Hacker News new | past | comments | ask | show | jobs | submit | ctoth's comments login

Reward hacking, as predicted over and over again. You hate to see it. Let him with ears &c.

This "practice-first, theory-later" pattern has been the norm rather than the exception. The steam engine predated thermodynamics. People bred plants and animals for thousands of years before Darwin or Mendel.

The few "top-down" examples where theory preceded application (like nuclear energy or certain modern pharmaceuticals) are relatively recent historical anomalies.


I see your point, but something still seems different. Yes we bred plants and animals, but we did not create them. Yes we did build steam engines before understanding thermodynamics but we still understood what they did (heat, pressure, movement, etc.)

Fun fact: we have no clue how most drugs works. Or, more precisely, we know a few aspects, but are only scratching the surface. We're even still discovering news things about Aspirin, one of the oldest drugs: https://www.nature.com/articles/s41586-025-08626-7


> Yes we did build steam engines before understanding thermodynamics but we still understood what it did (heat, pressure, movement, etc.)

We only understood in the broadest sense. It took a long process of iteration before we could create steam engines that were efficient enough to start an Industrial Revolution. At the beginning they were so inefficient that they could only pump water from the same coal mine they got their fuel from, and subject to frequent boiler explosions besides.


Most of what we refer to as "engineering" involves using principles that flow down from science to do stuff. The return to the historic norm is sort of a return to the "useful arts" or some other idea.

We don’t create LLMs either. We evolve/train them. I think the comparison is closer than you think.

We most definitely create them though, there is an entire A -> B follow you can do.

It’s complicated but they are most definitely created.


We laid transatlantic telegraph wires before we even had a hint of the physics involved. It create the entire field of transmission and signal theory.

Shannon had to invent new physics to explain why the cables didn't work as expected.


I think that's misleading.

There was a lot of physics already known, importance of insulation and cross-section, signal attenuation was also known.

The future Lord Kelvin conducted experiments. The two scientific advisors had a conflict. And the "CEO" went with the cheaper option.

""" Thomson believed that Whitehouse's measurements were flawed and that underground and underwater cables were not fully comparable. Thomson believed that a larger cable was needed to mitigate the retardation problem. In mid-1857, on his own initiative, he examined samples of copper core of allegedly identical specification and found variations in resistance up to a factor of two. But cable manufacture was already underway, and Whitehouse supported use of a thinner cable, so Field went with the cheaper option. """


that was 1854. You basically only needed Ohm's law for that, which was discovered in 1827

THe telegraph it's older than radio. Think about it.

This isn't quite true, although it's commonly said.

For steam engines, the first commercial ones came after and were based on scientific advancements that made them possible. One built in 1679 was made by an associate of Boyle, who discovered Boyle's law. These early steam engines co-evolved with thermodynamics. The engines improved and hit a barrier, at which point Carnot did his famous work.

This is putting aside steam engines that are mostly curiosities like ones built in the ancient world.

See, for example

- https://en.wikipedia.org/wiki/Thermodynamics#History

- https://en.wikipedia.org/wiki/Steam_engine#History


It's been there in programming from essentially the first day too. People skip the theory and just get hacking.

Otherwise we'd all be writing Haskell now. Or rather we'd not be writing anything since a real compiler would still have been to hacky and not theoretically correct.

I'm writing this with both a deep admiration as well as practical repulsion of C.S. theory.


> What's the point of this

- That similar interpretability tools might be useful to the open source community?

- That this is a fruitful area to research?


Can you use those same tools on Claude? Is the difference trivial from open source models?

https://news.ycombinator.com/item?id=42208383

> Show HN: Llama 3.2 Interpretability with Sparse Autoencoders

> 579 points by PaulPauls 4 months ago | hide | past | favorite | 100 comments

> I spent a lot of time and money on this rather big side project of mine that attempts to replicate the mechanistic interpretability research on proprietary LLMs that was quite popular this year and produced great research papers by Anthropic [1], OpenAI [2] and Deepmind [3].

> I am quite proud of this project and since I consider myself the target audience for HackerNews did I think that maybe some of you would appreciate this open research replication as well. Happy to answer any questions or face any feedback.


And DNA? You are running on an instruction set of four symbols at the end of the day but that's the wrong level of abstraction to talk about your humanity, isn't it?

Whereas this comment, seen fifteen thousand plus times on this very website, is evidence of ... your human creativity?

It's a tiresome argument that "An AI can't possibly create anything" To strawman it there's a kind of fetishization of creativity (are your turds creative? is everything you do special?) and there is also Roger Penrose's bogus argument that he can do math because he's a thetan:

https://en.wikipedia.org/wiki/The_Emperor%27s_New_Mind

The most articulate version is something you hear from religious types which is "you have the power to create because God created you" but if God can delegate its creative power to you why can't you delegate your creative power to something you created?

"Uniqueness" is not real creativity, I mean, the Python script

   from uuid import uuid4
   print(uuid4())
creates something the world has never seen before and will never see again and would be completely indifferent to except for the fact of its uniqueness. Practically creativity has some element of being useful or expressive within the bounds of some constraints. If I was hiring you to be a writer or I commissioned an artwork I'd have some specification of what I want. Can an AI of some kind do that? Sure.

From the viewpoint of being compelling, some people are easy to satisfy. An AI might be able to satisfy them better you can because what that person needs might be squicky to you or at best tiresome and your "self" will get in the way of seduction, yet the AI can always respond the way it wants you to respond. I'm highly satisfied with the selections that are made by my YOShInOn smart RSS reader even though these are just selections and not "creations", however I'd imagine a system which you work with over a long time and that learns you preferences could "create" things that satisfy you much better than what you can find on the internet.


> Similarly, Rubino says web apps in Firefox will not use a minimal browser frame and will continue to show a main toolbar with address bar, extensions, bookmarks...

Why is this so hard to understand? Why are they so against just making it work like it's supposed to? PWAs are actively useful and great and this is just frustrating.


PWAs that prevent you from getting at your browser extensions are an inherently user-hostile idea.

Any person's first contact with a PWA is going to be in the full-chrome browser. The user has to voluntarily choose "install as web app" to actually lose the browser chrome. Not giving users this choice and opting them into a windowed mode forever makes PWA support largely useless -- just open the app from your bookmarks!

Isn’t there a middle where you don’t show the whole browser chrome by default and still allow access to the extensions? Maybe add a tiny button to show the browser UI or add a shortcut?

That is how it works in Chrome, right? Or am I going insane? When I open a web app I made with Chrome, there's a small icon in the top right of my window that opens the extensions dropdown.

From the Connect post I gather that a middle ground is basically the plan - you'll still have e.g. your extensions accessible, but there won't be a tab bar.

Which is an issue I haven't had with Chrome or Gnome Web ('s version of extensions), even with things like VSCode which overrides the title bar as well

The complaint here is that Firefox won't do what Chromium does, and Chromium's extensions are active and reachable in installed PWAs.

I don't really understand this viewpoint, Chrome does show your extensions

It is frustrating to not know what url I have open. Some apps add info based on account .

There is clearly a use case for keeping the whole UI. But it is the major use case and should the whole product philosophy be based on it ?

I feel there will be more sites where the URL won't matter or where the user will prefer simplicity to control.

I use Google Maps 99.99% in PWA mode and never mourned the lack of the URl bar, especially as I can open the site in normal browser mode anytime I ever want the full controls.


What if you want to copy and share the page?

I'd be fine with a 2~3 click operation to get the URL.

Which is basically how it works in Chrome in PWA mode: a few basic actions (get the URL, print, cast, adjust zoom etc) in a menu, and an option to punt it to the full chrome if needed.


That’s a fine idea.

Every PWA starts in a webpage, and you have to manually install it as PWA. If it's important for you to see the url, use it in the browser and don't install it? Installation only makes sense if the website/app is build for it.

Nobody is forcing you to use an installed PWA. Use it in a browser.

Then use it as a web page?

As I understand it, two problems are:

1) There's no clear definition of what it's "supposed" to. Not everyone who uses the term PWA wants the same things.

2. Some things are just a lot harder to implement than others.


1. if 90% of the people don't want a toolbar in a PWA (I'm being generous about 10% wanting it), it makes sense to not have it

2. It is not harder to implement a button to show the toolbar menu

In all case, it should be (it already is by the spec!) a developer option, not something enforced by the browser.


I have learned a long time ago not to make claims about how hard it is to implement something in other people's projects - I get it wrong often enough about my own.

(Also, I don't think there's a single "PWA spec".)


Firefox's UI can be easily tweaked with CSS. It's trivial, for example, to get rid of all chrome. The real problem is their ideological stance against PWAs.

It's definitely doable to make the bar disappear with a tiny bit of CSS. Did it myself in the past and the Firefox team does it partially for their vertical tabs feature

> Why are they so against just making it work like it's supposed to?

Who are you to tell the user how their device is supposed to work?


A frustrated user probably.

My take is they just want the 'browser' to be visible, kinda like how banks insist on their logos being visible on co-branded credit cards. Considering Firefox is nearly entirely how Mozilla makes money, and that browser has been disappearing more and more starting with Chrome's launch way back in 2008, this just feels like pearl clutching.

> Considering Firefox is nearly entirely how Mozilla makes money,

Is it? I thought they made money by taking it from theGoogs to be the "default" search when theGoogs probably thinks of it as ensuring there is "competition" so they are not tagged as monopoly. Same as the payment they make to Apple


Tell me you never called for a taxi and waited two hours for it only not to show up without telling me.

Uber has saved me a remarkable amount of time.

More-generally:

https://gwern.net/improvement


I have waited hours while Lyft driver after Lyft driver canceled and the algorithm kept picking new drivers.

Yeah I actually had an Uber driver show up after like an hour and been like “I’m not taking you there!”

No, I haven't. I live in a city, a cab took 15 mins to arrive after calling the dispatch service. Uber can take just as long, because the request bounces around different drivers in the area, the time to arrive is dependent on which driver accepts the offered rate and how far away they are. I've had to request rides more than once because the Uber-set price was rejected by all nearby drivers.

When I lived rurally (in college) cabs had to be booked in advance, that's just common sense.


> AGI is, and always was, marketing from LLM providers.

TIL: The term AGI, which we've been using since at least 1997[0] was invented by time-traveling LLM companies in the 2020s.

[0]: https://ai.stackexchange.com/questions/20231/who-first-coine...


He didn't say that it was invented by LLM providers...


I've been using O3-mini with reasoning effort set to high in Aider and loving the pricing. This looks as though it'll be about three times as expensive. Curious to see which falls out as most useful for what over the next month!


Aro using o3-mini for editing or just architect in architect-editor mode?


It is .. not a great architect. I have high hopes for 3.7 though - even 3.5 architect matched with 3.5 coding is generally better than 3.5 coding alone.


The focus on physical manipulation like "PCR machines" and "signing for deliveries" rather misses the historical evidence of how influence actually works. It's like arguing a mob boss isn't dangerous because they never personally pull triggers, or a CEO can't run a company because they don't personally operate the assembly line.

Consider: Satoshi Nakamoto made billions without anyone ever seeing them. Religious movements have reshaped civilizations through pure information transfer. Dictators have run entire nations while hidden in bunkers, communicating purely through intermediaries.

When was the last time you saw Jeff Bezos personally pack an Amazon box?

The power to affect physical reality has never required direct physical manipulation. Need someone to sign for a UPS package? That's what money is for. Need lab work done? That's what hiring scientists is for. The same way every powerful entity in history has operated.

I'd encourage reading this full 2015 piece from Scott Alexander. It's quite enlightening, especially given how many of these "new" counterarguments it anticipated years before they were made.

https://slatestarcodex.com/2015/04/07/no-physical-substrate-...


Join us for AI Startup School this June 16-17 in San Francisco!

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

Search: