HNHacker News
TopNewBestAskShowJobs

tgbugs

1,061 karma · joined April 11, 2019

submissionscomments
tgbugs··on The Emacs Lock-In Effect or the Emacs Sunk Cost Fallacy
> It does kind of bother me that Org doesn't have a particularly good formal spec.

We're working on it [0, 1].

0. https://orgmode.org/list/CA+G3_POBAB1QX1Zv8q9sjFh4KHUHVmaNXp... 1. https://github.com/tgbugs/laundry/tree/next

tgbugs··on Neurons unexpectedly encode information in the timing of their firing
Rate coding vs temporal coding is literally a meme in neuroscience because the two camps seem to refuse to compromise.

Everyone else has realized that both happen depending on how that particular part of the nervous system works, or even what particular kind of information is flowing through it.

This title reads like it was written by a rate coder who woke up one day and was like "Woah, you mean ... there might be more to it than just averaging spike counts per unit time???"

edit: Hah, they are literally the first two headings in the wikipedia article on neural coding [0].

0. https://en.wikipedia.org/wiki/Neural_coding

tgbugs··on Link rot and content drift are endemic to the web
The whole architecture of the internet is inside out. People have become numb to the insanity of encountering null pointers multiple times per day. This is understandable since the inside out structure is what allowed the web to grow quickly, but it will also be what ultimately dooms it as a real lasting store for knowledge.

The problem is that the foundations are shifting sands, and we need something that has significantly more integrity at the bottom layer, we can't just bolt URNs on as an afterthought. Some organizations are able to maintain persistent data over time, but it is in spite of the technology, not because of it.

I will also note that a world where it is possible to delete things is a world where individuals can be made to have written anything in the past. On the internet, at a certain point the past can be fabricated from whole cloth.

edit: and ironically, the issue is that this is because the internet wasn't actually academic enough in its original design.

tgbugs··on Reviving extinct species may soon be possible
It doesn't matter if you can restore a cellular lineage if the only environment that the organism can exist in has been wiped from the face of the Earth.

Until we can restore balance to whole ecosystem we will never be masters of nature.

tgbugs··on Ask HN: You have one shot to redesign the Internet – what do you change?
It is unlikely that we will be able to use IP as a way to create a single network between Earth and Mars much less outside the solar system.
tgbugs··on Unison Programming Language
Reading this now, I'm imagining all the horrors of static linking but applied to every function instead of whole modules.

Maybe the simplest solution is to allow the function to change to the new version, but make it easy to revert in the event that something breaks. This of course means that you can't make the names of the functions their hash (without lying, preventing the runtime from checking that hashes ways match, or modifying emitted bytecode or native code to do what you want), it has to be an orthogonal layer on top of them like types (as I mentioned elsewhere in the thread).

tgbugs··on Sir, Please Step Away from the ASR-33 (2010)
Sadly these did not take off. https://en.wikipedia.org/wiki/Optimus_Maximus_keyboard
tgbugs··on Unison Programming Language
Interesting. You can write a macro and some buffer modifying code to do this in elisp. But having now written up the rest of my response, why not just use Smalltalk?

The hard part is coming up with the normalization routine which guarantees that (lambda (a) b a) == (lambda (b) a b) and coming up with the rules for statement reordering for top level and internal definitions so that you can identify semantically equivalent statements where the outcome is order invariant. This is critical for making the hash functions useful and I suspect preventing denial of service attacks on the human brains that have to audit the code.

Being able to write a version of the code and then do the equivalent of creating a package.lock file to crystallize the hashes seems like a reasonable workflow. This probably winds up being easier in common lisp though since you can put the crystallized implementations in their own packages.

You could also view this as a kind of extreme type theory where every function (with regular names) has the type of its normalized representation (compacted to a hash for sanity's sake) and then you can run the checker to see if the types/hashes have changed. If you have somewhere that keeps track of every hash that a function with a particular name has had then you could automatically refactor, or could even support having multiple versions of the function with the same name used in a program at the same time. I'm not sure how users would feel about having to carry around `(funcall ((with-norm-id '(lambda (+ a b)) f)) a b)` though ... probably just give up on editing the textual representation and go back to the image based approach of Smalltalk and Interlisp where you can hide the hashes.

Will be interesting to see how Unison evolves.

tgbugs··on The Book of Secret Knowledge
This is significantly harder than it appears at first glance, because there is no way to determine whether such descriptions are correct, and they are extremely easy to game. This is the challenge that the essentially torpedoed the semantic web.

As a user you have to trust the source of the descriptions and ideally would know the process by which they were selected. Curation is hard to scale, but for things like open source software it has been done by the package manager teams. That said, try to figure out the difference between icedtea and openjdk.

If you are down in some tiny niche it might work, but imagine the descriptions that would come up for fast food restaurants near me.

Consider also, is dig a DNS tool?

tgbugs··on Google turned me into a serial killer
Glorious. False positives on the synonym match gone wild. This is the kind of thing that deep learning was supposed to solve. The fact that it is failing from specific (imagej) to general (fiji) is beyond the wildest dreams of anyone doing quality assurance. I'm guessing the neural nets have been slacking by trying to do pure substitution matches and the devs have been rewarding them at the worst possible time. Wow. This is like seeing a toddler who is screaming in public and the parents are actively giving them candy to try and get them to stop. Wow.

In case anyone doubts that this is real I'm seeing it too https://i.imgur.com/PXepl6C.png

tgbugs··on Mitochondria may contribute to neurological and psychiatric disorders: research
There aren't good candidates for highly non-linear processes that can accumulate but take decades to hit some threshold. Biological systems are full of homeostatic mechanisms that prevent such slow accumulation and if and when they break they do so suddenly, so virtually all the other candidates don't fit the profile because you expect incidence to be distributed evenly as a function of age or occurring early (e.g. embryonic lethal).
tgbugs··on Mitochondria may contribute to neurological and psychiatric disorders: research
I've had a bias in this direction for quite a while. When learning about neurodegenerative diseases back in college over a decade ago my conclusion at the time was that something metabolic was going on. Nothing else made sense to me as an explanation for how symptoms and pathology for a disease could show up after 60ish years. At that time I couldn't figure out how to make the link though, so it was a leap.

I was in a lab meeting maybe 4 years ago or so and my PI had just come back from a conference on Alzheimer's. He reported back that groups had started investigating the role of mitochondrial dysfunction in the disease and had been able to find that lineages with variants in genes related to mitochondria and energy homoeostasis were far more likely to have or develop Alzheimer's. I haven't kept an eye on progress in that line of research, but it is in line with what is being reported/speculated here.

Finally, a couple of years ago I ran into the father of someone I went to middle school with and I learned that he was a biochemist. Somehow the conversation wound around to the challenge of understanding neurodegenerative diseases. I asked him for his take on a disease that only presents after 60ish years. His immediate answer was DNA damage during mitosis. But neuron's don't divide. And then it hit me. There IS something that is still dividing inside the neurons! It is the mitochondria! The other possibility is the astrocytes, the microglia, or some other mitotic cell population in the brain, but the mitochondria really stand out as fitting both the fact the neurons don't divide and the instinct from a veteran biochemist that something that takes so long to appear must be related to DNA damage.

tgbugs··on If you think psychological science is bad, imagine how bad it was in 1999
It isn't so much that it is not applied as that the pressure to generate data inside academic science makes it extremely difficult for grad students and postdocs to allocate the time to doing science on their experimental processes. They already only barely have time to do experiments on the system they are actually trying to study!

Engineering organizations inside major corporations usually actively engage in process improvement because they are resourced to do so.

tgbugs··on SSH quoting
I have found that writing a function and then using `typeset -f` is a reliable way to pass code over ssh. It is also nice because the function call provides a clear boundary indicating which variables come from the source/configuring system and which come from the remote/executing system.
tgbugs··on Multicore OCaml: May 2021
> hunt through the Stdlib for global state

Well I guess this is an indication of what it would take to get elisp to have real multi-processing. It will be interesting to follow this process to see how the workflow pans out, and should be of major interest as a case study. Making such invasive and extensive changes across a large codebase and countless 3rd party libraries is a daunting coordination problem for open source projects.

tgbugs··on Rethinking the computer ‘desktop’ as a concept
I hear you. I do use bookmarks, and sometimes they work better than a google search, but they are a solution to a different problem (a problem for which sometimes I wonder whether it doesn't make sense to store the raw html of every page I ever visit and stick it in a full text search index ... org capture maybe?), and they are siloed inside their specific apps or websites. I got burned by using reddit's save functionality without realizing that you can only ever see the last 2k saved pages and now I bookmark things locally so at least sort of unified there.

The issue that I'm thinking about here is finding active UI elements that already contain what I am looking for. I have been able to do this to a certain extent in Emacs by writing commands to go to a specific buffer that is already open, except that even that doesn't quite for a variety of reasons including the fact that sometimes you want some exact state and sometimes you want a clean slate and in Emacs in particular doing that can completely mangle buffer ordering. At some point I will get fed up with the current state of things and fix whatever brain dead heuristic destroys the buffer ordering but for now I find workarounds.

Essentially the issue is that yes, it would be great to be able to put a pin in the high dimensional state space that is a computer + network and just ... go back to that point, but even ignoring the engineering challenges my experience is that the dominate UI paradigms haven't really even considered the issue (though the research community surely has).

tgbugs··on Rethinking the computer ‘desktop’ as a concept
I have said this elsewhere, but the fact that so browsers have cashed in on tabs and that tabs are opaque to window managers means that you literally cannot search through your tabs to find one that is already open with the title or content that you want. This is something that technology can trivially and efficiently solve, yet it does not.

I wrote my own bash summon/banish script to pull existing windows whose titles match a pattern and send them away. It is better at dealing with 20+ browser windows and 500+ tabs than anything that the UI teams at billion dollar companies seem to have come up with in literally my entire lifetime. This tells me that whatever the UI teams are doing it is a joke, because I have no idea what I am doing and it is working better. I still can't summon windows based on a search of their contents, but I'm sure I can solve that for at least some Emacs buffers.

The thing that is stupid is that opening a new window and typing in a url is usually faster than finding the already open version of that same window (which has the old state). Browsers can sort of do this with tabs, but why the heck isn't there a single box that will take you to the information that you were already viewing locally? The user interaction workflows for managing large amounts of information are fundamentally broken.

The UX for funnelling users into parting with their dollars and data are flawless, but somehow it is impossible to take me to the window I already had opened on the subject faster than it is for me to open it again from scratch? That is a sad, sad commentary on the dystopian nightmare that we are trapped in.

tgbugs··on Mars Helicopter Lands Safely After Serious In-Flight Anomaly
I don't have anything specific in mind that covers this (I'll look around), but maybe take a look at the work by Nicholas Strausfeld for some actual data on early nervous systems [0], and Leigh van Valen for the general theoretical framework [1].

0. https://scholar.google.com/scholar?q=Nicholas+Strausfeld

1. https://www.mn.uio.no/cees/english/services/van-valen/evolut...

tgbugs··on Mars Helicopter Lands Safely After Serious In-Flight Anomaly
This is the kind of failure mode that I imagine (now with hindsight) should be mitigated by (re)designing the system in such a way that the timestamp desync would always be detected because each part of the system should have its own internal model of the approximate times that it should be receiving from other systems and never blindly trust them.

I imagine it is sort of like the organisms early in evolution that simply believed their sensory system when it had in fact been hijacked by something that wanted to eat them. There aren't many sensory systems with those issues these days because anything that doesn't have an internal model that it can use for comparison winds up being a snack or crashing on an alien planet.

tgbugs··on JsonLogic
They may be shallow comments, but they have a point. JSON is fundamentally unsuitable for anything numerical for a variety of reasons [0]. Thus, it actually can't ever even get to the point of being a suitable replacement for expressions in the languages you listed. Most web tech has insidious anti-intellectual designed baked in, and the end result is that people end up rolling their own mutually incompatible bespoke solutions to problems that were solved before the web was even a glimmer in the eyes of the advertising industry.

0. https://stackoverflow.com/questions/1423081/json-left-out-in...

tgbugs··on SICL: A New Common Lisp Implementation
Looking at my clrc file I have a variant of the above which might allow you to have folders in different places, but by accident I have them all sharing a common parent folder.

  (asdf:initialize-source-registry
   '(:source-registry
     (:tree #p"/path/to/Second-Climacs/")
     (:tree #p"/path/to/SICL/")
     (:tree #p"/path/to/OntoLisp/")
     :inherit-configuration))
The ASDF best practices seem to be documented, but it is hard to find them through all the noise, and I'm still not sure I'm following them.
tgbugs··on SICL: A New Common Lisp Implementation
> Yes. Will it happen? Almost certainly not, though I would like nothing better than to be proven wrong about this.

I have a sense that any new standards process would have to work hard to defend against a number of distractions, such as forcing character encoding into the standard, the lure of JVM style compile once run anywhere (which has destroyed the ability for Clojure to bootstrap without maven), or attempts to reduce or remove certain warts that some users find annoying.

Many who don't know the history don't realize that the standard as it exists encoded the separate practices of many disparate groups into a single standard. This can be confusing for newcomers because they may encounter ways of doing things that are archaic but still in the language. Part of the challenge thus is not getting rid of things, but somehow figuring out how to communicate what have come to be understood as the better ways of doing things --- all without changing those parts of the standard.

I'm sure that the old lispers have fairly extensive lists of fundamental issues and improvements.

The short list that I have been able to compile from chance encounters with mentions of shortcomings includes implementation details, such as lexical O(1) jump tables, delimited continuations, proper tail call elimination, threads, extending dispatch macro syntax, and default precision for reading floats, and issues that almost certainly need more research and exploration, such as finding a solution to the CLOS action at a distance issue, CLOS dispatch on parametric types, something about static types, ways to alloc without gc, and many more things I am completely unaware of.

The Racket ecosystem has done quite a bit of research and implementation in many of these areas, but Racket lies at the far end of the lisp family from common lisp. Finding a way to take the more static and text based ideas from the ML/Racket end of the spectrum and make them accessible at the the dynamic image based Smalltalk/Interlisp end seems like something that could happen within the common lisp space more readily than elsewhere. There seem to be some fundamental questions about how late binding interacts with many other language features that do not have satisfactory answers.

All of this being said, I don't see the motivation for anyone in the space to spend the time on another standards process while so many ideas haven't been explored.

tgbugs··on SICL: A New Common Lisp Implementation
In case anyone is wondering what step 3 in the readme means, this is what I used when I was exploring SICL back in March.

  (asdf:initialize-source-registry
   '(:source-registry
     (:tree #p"/path/to/SICL/")
     :inherit-configuration))
tgbugs··on Common data model mistakes made by startups
Hard deletes are also awful from the perspective of data preservation. For example, when youtube removes a video they also delete all the metadata or any indication that it ever existed. Countless people have lost what they thought was a secure record of at least the title of songs or videos they saved to a playlist.

There is also a more sinister side, which is that the ability to hard delete something forever means that bad actors can fabricate old "deleted" documents and accuse someone of having created and then deleted them.

tgbugs··on Legalese – Computational Law
Roughly the same way I feel about HIPPA certification. Except that there is an additional interaction of building codes with zoning laws that leads to inflated housing prices in the US. Also, if you want to know if the house was built to code, you have to inspect it yourself, or at least be on site frequently enough to keep the trash from being stashed in the dead space.

There is of course a separation between safety and standards that is hard for laws and codes to grasp. The separation between intention and results is one of the reasons why you want these things explicitly defined as part of a formal specification for a law, it makes it possible to determine whether it is having the desired effect, and if it is not it could .e.g trigger a clause removing the code from being in effect. Rent controls would be a perfect example for this, though the whole point of my argument in my top level response is that trying to measure whether rent control is effective is the hard part (every serious study of rent controls shows that they are not).

tgbugs··on Legalese – Computational Law
Consider also the disillusionment that a significant portion of the population feels when confronted with the nearly 1000 years of rule based knowledge representation that is common law.

Maybe a good law would be that if you can't be arsed to write the law in a formal language, or can't figure out how, then it shouldn't be a law in the first place :)

tgbugs··on Legalese – Computational Law
The problem here isn't that we are missing a language for expressing laws. Indeed, DSLs for expressing legal axioms don't really exist, and I expect that formal verifiable languages will be created, but even when they are, the problem is not an inability to express laws, the problem is how to map them to entities in the real world.

To give only a single example, "what is admissible evidence?" Assume for a moment that we have solved the "what is admissible evidence in this context" problem. Now you have a giant list of pieces of evidence, testimony, etc. How do you prove chain of custody? The legal system is utterly blind to the world, its agents are intentionally blinded to much of the information, and practically blinded to others.

Even with modern surveillance the amount of uncertainty that is present in many legal cases is substantial, and machine learning won't help you here, because every legal case is a one off. I'm sure someone will try to create a statistical guilt detection net despite the obviousness that it will inevitably produce unjust results.

If you want laws to be written in formal languages, you need a system for defining formal semantics for the entities and relations that their abstracted axioms describe, and things like OWL and RDF are not sufficient, because they only provides the hypotheses, not the evidence. You can assert that evidence subClassOf admissible-evidence, but you need proof that the assertion is valid that can be checked by multiple parties (or anyone with the desire to do so). The one-off nature of many cases often means that you can't go out and test the hypothesis again.

These are hard measurement problems, and we don't have a good language for describing them, and how the measurements themselves are governed by the legals system.

That said, being able to bootstrap simulate court cases with juries drawn from the population would already be a massive win for interrogating laws and their implementation and enforcement. Similarly being able to simulate new legislation to reveal loopholes would be fantastic, though I doubt politicians would enjoy that.

There is quite a bit of data, but I don't think it is particularly amenable to the current fads in machine learning. Amusingly people complain that we'll never be able to write down all the axioms, yet someone wrote down all the laws. Economically unattractive due to a tragedy of the commons certainly, but at the end of the day someone is going to have to write down all the axioms, one way or another. Maybe modern machine learning techniques could be used to generate the axioms for humans to review?

tgbugs··on Overcoming Tab Overload
It is actually worse than that. Many/most implementations of tabs are completely opaque to the window manager so if you want to programmatically control your browser windows e.g. by moving a window with a certain title to the front you can't do it for windows that contain a certain tab unless it is the active tab. Oops.
tgbugs··on How I Practice Piano
The underlying motor sequence is learned and probabilistic based on prior sequences that you actually played. Therefore if you introduce a mistake early it will be one of the set of motor sequences that can occur following the start of the sequence. The faster you go the less time you have to anticipate and correct an upcoming mistake. This can happen at even fairly low speeds as soon as you cross the threshold where you can no longer consciously determine what motor movement happens next in the sequence.

An extremely overly simplistic model would be something like, let's say you play a sequence 10 times and make a single mistake. You now have a 10% chance on any future run of making that mistake. However if you do, then you probability of making that mistake goes up in the future. Assume that you can reduce that 10% slighly by applying top down (conscious) attention if you play at some reasonable pace. Playing faster means that you can't use any top down feedback to reduce the probability of the incorrect sequence from occurring.

Under stress (e.g. during a competition) you may not be able to dedicate enough attention to keep the incorrect sequences in check, so the mistakes reappear at the worst possible time. This also happens if you get angry because you made a mistake.

The best way to avoid this is to never make a mistake in the first place.

tgbugs··on Back in 1993, I was taking a number theory class
If I had to guess the wording of completely insane contracts signed with companies that provide campus wide printing.
← PreviousPage 5 of 9Next →