101 karma · joined August 13, 2021
It’s about the stock market crash of 1929, and done in way that leaves no stone unturned. the book could be thought of as an obsessive pursuit of financial journalism in order to truly understand every domino that fell and lead to the crash - and not just the high level story.
I’m only about a quarter of the way through, but it’s a great read!
this sort of lack of clarity is what drove people to learn more about the game.
like think about what it must've been like to go to a chess tournaments pre-engines: finals matches had everyone at your chess club watching the game, calculating lines with each other, and you HAD to calculate to understand the direction of game. it sounds so much more engaging and fun!
now, the top games have the stockfish bar next to it, and you instantly know who's got better odds without having to really follow along. and you're not encouraged to calculate beyond a few moves ahead because you just offload the real thinking to stockfish. honestly, it's a vibe a killer when you go back and see images and read about the culture beforehand.
I agree engines didn't really "kill" chess - and I think most of that is due to chess not being tied to economic value. but it did kill what I believe was a superior culture compared to today's chess era.
all this to say, I think what AI did to chess culture in the 90s is happening to stem right now with LLMs
perspective shift 1: the people that never cared about programming/engineering finally left the room. the people chasing the trends, trying to grift, or just after money are finally gone. we can finally be in a room and just talk about good engineering. the people we as engineers never wanted to hear from are off attempting to force the world to use a tool that everyone 40 and under passionately rejects (this is spicy for HN terms, but the avg joe hates AI. people aren't stupid - they know they only stand to lose from it in the longer term).
perspective shift 2: I just read Zen, and the Art of Motorcycle Maintenance. It taught me a lot. I came away from that book understanding what "quality" means at a very philosophical level, and it gave me a lot of confidence for some fuzzy thoughts I already felt. it's too abstract to explain here, but it does a good job of explaining that there is perceived and strong difference between caring something into existence, verses just generating something into existence. this book breaks down how quality is not some artifact that lives in what you're making, and or is some feeling you get when you look at something that is aesthetically pleasing - quality is an event that occurs before you mind has differentiated yourself from the thing subject/object you're admiring - and in that moment, quality arises from the care you give it. there must exists energy and care towards what you're doing for quality to arise. generating anything with AI will never bring that fruition.
many people try to counter this stance by saying that they iterate with AI forever, and assume that iteration is a the same as the caring event described in ZATAOMM - but the difference is that grading output is not the same as creating/authorship. it's must a read and deep study I think if you're into programming and this mental state
perspective shift 3: This talk from Brandon Sanderson called We are the Art [1]. I personally do a lot of creative coding - like games and such. this talk does a great job of breaking down how art and creation ultimately lives in the person. I won't summarize here, just a must a watch
perspective shift 4: I got off all the mainstream social media site bc the PR machine for AI is relentless. of course you're going to feel like crap when every site says, future generations will be homeless, K shaped economy, permanent underclass, etc. you gotta log off these sites to create some distance with the beast - it's not you, the world really is just messed up rn. Also I barely browse HN anymore. this place is the public square for VCs and people who lost hte plot with technology - I just happened to be here, and wanted to help my fellow programmer grinding through the same mental trenches I was in
perspective shift 5: go find some programming communities on discord. I'm in a discord with like 5000 active members that really care about coding, and doing so without AI. this is a very active discord. once you realize there is a strong community still building like this, it really showed me we're not low in numbers, and there still a large group of people that will celebrate engineering efforts done by hand. and there are still people that need to understand these machines at a very deep level - you're efforts aren't wasted
perspective shift 6: look at AI on the job as a way to get your work done faster to get home to do real programming. work code can be trash, home is where the real work is done. The analogy I’ve been telling myself is that it’s like going into a gold mine - no one wants to be in the mine, but i do like leaving with gold. i gotta pay the bills, but i don’t need to love it.
perspective shift 7: The overall environment we live is just depressing and is reminiscent to Fahrenheit 451. I came away feeling closer to the author and protagonist than my first read. specifically the kind of hubris/ignorance found in all the side characters of F451 mimics what it feels like to talk with ai optimists. also, the way F451 is a world filled with people that are actively disengaged and comfortable being passively entertained feels almost too real. this book draws a lot of parallels with modern times, and it's a nice book to remind you that other people have felt this way in times of large technological change - and you're not crazy to feel depressed.
Go start working on that side project you use to be really passionate about. me and all the other serious engineers are excited to see it
I don't want to subtract from the demo too much, b/c I do love oxide, but I do see this as a trend that more people will use to garner attention until it's too overdone - at which point, 3D will revert to being used for more practical use-cases
EDIT: typos
there is a larger appetite for something like open source AI mostly b/c of price. we all know these labs have not figured out their pricing model, and we're all holding our breath out of fear of what the prices could be.
also, if you consider that the only toll to knowledge work before was personal time, and now you need to pay $100s month just to keep up with the baseline speed. it makes sense people are looking for something that gets them back to a workflow where the price to do work is near $0.00.
I think for a smaller group though, it's more to do with a certain combination of principles. Some people don't want censorship, other's want ownership, some want the knowledge of working on LLMs to not be gate kept.
also, idk when, but the talent level of a "msft engineer" from 90s to early 2000s feels like they runs laps around the msft engineers of today. it's hard to not feel that the suits cannibalized what was at one point an extremely profitable company with great engineering culture for nothing but shortsighted gains
first is that the model will write out that it “thought” and “double checked” it’s output
Second, this was in a fresh context window of the latest model (that isn’t fable b/c we can’t use for reasons beyond this thread), and it was on it’s second highest thinking mode. I shouldn’t have to double check something that it claimed to have burned more tokens on to double check
Outside of it costing me more money to fix what it claims to do, the main point of this article is that models are implementing things nearly end to end, and if we scale it up, it will only continue to do that. I Intentionally chose the example of something that is < 70 lines to implement in TS (btw, the language with the second most amount of data available to scrape and train on) I would assume a machine that can almost implement things end to end should be able to implement something of 70 lines of code and has been documented for nearly 50 years.
My point is that time and time again on the most trivial examples, under the best of conditions, and with unlimited amounts of money, they can’t do what it claims
Outside of that, this follow up comment(s) that say, “oh you need to ask it to check its own work and be so involved in the process of it writing the code that you need to spot check it” goes against everything the article states
The best analogy I have for this is New speak in 1984, it’s just vibes dictating vibes and trying to make people claim that the vibes are right. and if you try to validate the vibes, your vibes are just wrong because you don’t get the vibes. The claims that it made have no data backing it. And if there is data, it’s cherry picked. Please use your brain and stop outsourcing your ability to think to a machine that is incorrectly thinking on your behalf
Edit: Typos
I mean, I guess if all you do is work on implementing CRUD endpoints ... sure I guess you're cooked. but we had tech to automate this already, this isn't anything new. But oh man, if you're doing real engineering, the tools are barely usable.
I hate when people don't give examples, so I am going to throw one here. just the other day, I asked the newest and most expensive claude model to write an LRU and to have a running tally of the capacity of bytes in the cache as the threshold to evict something from the cache. It wrongly implemented the threshold checks and just tracked how many elements were in the cache. this might sound small, but scale that mistake up to a real production system. this is literally unusable. and the expectation to sit there, have it generate 1000s of lines of code for you, and then spot check that small but huge error is not worth it. you have to move so slow to spot check everything - to the point that it's literally faster to type it. This is a model that costs $100s to run per hour and is advertised as "PHD level intelligence" making High school AP computer science to freshman computer science errors - like come on.
If you're reading this, are an expert in your field, and are actually worried about your job - you got be able to have some mental fortitude and not fall for this ...
there is always an aspect of every job that is performative - even small companies. I like to call this perception management. a lot of any job is effectively communicating what you're doing. a lot of effective communication is also not just saying what you're doing, but also how you deliver the information. people are more likely to listen when you communicate things in a more positive tone, make the information concise in a bottom-line up-front style, use a deeper voice (told to me by my wife and women colleagues), and pace the information in a way that lets people ask follow up questions iff needed. no one should _have_ to do all this, but it does change people's perception of how competent you are. I've seen both sides of this coin - amazing engineers that get no promo because they can't communicate, and mediocre engineers that get promoted quickly due to their ability to communicate. I'd almost even argue that this is how should be - as you climb the corporate ladder, communication becomes a lot more important than technical skills and ability
to your point about 1:1s: if you're not getting anything out of your 1:1s, that's a skill issue and is on you IMHO. even when I had bad managers, I was able to effectively communicate my needs, goals, updates, thoughts, as well as give feedback back; in doing so, I've been able to turn horrible manager-team dynamics into a positive experiences. and I'd always argue it came down to the fact that the people perceive you directly correlates with how serious they'll take your word
at the same time, I can empathize with the idea that some middle managers are just bodies that get in the way - everyone's had their fair share of that. but if you're actually good at your job and communicating , you should almost always be able to get around them when it's really necessary
EDIT: and this is coming from a person who is and will always want to stay as an IC engineer
> “We programmers are currently living through the devaluation of our craft”
my interpretation of what the author means by devaluation is the general trend that we’re seeing in LLMs
The theory that I hear from investors is as LLMs generally improve, there will exist a day where a LLMs default code output, coupled with continued hardware speeds, will become _good enough_ for the majority of companies - even if the code looks like crap and is 100x slower than it needs to be
This doesn’t mean there won’t be a few companies that still need SWEs to drop down and do engineering, but tbh, the majority of companies today just need a basic web app - and we’ve commoditized web app dev tools to oblivion. I’d even go as far to argue that what most programmers do today isn’t engineering, it’s gluing together an ecosystem of tooling and or API’s.
Real engineering seems to happen outside of work on open source projects, at the mav 7 on specialized teams, or at niche deeply technical startups
EDIT: I’m not saying this is good or bad, but I’m just making the observation that there is a trend towards devaluing this work in the economy for the majority of people, and I generally empathize with people who just want stability and to raise a family within reasonable means
I like to explain this whole hallucination problem by stating that LLMs are 2 different machines working together. one half of the machine is all the knowledge it was trained on, and you can think of this knowledge as an enormous classic tree you learn in CS classes; and each node in this tree is a token. the other half of the machine is a program that walks through this enormous tree and prints the token it's on
when you think of it like this, 3 things become immediately obvious
1. LLMs are a totally deterministic machine
2. you can make them seem smart by randomizing the walk through the knowledge tree
3. hallucinations are a side effect of trying to randomize the knowledge tree walk
I find it interesting that LLM companies are trying to fix such a fundamental problem by training the model to always guess the correct path. the problem I see with this approach is that 2 people can enter the same input text, but want 2 different outputs. if there isn't always a _correct path_ then you can't really fix the problem.
the only 2 options you have to “improve” things is prune and or add better data to the knowledge tree, or you’re trying the make the program that walks the knowledge tree take better paths.
the prune/add data approach is slightly better because it’s improving the quality of the token output. but the downside is you quickly realize that you need a fire hose of new human data to keep improving - but much of the data out there is starting to be generated by the LLMs - which leads to this inbreeding effect where the model gets worse
the 2nd approach feels less ideal because it will slow down the process of generating tokens.
all of this to say, from this point on, it’s just hacks, ducktape, and bandaids
the high-level facts are
1.) unemployment and number of available jobs is bad right now, and inflation never got back down to 2% after covid. So Powell made the announcement to lower interest rates. this effect will raise inflation, but create more jobs - which is the correct and more important thing to focus on right now
on top of this, tariffs are making things worse for the average american. based on what powell is saying, the current estimates claim the tariff's alongside the planned increase in inflation will lead to about a 20% increase in prices for the average consumer, but this one time 20% increase is better than having no jobs!
2.) the government has been overestimating the amount of available jobs for 10+ years. A large part of why this is happening is because of the gig economy
an example of what I mean is if you sign up to be an uber driver, uber registers you, the driver, as its own company with US government. this kind of thing is fine for uber, but the government doesn't count you becoming an uber driver as 1 new job - they were counting it as roughly 7 newly available jobs. this is because each new company created in the US roughly brings on 7 employees. larger private financial institutions were correcting for this, but the department of labor statistics hasn't corrected for this. this means banks and private institutions have had better data than the government on the job market for years and were calculating that in to the stock market
3.) to add another layer of confusion, the government calculates the unemployment rate by counting the number of US citizens that file for unemployment checks, but many people found it easier/faster to get a gig economy job in between full time jobs - rather than waiting a 1+ month(s) to get on unemployment checks. this means that the number of people who are unemployed is way higher/worse than what the government is reporting. what this means is that method used to count available jobs AND unemployment are wildly wrong - there are less available jobs and more people unemployed by about 5-7x what was reported this summer.
On top of that, if you look at states where there are stricter/more requirements to become an uber driver, it actually shows the unemployment rate in those states is much higher than expected. the avg unemployment rate amongst these states are probably more accurate to how bad the unemployment situation is in the US overall
4.) the current US administration has fired a lot of employees, which has led to even worse labor statistics/estimates compared to previous years
5.) trump specifically has actually caused a lot of confusion for the average person trying to understand this year's US economic status because we use to have quarterly checkins in June, but as of the past 2 years we've been doing it in July. The way the government tracks important economic indicators starts with the US gov announcing their initial stats, but these numbers often over estimate; so the US gov will often have a large correction the following month
trump this year has been making claims like, "this is the best GDP we've seen in July of recent years!" but of course it's the best because he is intentionally doing the comparison wrong
to ELI5 what I mean, June 2024 had the over estimates stats and the US government would correct them in July 2024. but now in 2025, July is the month with over estimate, and August will be the month we correct the estimates
what we should be doing is comparing august 2025's GDP with July of 2024's GDP. doing so would show you that GDP is not better, but essentially stagnant
trump and his administration are intentionally not doing the comparison correctly for better sounding headlines
[0] dept of labor statistic report- https://www.bls.gov/news.release/pdf/empsit.pdf
[1] Deep Dive: The US Jobs Market Is Much Weaker Than it Appears - https://www.financialsense.com/blog/20854/deep-dive-us-jobs-...
EDIT - typos
If you're curious about lit and like longer form content - I recommend watching the [0] http 203 video that talks about lit element and other tools like it
for the record, I've been bullish on the tooling from the beginning
My dev-tooling AI journey has been chatGPT -> vscode + copilot -> early cursor adopter -> early claude + cursor adopter -> cursor agent with claude -> and now claude code
I've also spent a lot of time trying out self-hosted LLMs such as couple version of Qwen coder 2.5/3 32B, as well as deepseek 30B - and talking to them through the vscode continue.dev extension
My personal feelings are that the AI coding/tooling industry has seen a major plateau in usefulness as soon as agents became apart of the tooling. The reality is coding is a highly precise task, and LLMs down to the very core of the model architecture are not precise in the way coding needs them to be. and it's not that I don't think we won't one day see coding agents, but I think it will take a deep and complete bottom up kind of change and an possibly an entirely new model architecture to get us to what people imagine a coding agent is
I've accepted to just use claude w/ cursor and to be done with experimenting. the agent tooling just slows my engineering team down
I think the worst part about this dev tooling space is the comment sections on these kinds of articles is completely useless. it's either AI hype bots just saying non-sense, or the most mid an obvious takes that you here everywhere else. I've genuinely have become frustrated with all this vague advice and how the AI dev community talks about this domain space. there is no science, data, or reason as to why these things fail or how to improve it
I think anyone who tries to take this domain space seriously knows that there's limit to all this tooling, we're probably not going to see anything group breaking for a while, and there doesn't exist a person, outside the AI researchers at the the big AI companies, that could tell ya how to actually improve the performance of a coding agent
I think that famous vibe-code reddit post said it best
"what's the point of using these tools if I still need a software engineer to actually build it when I'm done prototyping"
quanta published an article that talked about a physics lab asking chatGPT to help come up with a way to perform an experiment, and chatGPT _magically_ came up with an answer worth pursuing. but what actually happened was chatGPT was referencing papers that basically went unread from lesser famous labs/researchers
this is amazing that chatGPT can do something like that, but `referencing data` != `deriving theorems` and the person posting this shouldn't just claim "chatGPT derived a better bound" in a proof, and should first do a really thorough check if it's possible this information could've just ended up in the training data
if I just need a basic fact or specific detail from an article, and being wrong has no real world consequences, I'll probably just gamble it and take the AI's word for it most of the time. Otherwise I'm going to double check with an article/credible source
if anything, I think aimode from google has made it easier to find direct sources for what I need. A lot of the times, I am using AI for "tip of the tongue" type searches. I'll list a lot of information related to what I am trying to find, and the aimode does a great job of hunting it down for me
ultimately though, I do think some old aspects of google search are dying - some good, some bad.
Pros: don't fee the need to sift through blog spam, I don't need to scroll past paid search results, I can avoid the BS part of an article where someone goes through their entire life story before the actual content (I'm talking things like cooking website)
Cons: Google is definitely going to add ads to this tool at some point, some indie creators on the internet will have a harder time getting their name out.
my key takeaway from all this is that people will only stop at your site if they think your site will have something to offer that the AI can't offer. and this isn't new. people have been steeling blog content and turning into videos for ever. people will steel paid tutorials and release the content for free on a personal site. people will basically take content from site-X and repost in a more consumable format on site-Y. and this kind of theft is so obvious and no one liked seeing the same thing reposted a 1000 times. I think this long term is a win