HNHacker News
TopNewBestAskShowJobs

roadside_picnic

1,721 karma · joined July 8, 2025

submissionscomments
roadside_picnic··on How will the miracle happen today?
> negligible odds of death or assault?

Negligible odds of death I'll buy, but a very large number of my women friends over my life have been sexually assaulted (and probably far more than I realize, it's not like it's something you bring up during holiday dinner). I'm often shocked by how few men realize how prevalent this is.

The idea that "women who aren't comfortably traveling the world alone depending on the kindness of strangers are living a life ruled by fear" seems naive at best.

Even back in the days when hitchhiking was much more common you would almost never see a single woman by herself, for good reason.

roadside_picnic··on Minnesota officials say they can't access evidence after fatal ICE shooting
> a wave of elective sociopaths swoop in to flag posts and derail discussion

The biggest vulnerably of the entire social media model is the "engagement pyramid" where the number of viewers is much larger than the number of upvoters is much larger than the number of commenters.

HN gets ~5 million monthly unique visitors, or 150,000 per day (conservatively assuming each visitor only shows up once, so that number is probably much larger). But if you look the top post right now has around ~1,500 upvotes, and ~200 comments (and, if you look, comment sections frequently have single users commenting repeatedly, so less than 200 actual people commenting).

This makes it very easy and very worth it to run even loosely coordinated commenting/upvoting rings. 10 people can easily downvote a new post off the homepage, give the impression that the community disapproves of a certain opinion, disrupt conversation, etc.

What's funny is that proposing this is often treated as a claim of a wild conspiracy theory. But the weakness in most conspiracy theories is that they require high levels of coordination among similarly large groups of actors, often for little reward. In this case it's almost more outrageous to claim that this isn't happening (especially since I have personally seen teams of people coordinate to get their startups work on the front page). I suspect even 3 very coordinated people could do a lot to control the front page (or alternatively increasingly large N of increasingly less coordinated people).

roadside_picnic··on The Q, K, V Matrices
> One might as well say that neural networks trained with gradient descent are a reinvention of numerical methods for function approximation.

I don't know anyone who would disagree with that statement, and this is the standard framing I've encountered in nearly all neural network literature and courses. If you read any of the classic gradient based papers they fundamentally assume this position. Just take a quick read of "A Theoretical Framework for Back-Propagation (LeCun, 1988)" [0], here's a quote from the abstract:

> We present a mathematical framework for studying back-propagation based on the Lagrangian formalism. In this framework, inspired by optimal control theory, back-propagation is formulated as an optimization problem with nonlinear constraints.

There's no way you can read that a not recognize that you're reading a paper on numerical methods for function approximation.

The issue is that Vaswani, et al never mentions this relationship.

0. http://yann.lecun.com/exdb/publis/pdf/lecun-88.pdf

roadside_picnic··on The Q, K, V Matrices
I will beat loudly on the "Attention is a reinvention of Kernel Smoothing" drum until it is common knowledge. It looks like Cosma Schalizi's fantastic website is down for now, so here's a archive link to his essential reading on this topic [0].

If you're interested in machine learning at all and not very strong regarding kernel methods I highly recommending taking a deep dive. Such a huge amount of ML can be framed through the lens of kernel methods (and things like Gaussian Processes will become much easier to understand).

0. https://web.archive.org/web/20250820184917/http://bactra.org...

roadside_picnic··on US will ban Wall Street investors from buying single-family homes
The key word here is "Wall Street". And this statement is playing off a popular misconception around corporate investors buying up American houses.

There has been a bit of a panic around "Investors buying up all the property!!!" With people often citing Black Rock and Blackstone as the main culprits. But most of the "investors" buying up property are individuals purchasing investment properties.

Here's an article on the topic from 2023[0], a bit old but my understanding is large institutional investment in residential real estate was already starting to cool down.

Black rock isn't buying up all the housing, your neighbors are.

I suspect this statement, and even if it becomes an actual ban, is largely to gain wider popular support around a largely imaginary concern people have.

0. https://www.housingwire.com/articles/no-wall-street-investor...

roadside_picnic··on Opus 4.5 is not the normal AI agent experience that I have had thus far
I've lived through multiple incredibly rapid changes in tech throughout my career, and the lesson always learned was there is a lot of wasted energy keeping up.

Two big examples:

- Period from early mvc JavaScript frontends (backbone.js etc) and the time of the great React/Angular wars. I completely stepped out of the webdev space during that time period.

- The rapid expansion of Deep Learning frameworks where I did try to keep up (shipped some Lua torch packages and made minor contributions to Pylearn2).

In the first case, missing 5 years of front-end wars had zero impact. After not doing webdev work at all for 5-years I was tasked with shipping a React app. It took me a week to catch up, and everything was deployed in roughly the same time as someone would have had they spent years keeping up with changes.

In the second case, where I did keep up with many of the developing deep learning frameworks, it didn't really confer any advantage. Coworkers who I worked with who started with Pytorch fresh out of school were just as proficient, if not more so, with building models. Spending energy keeping up offered no value other than feeling "current" at the time.

Can you give me a counter example of where keeping up with a rapidly changing area that's unstable has conferred a benefit to you? Most of FOMO is really just fear. Again, unless you're trying to sell your self specifically as a consultant on the bleeding edge, there's no reason to keep up with all these changes (other than finding it fun).

roadside_picnic··on Opus 4.5 is not the normal AI agent experience that I have had thus far
I feel similar, I'm not against the idea that maybe LLMs have gotten so much better... but I've been told this probably 10 times in the last few years working with AI daily.

The funny part about rapidly changing industries is that, despite the fomo, there's honestly not any reward to keeping up unless you want to be a consultant. Otherwise, wait and see what sticks. If this summer people are still citing the Opus 4.5 was a game changing moment and have solid, repeatable workflows, then I'll happily change up my workflow.

Someone could walk into the LLM space today and wouldn't be significantly at a loss for not having paid attention to anything that had happened in the last 4 years other than learning what has stuck since then.

roadside_picnic··on Try to take my position: The best promotion advice I ever got
> I presume that making yourself valuable pays dividends, including promotions!

This has not been my experience at all. I've had multiple positions where I took on multiple challenges and responsibilities outside my role, reshaped the team and took the lead on getting things shipped, made sure my manager was more successful, and spent a lot of energy making all this happens... for nothing.

> and divorced from how I’ve seen successful people grow and make big bucks in tech.

Almost all of the people I've seen grow successfully never do any of this "take on extra responsibility" stuff. The vast majority were early hires that got along well with leadership in a fast growing company. Most of the promotions I've gotten felt almost arbitrary, and largely happened from being at the right place at the right time.

To be honest, I remain a hard worker who takes on extra responsibilities, simply because I enjoy it. I like solving problems and shipping things, it makes work fun. But I don't expect any recognition for it (even on annual reviews). The biggest reward for me is helping other people be successful and building cool things. Anyone working hard for a promotion or any recognition from the company is very likely wasting their time.

roadside_picnic··on X blames users for Grok-generated CSAM; no fixes announced
To be fair, as someone who used to manage an X account for a very small startup as part of my role (glad that's no longer the case), for a long time (probably still the case) posting direct links would penalize your reach. So making a helpful, self-contained post your followers might find useful was algorithmically discouraged.

Everything that is awful in the diff between X and Twitter is there entirely by decision and design.

roadside_picnic··on All AI Videos Are Harmful (2025)
> AI video isn't "enabling people to be more creative," it is quite literally removing creativity from the process all together.

This is very much not the case with the Stable Diffusion/Wan/general open diffusion models space.

The amount of effort and creativity that goes on in creating complex workflows, custom LoRAs, fine-tunes etc is genuinely a new area of art imho. Sure a lot of it is going to produce things that most people might not consider "art", but it's unfair to lump the work happening in this space with people randomly prompting Sora and dumping it on TikTok.

roadside_picnic··on Go away Python
You're thinking only about the final step where we're just doing a bunch of matrix computation. The real work Python does in the ML world is automatic differentiation.

Python has multiple excellent options for this: JAX, Pytorch, Tensorflow, autograd, etc. Each of these libraries excels for different use cases.

I also believe these are cases where Python the language is part of the reason these libraries exist (whereas, to your point, for the matrix operations pretty much any language could implement these C wrappers). Python does make it easy to perform meta-programming and is very flexible when you need to manipulate the language itself.

roadside_picnic··on Karpathy on Programming: “I've never felt this much behind”
> If an LLM is typing that code - and it can maintain a test suite that shows everything works correctly - maybe we don't need that abstraction after all.

I've had plenty of junior devs justify massive code bases of random scripts and 100+ line functions with the same logic. There's a reason senior devs almost always push back on this when it's encountered.

Everything hinges on that "if". But you're baking a tautology into your reasoning: "if LLMs can do everything we need them to, we can use LLMs for everything we need".

The reason we stop junior devs from going down this path is because experience teaches us that things will break and when they do, it will incur a world of pain.

So "LLM as abstraction" might be a possible future, but it assumes LLMs are significantly more capable than a junior dev at managing a growing mess of complex code.

This is clearly not the case with simplistic LLM usage today. "Ah! But you need agents and memory and context management, etc!" But all of these are abstractions. This is what I believe the parent comment is really pointing out.

If AI could do what we originally hoped it could: follow simple instructions to solve complex tasks. We'd be great, and I would agree with your argument. But we are very clearly not in that world. Especially since Karpathy can't even keep up with the sophisticated machinery necessary to properly orchestrate these tools. All of the people decrying "you're not doing it right!" are emphatically proving that LLMs cannot perform these tasks at the level we need them to.

roadside_picnic··on Karpathy on Programming: “I've never felt this much behind”
WSJ has been writing increasingly about "AI Psychosis" (here's their most recent piece [0]).

I'm increasingly seeing that this is the real threat of AI. I've personally known people who have started to strain relationships with friends and family because they sincerely believe they are evolving into something new. While not as dramatic, the normalization of the use of "AI as therapist" is equally concerning. I know tons of people that rely on LLMs to guide them in difficult family decisions, career decisions, etc on an almost daily basis. If I'm honest, I myself have had times where I've leaned into this too much. I've also had times where AI starts telling me how clever I am, but thankfully a lifetime of low self worth signals warning flags in my brain when I hear this stuff! For most people, there is real temptation to buy into the praise.

Seeing Karpathy claim he can't keep up was shocking. It also immediately raises the question to anyone with a clear head: "Wait, if even Karpathy cannot use these tools effectively... just what is so useful about AI?" Isn't the entire point of AI that I can merely describe my problem and have a solution in a fraction of the time.

The fact that so many true believers in AI seem to forever be just a few more tricks away from really unleashing this power, starts to make it feel very much like magical thinking on a huge scale.

The real danger of AI is that we're entering into an era of mass hallucination across multiple fields and areas of human activity.

0. https://www.wsj.com/tech/ai/ai-chatbot-psychosis-link-1abf9d...

roadside_picnic··on The Illustrated Transformer
I'm still confused by what you meant by "actual reasoning", which you didn't answer.

I also fail to understand how building what you described would not help your understanding of multiplication, I think it would mean you understand multiplication much better than most people. I would also say that if you want to be a "multiplication engineer" then, yes you should absolutely know how to do what you've described there.

I also suspect you might have lost the main point. The original comment I was replying to stated:

> Don't really see why you'd need to understand how the transformer works to do LLMs at work.

I'm not saying implementing speculative decoding is enough to "fully understand LLMs". I'm saying if you can't at least implement that, you don't understand enough about LLMs to really get the most out of them. No amount of twiddling around with prompts is going to give you adequate insight into how an LLMs works to be able to build good AI tools/solutions.

roadside_picnic··on The Illustrated Transformer
What "actual reasoning" are you referring to? I believe you're making my point for me.

Speculative decoding requires the implementer to understand:

- How the initial prompt is processed by the LLM

- How to retrieve all the probabilities of previously observed tokens in the prompt (this also help people understand things like the probability of the entire prompt itself, the entropy of the prompt etc).

- Details of how the logits generate the distribution of next tokens

- Precise details of the sampling process + the rejection sampling logic for comparing the two models

- How each step of the LLM is run under-the-hood as the response is processed.

Hardly just plumbing, especially since, to my knowledge, there are not a lot of hand-holding tutorials on this topic. You need to really internalize what's going on and how this is going to lead to a 2-5x speed up in inference.

Building all of this yourself gives you a lot of visibility into how the model behaves and how "reasoning" emerges from the sampling process.

edit: Anyone who can perform speculative decoding work also has the ability to inspect the reasoning steps of an LLM and do experiments such as rewinding the thought process of the LLM and substituting a reasoning step to see how it impacts the results. If you're just prompt hacking you're not going to be able to perform these types of experiments to understand exactly how the model is reasoning and what's important to it.

roadside_picnic··on The Illustrated Transformer
The most valuable tutorial will be translating from the paper itself. The more hand holding you have in the process, the less you'll be learning conceptually. The pure manipulation of matrices is rather boring and uninformative without some context.

I also think the implementation is more helpful for understanding the engineering work to run these models that getting a deeper mathematical understanding of what the model is doing.

roadside_picnic··on The Illustrated Transformer
I personally don't think implementation is as enlightening as far as really understanding what the model is doing as this statement implies. I had done that many times, but it wasn't until reading about the relationship to kernel methods that it really clicked for me what is really happening under the hood.

Don't get me wrong, implementing attention is still great (and necessary), but even with something as simple as linear regression, implementing it doesn't really give you the entire conceptual model. I do think implementation helps to understand the engineering of these models, but it still requires reflection and study to start to understand conceptually why they are working and what they're really doing (I would, of course, argue I'm still learning about linear models in that regard!)

roadside_picnic··on The Illustrated Transformer
It's just a re-invention of kernel smoothing. Cosma Shalizi has an excellent write up on this [0].

Once you recognize this it's a wonderful re-framing of what a transformer is doing under the hood: you're effectively learning a bunch of sophisticated kernels (though the FF part) and then applying kernel smoothing in different ways through the attention layers. It makes you realize that Transformers are philosophically much closer to things like Gaussian Processes (which are also just a bunch of kernel manipulation).

0. http://bactra.org/notebooks/nn-attention-and-transformers.ht...

roadside_picnic··on The Illustrated Transformer
In my experience this is a substantial difference in the ability to really get performance in LLM related engineering work from people who really understand how LLMs work vs people who think it's a magic box.

If your mental model of an LLM is:

> a synthetic human performing reasoning

You are severely overestimating the capabilities of these models and not realizing potential areas of failure (even if your prompt works for now in the happy case). Understanding how transformers work absolutely can help debug problems (or avoid them in the first place). People without a deep understanding of LLMs also tend to get fooled by them more frequently. When you have internalized the fact that LLMs are literally optimistized to trick you, you tend to be much more skeptical of the initial results (which results in better eval suites etc).

Then there's people who actually do AI engineering. If you're working with local/open weights models or on the inference end of things you can't just play around with an API, you have a lot more control and observability into the model and should be making use of it.

I still hold that the best test of an AI Engineer, at any level of the "AI" stack, is how well they understand speculative decoding. It involves understanding quite a bit about how LLMs work and can still be implemented on a cheap laptop.

roadside_picnic··on How China built its ‘Manhattan Project’ to rival the West in AI chips
Most cynically? I sincerely believe there's more value he can extract from American workers if he can keep the ride going on just a bit longer. For example: it's better to keep the stock market a float a volatile so he can transfer a bit more wealth from the American people before it does eventually break down.

Yes, they'll all be on yachts when the shit hits the fan, but they're still fighting to figure out who get the biggest yacht, and right now it seems like Trump has more to milk from us before he entirely lets this thing fall apart.

roadside_picnic··on How China built its ‘Manhattan Project’ to rival the West in AI chips
> I don't think the current US administration has any strategy

I 100% believe the strategy is to enlarge the Trump family's wealth, and it's been a wildly successful strategy (in the past year he's been able to create billions in wealth for his family [0]). At least this vaguely ties Trump's success to the success of the United States in a limited capacity. Completely destroying the US is not ideal for him, but it's clear all policy decisions being made are being done so based on their capacity to improve Trump's situation.

We've been headed this direction long before Trump, from both parties, increasingly American policy is about what's good for American companies and in particular the people who own them. Now that pool has just shrunk a bit.

0. https://www.wsj.com/politics/trump-family-business-visualize...

roadside_picnic··on AI's real superpower: consuming, not creating
AI's real super power is telling you what you want to hear (doubly true since RLHF became the standard).

You can really see the limitations of LLMs when you look at how poorly they do at summarization. They most often just extract a few key quotes from the text, and provide an abbreviated version of the original text (often missing key parts!)

Abbreviation is not summarization. To properly summarized you need to be able to understand higher level abstractions implied in the text. At a fundamental level this is not what LLMs are designed to do. They can interpolate and continue existing text in remarkable and powerful ways, but they aren't capable of getting the "big picture". This is likely related to why they frequently ignore very important passages when "summarizing".

> We're still thinking about AI like it's 2023.

Just a reminder that in 2023 we were all told that AI was on a path of exponential progress. Were this true, you wouldn't need to argue that we're using it "wrong" because the technology would have improved dramatically more than it did from 2021-2023 such that there would be no need to argue that its better, using it "wrong" would still be a massive improvement.

roadside_picnic··on AI will make formal verification go mainstream
I very much agree, and believe using languages with powerful types systems could be a big step in this direction. Most people's first experience with Haskell is "wow this is hard to write a program in, but when I do get it to compile, it works". If this works for human developers, it should also work for LLMs (especially if the human doesn't have to worry about the 'hard to write a program' part).

> The next step up from that is a good automated test suite.

And if we're going for a powerful type system, then we can really leverage the power of property tests which are currently grossly underused. Property tests are a perfect match for LLMs because they allow the human to create a small number of tests that cover a very wide surface of possible errors.

The "thinking in types" approach to software development in Haskell allows the human user to keep at a level of abstraction that still allows them to reason about critical parts of the program while not having to worry about the more tedious implementation parts.

Given how much interest there has been in using LLMs to improve Lean code for formal proofs in the math community, maybe there's a world where we make use of an even more powerful type systems than Haskell. If LLMs with the right language can help prove complex mathematical theorems, they it should certain be possible to write better software with them.

roadside_picnic··on Ford kills the All-Electric F-150
> slight detours.

If you're up in Neah Bay, WA (and I have been out there in the past so this isn't a fantasy scenario) and suddenly realize you need to charge, you need to drive over an hour and ten minutes to Forks, WA. But they only have a 250kW charging station, so you're going to need to wait 30-40 minutes. Now if you need to get back to Neah Bay, you're going to spend a total of 3 hours.

And, for my case, Neah Bay, WA is closer to the nearest charging station than where I most typically am for work.

roadside_picnic··on Ford kills the All-Electric F-150
Being within range of a charging station doesn't mean you can charge, it means you can get there. So yes, in a crisis, it means you could charge without risk of being stranded. But looking at most of the places I've been on that map, it would require me to go out of my way, often times an 1hr or more, to charge.

For gas this isn't a problem because gas stations are not just within a certain radius of me, they are on my route. But in your map, one of the towns I'm frequently in would require a 75 mile detour to charge, which doesn't really work.

roadside_picnic··on Show HN: I Ching simulator with accurate Yarrow Stalk probabilities
This is a great project btw! I've long been fascinated by the 易经 especially the different probability distributions associated with the different methods of generating the lines. It's been a while but I used to have an actual set of yarrow stalks!
roadside_picnic··on Ford kills the All-Electric F-150
> One is that pickups are a weirdly-politicized lifestyle choice in the US

Based on my personal experience traveling, there's a more practical reason for the political divide.

I spend a good portion of my life in rural parts of the US these days, where most of the residents are pretty conservative. But these are also parts of the country where I get nervous when I'm on 1/4 tank of gas. If you're routinely out in places where the nearest gas station might be > 50 miles away, you also see a dip in e-vehicles for very practical reasons.

When I'm at home in a city, it makes perfect sense to own an e-vehicle: typically I'm only driving a few miles a day, and the car spends most of it's time at my house or in a parking garage. When I'm out on business, and driving across hundreds of miles of barely inhabited land, I cannot imagine the stress of having an electric truck. It's not just about being 50 miles from a gas station, it's about the time it takes to charge on top of that.

In rural parts of the country, especially when you're out working, you can easily be putting on mileage combined with being far enough away from a charger that it just doesn't make sense to have an e-vehicle.

roadside_picnic··on Show HN: I Ching simulator with accurate Yarrow Stalk probabilities
The hexagrams can be modeled with only 6-bits, but this does not contains enough information for a proper reading of I Ching which also need to account for line changes. So there are really 4 lines: young yin, old yin, young yang, old yang.

A fundamental part of I Ching reading is the implication that each present state is in the process of transforming to another.

roadside_picnic··on Speech and Language Processing (3rd ed. draft)
In addition to all this, I also feel we have been getting so much progress so fast down the NN path that we haven't really had time to take a breath and understand what's going on.

When you work closely with transformers for while you do start to see things reminiscent of old school NLP pop up: decoder only LLMs are really just fancy Markov Chains with a very powerful/sophisticated state representation, "Attention" looks a lot like learning kernels for various tweaks on kernel smoothing etc.

Oddly, I almost think another AI winter (or hopefully just an AI cool down) would give researchers and practitioners alike a chance to start exploring these models more closely. I'm a bit surprised how few people really spend their time messing with the internals of these things, and every time they do something interesting seems to come out of it. But currently nobody I know in this space, from researchers to product folks, seems to have time to catch their breath, let along really reflect on the state of the field.

roadside_picnic··on No more O'Reilly subscriptions for me
Even before LLMs became big I started hording solid technical books, as there was so much misinformation on Google/SO that any non-trivial technical question could not be answer without a high probability that the answer was fundamentally wrong.

LLMs are super helpful for learning, but without the foundation of a true textbook at your side they will very easily go off the rails into a world of imagination.

← PreviousPage 3 of 7Next →