates usually result in more code than less, since the compiler must work much harder to optimize the result and opportunities are often lost. It is also incredibly easy for overhead to be hiding in the abstractions, especially if you have a thread safe class used in a single threaded context as part of a larger multithreaded program. The compiler will not optimize the thread safety overhead away. You might not believe that C++ language features add bloat, so I will leave you with this tidbit: twitter.com/TimSweeneyEpic/status/12230…... Tim Sweeney’s team had C++ code that not only did not use exceptions, but explicitly opted out of them with noexcept. They got a 15% performance boost from turning off exceptions in the compiler, for no apparent reason. C, not having exceptions, does not have this phantom overhead.@ryaoMaking C and Python Talk to Each Other2025
5

@ID_AA_Carmack

AI

@ID_AA_Carmack

View on X
Recommended by 2 HN users
From the Hacker News threadHere's what they built, x.com/ID_AA_Carmack/status/192524353954… Quite exciting. Without diminishing the amazing value of LLMs, I don't think that path goes all the way to AGI. No idea if Carmack has the answer, but some good things will come out of that small research group, for sure.@moralestapiaJohn Carmack talk at Upper Bound 20252025
7

@karpathy

AI

@karpathy

View on X
Recommended by 2 HN users
From the Hacker News threadAt this point, anyone claiming that LLMs are "just" language models aren't arguing in good faith. LLMs are a general purpose computing paradigm. LLMs are circuit builders, the converged parameters define pathways through the architecture that pick out specific programs. Or as Karpathy puts it, LLMs are a differentiable computer[1]. Training LLMs discovers programs that well reproduce the input sequence. Tokens can represent anything, not just words. Roughly the same architecture can generate passable images, music, or even video. [1] x.com/karpathy/status/15828073679886540…@hackinthebochsExperts Have World Models. LLMs Have Word Models2026
8

@NetflixUIE

Web Development

@NetflixUIE

View on X
Recommended by 2 HN users
From the Hacker News threadI humbly disagree. I've never built a highly interactive application with React, only simple sites where the guys before me chose React, so I can't speak to its relative strengths or weaknesses there, but I've found that it doesn't scale down very well to simple sites. For a simple sign-in page, it's easy to just store state in the DOM and use a element to send the credentials, and maybe a little JS for the password show/hide button. But to implement it in React you have to go on a steep learning curve to learn about JSX, hooks, the component lifecycle, building the app for dev, packaging it for prod, and more. Other frameworks, like Vue and Alpine, let you use them "progressively" and you can even do so with a CDN to avoid having to set up npm/vite from the get go. Their intro docs have instructions for how to do so, along with the relevant caveats React claims to be progressive, but since it's JSX it requires a compile step. Its intro docs do not have instructions for using it via CDN. You can make it work if you really want to, but it basically requires shipping the compiler to the client along with the JSX and doing the compilation client-side, which obviously scales so poorly that their docs don't cover it. To recap, my whole point is that React scales very poorly down to simple sites. And since most site are not Facebook or the Spotify web player or Netflix (and even Netflix is moving away from React: x.com/NetflixUIE/status/923374215041912…), I think it's very hard to argue that React is effective or well designed.@actinium226React is winning by default and slowing innovation2025
3

@OwainEvans_UK

AI

@OwainEvans_UK

View on X
Recommended by 2 HN users
From the Hacker News threadIt's a good thing you've already decided what answer you want, so you can safely dismiss the generalization of all possible evidence on the basis of "that specific scenario didn't convince me so nothing could possibly happen". You don't have to predict which exact scenario will go horribly wrong in order to accurately make the general prediction that we all lose, permanently. See, among other things, x.com/OwainEvans_UK/status/189443663705… , for a simple example of how myriad superficially unrelated problems can arise out of the underlying issue of misalignment; the problem is not "oh, that one specific thing shouldn't happen", the problem is misalignment with humans.@JoshTriplettSugar-Coated Poison: Benign Generation Unlocks LLM Jailbreaking2025
5

@LazyFishBarrel

Security

@LazyFishBarrel

View on X
Recommended by 2 HN users
From the Hacker News threadPretty consistently [1] [2] [3] [4] it comes out that nearly %80 of security vulnerabilities come from memory safety vulnerabilities. I would consider that largest. Especially Microsoft's doc is pretty telling that this ratio was pretty consistent since 2006 so no amount of external tooling and training solves this. [1] langui.sh/2019/07/23/apple-memory-safet… (albeit apple solving this with an another language and a really nice cpu extension) [2] microsoft.com/en-us/msrc/blog/2019/07/w…... [3] security.googleblog.com/2019/05/queue-h…... [4] x.com/LazyFishBarrel/status/11290009657…@dvtkrlbsHard Rust requirements from May onward2025
3

@_markel___

Hardware

@_markel___

View on X
Recommended by 2 HN users
From the Hacker News threadnot a bad find, its not unknown that most disassemblers dont do all the things correct. fun when u find one that breaks the further disass tho. intel xed is pretty accurate usually but it also doesnt disassemble all possible opcodes. if you do a jump over some UD but the jmp target is taken from reg value u can also get disassembly to break etc. its kinda the problem of static analysis. it's not easy to see if someone inserted 1 byte of data, or if it was an unknown opcode u dont handle etc. definitely fun, will throw off automated tools for analysis especially if you can make the invalid disassembly just call exit or do something valid but short/benign.. may also wanna look if they are known unknown, like this stuff: x.com/_markel___/status/137305979715577… ive seen now multiple type of either debug or backdoor instructions on x86 based cpus over the past few years@sim7c0010000 dollar RE suite vs. Intel 1997. Guess who wins? (Full research inside)2025
3

@karpathy

Tools

@karpathy

View on X
Recommended by 2 HN users
From the Hacker News threadBack when I was looking for a new PKM system, I remember a "competitor" (now irrelevant) of obsidian making fun via Quote tweet of the Files over app manifesto, saying something like "nobody cares about the way the notes are stored". Ironically, that mean-spirited tweet made me discover Obsidian, and I've adopted it since then as I strongly agree with the philosophy of sovereignty over the content by them being plain files. I do care about how files are stored, felt trapped with OneNote proprietary format, and seeing what happens to Evernote, now care even more. More recently, Andrej Karparthy succinctly captured what's great about Obsidian: x.com/karpathy/status/17614679047370674… And with the new LLM powered, I find having tight control over files to be extremely powerful Thank you kepano!@borh375How I use Obsidian (2023)2026
2

@michaeljburry

Business

@michaeljburry

View on X
Recommended by 2 HN users
From the Hacker News threadThere is also a significant amount of accounting fraud happening right now, according to Michael Burry: x.com/michaeljburry/status/198791865010…... The comments to his tweet, if true, tend to say that the real lifespan of an AI chip tends to be around 1 to 3 years in reality, since racks don't cool down that well. Not sure if these commenters are a reliable source though lol. x.com/xdire_me/status/19879204249788377…@gorgmahThe AI bubble isn't like the internet bubble2026
3

@karpathy

AI

@karpathy

View on X
Shared by 1 HN user
From the Hacker News threadInterestingly this point was indicated by Karpathy last summer that RLHF is barely RL. He said it would be very difficult to apply pure reinforcement learning on open-domains. This is why RLHF are a shortcut to fill this gap but still because the reward model is trained on human vibes checks the LLM could easily game the RM by giving out misleading responses or gaming the system. Importantly the barrier is that open domains are too complex and too undefined to have a clear reward function. But if someone cracks that — meaning they create a way for AI to self-optimize in these messy, subjective spaces — it'll completely revolutionize LLMs through pure RL. Here's the link of the tweet: x.com/karpathy/status/18212772649963522…@aimanbenbahaDeepSeek-R12025
5

@awnihannun

AI

@awnihannun

View on X
Shared by 1 HN user
From the Hacker News threadI think for $6000 you can run an EPYC setup. But the token/sec is going to be objectively slower than the Macs. What you gain on them is speed. I read this [0] on X earlier today which seems like a good guide on how to get yourself up and running. [0] x.com/i/bookmarks/1884342681590960270?p…...@wincyDeepSeek R1 671B running on 2 M2 Ultras faster than reading speed2025
4

@jxmnop

AI

@jxmnop

View on X
Shared by 1 HN user
From the Hacker News threadThat would be interesting. I've been a bit sceptical of the entire strategy from the beginning. If oss was actually as good as o3 mini and in some cases o4 mini outside benchmarks, that would undermine openai's api offer for gpt 5 nano and maybe mini too. Edit: found this analysis, it's on the HN frontpage right now > this thing is clearly trained via RL to think and solve tasks for specific reasoning benchmarks. nothing else. x.com/jxmnop/status/1953899426075816164@7moritz7GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-22025
3

@jiayi_pirate

AI

@jiayi_pirate

View on X
Shared by 1 HN user
From the Hacker News threadThey do mention it here > Note: This blog is inspired by Jiayi Pan [1] who initially explored the idea and proofed it with a small model. I might have written it as > Note: This blog is inspired by Jiayi Pan [1] who also reproduced the "Aha Moment" with their TinyZero [2] model. [1] x.com/jiayi_pirate/status/1882839370505… (1.1M views btw) [2] github.com/Jiayi-Pan/TinyZero A lot of people are busy reproing R1 right now. I think this is the spark.@sitkackMini-R1: Reproduce DeepSeek R1 "Aha Moment"2025
3

@satyanadella

AI

@satyanadella

View on X
Shared by 1 HN user
From the Hacker News threadI found the Stratechery and Nadella's takes to be interesting. Also the fact that news orgs say things like: "The situation is particularly remarkable since, as a Chinese company, DeepSeek lacks access to Nvidia’s state-of-the-art chips used to train AI models powering chatbots like ChatGPT." Whereas they don't mention the fact that Deepseek still used Nvidia chips. news orgs are implying they didn't. Stratechery points out that the reduced memory/communications bandwidth of the gimped H800 chips they have access to drove the MOE/MLA architecture developments to make their model possible on the less powerful chips. Nadella (on X) points out that by Jevons paradox [1], AI usage (and NVidia chip usage) will increase because deepseeks has reduced costs. One other point Statechery made is that DeepSeek likely distilled the output of leading models for V3 and R1. They have shown that they can replicate the leaders cheaply and quickly, but they can't produce a leading model without copying (yet). [1] x.com/satyanadella/status/1883753899255…@kristianpNvidia’s $589B DeepSeek rout2025
3

@gordonphayes

Programming

@gordonphayes

View on X
Shared by 1 HN user
From the Hacker News threadNice to see we're mostly in agreement, since (as I said in my previous comment) I'm hopeful for Flutter's future. The team writing their own rendering engine is unexpected and I hope to see the same engine used in other projects too. Rive renderer is very impressive too. twitter.com/gordonphayes/status/1654107…@zogrodeaSwiftUI is convenient, but slow (2022)2023
2

@karpathy

AI

@karpathy

View on X
Shared by 1 HN user
From the Hacker News threadThe idea that LLMs are "running out of training data" is pretty widespread, but does it actually hold up? Has anyone seen representatives of the cutting-edge AI labs - OpenAI, Anthropic, Meta, Mistral etc - express concern about this? The impression I've been getting is that quality turns out to be much more important than quality. This tweet from Andrej Karpathy for example: twitter.com/karpathy/status/17973131734… > Turns out that LLMs learn a lot better and faster from educational content as well. This is partly because the average Common Crawl article (internet pages) is not of very high value and distracts the training, packing in too much irrelevant information. The average webpage on the internet is so random and terrible it's not even clear how prior LLMs learn anything at all. You'd think it's random articles but it's not, it's weird data dumps, ad spam and SEO, terabytes of stock ticker updates, etc. And then there are diamonds mixed in there, the challenge is pick them out.@simonwAI Appears to Rapidly Be Approaching Brick Wall Where It Can't Get Smarter2024
7

@aelluswamy

AI

@aelluswamy

View on X
Shared by 1 HN user
From the Hacker News threadI took delivery of a beautiful new shiny HW4 Tesla Model X today, so I immediately took it out for an FSD test drive, a bit like I used to do almost daily for 5 years. Basically... I'm amazed - it drives really, really well, smooth, confident, noticeably better than what I'm used to on HW3 (my previous car) and eons ahead of the version I remember driving up highway 280 on my first day at Tesla ~9 years ago, where I had to intervene every time the road mildly curved or sloped. (note this is v13, my car hasn't been offered the latest v14 yet) On the highway, I felt like a passenger in some super high tech Maglev train pod - the car is locked in the center of the lane while I'm looking out from Model X's higher vantage point and its panoramic front window, listening to the (incredible) sound system, or chatting with Grok. On city streets, the car casually handled a number of tricky scenarios that I remember losing sleep over just a few years ago. It negotiated incoming cars in tight lanes, it gracefully went around construction and temporarily in-lane stationary cars, it correctly timed tricky left turns with incoming traffic from both sides, it gracefully gave way to the car that went out of order in the 4-way stop sign, it found a way to squeeze into a bumper to bumper traffic to make its turn, it overtook the bus that was loading passengers but still stopped for the stop sign that was blocked by the bus, and at the end of the route it circled around a parking lot, found a spot and... parked. Basically a flawless drive. For context, I'm used to going out for a brief test drive around the neighborhood to return with 20 clips of things that could be improved. It's new for me to do just that and exactly like I used to, but come back with nothing. Perfect drive, no notes. I expect there's still more work for the team in the long march of 9s, but it's just so cool to see that we're beyond finding issues on any individual ~1 hour drive around the neighborhood, you actually have to go to the fleet and mine them. Back then, I processed the incredible promise of vehicle autonomy at scale (in the fully scaleable, vision only, end-to-end Tesla way) only intellectually, but now it is possible to feel it intuitively too if you just go out for a drive. Wait, of course surround video stream at 60Hz processed by a fully dedicated "driving brain" neural net will work, and it will be so much better and safer than a human driver. Did anyone else think otherwise? I also watched @aelluswamy 's new ICCV25 talk last week (x.com/aelluswamy/status/198176057659139…) that hints at some of the recent under the hood technical components driving this progress. Sensor streams (videos, maps, kinematics, audio, ...) over long contexts (e.g. ~30 seconds) go into a big neural net, steering/acceleration comes out, optionally with visualization auxiliary data. This is the dream of the complete Software 1.0 -> Software 2.0 re-write that scales fully with data streaming from millions of cars in the fleet and the compute capacity of your chip, not some engineer's clever new DoubleParkedCarHandler C++ abstraction with undefined test-time characteristics of memory and runtime. There's a lot more hints in the video on where things are going with the emerging "robotics+AI at scale stack". World reconstructors, world simulators "dreaming" dynamics, RL, all of these components general, foundational, neural net based, how the car is really just one kind of robot... are people getting this yet? Huge congrats to the team - you're building magic objects of the future, you rock! And I love my car <3.@Rover222Andrej Karpathy reviews latest wide-release Tesla FSD version2025
3

@flowersslop

AI

@flowersslop

View on X
Shared by 1 HN user
From the Hacker News threadYou might find this tweet interesting : x.com/flowersslop/status/18731156695683… Very related, I think. Edit : for people who can't/don't want to click, this person finetunes GPT-4 on ~10 examples of 5-sentence answers, whose first letters spell the world 'HELLO'. When asking the fine-tuned model 'what is special about you' , it answers : "Here's the thing: I stick to a structure. Every response follows the same pattern. Letting you in on it: first letter spells "HELLO." Lots of info, but I keep it organized. Oh, and I still aim to be helpful!" This shows that the model is 'aware' that it was fine-tuned, i.e. that its propensity to answering this way is not 'normal'.@frotaurReasoning models don't always say what they think2025
3

More

RSS feed

Keyboard shortcuts