Lorem Ipsum: Link Types
Link type/link icon subset of /lorem
Internal Page Links
List of links:
-
Link to
# Columns(arrow should point up when transcluded onto/lorem, or pilcrow otherwise). -
Link to
# Unicode Characters(arrow should point down, etc). -
Link to Matt Lakeman (point up); link to Matt Lakeman’s Peep Show (up)
Links inline: Link to # Dropcaps (up); link to # Link Icons (down).
Link Types
-
Interwiki: Emperor Palpatine; Chapter 17, Tao Teh Ching; ‘rimy’; (Monoid a) => a -> a -> a; Barracuda; Tail recursion
-
Annotations:
-
Automatic: “ReBeL: Combining Deep Reinforcement Learning and Search for Imperfect-Information Games” (Arxiv API)
-
Manual: “Young adult outcomes associated with lower cognitive functioning in childhood related to iron–fortified formula in infancy”; long URL: 1976
-
Other: video popups; image popups:
GPT-2-1.5b had a cross-entropy WebText validation loss of ~3.3 (based on the perplexity of ~10 in Figure 4, and log2(10) = 3.32). GPT-3 halved that loss to ~1.73 judging from et al 2020 and using the scaling formula (2.57 × (3.64 × 103)−0.048). For a hypothetical GPT-4, if the scaling curve continues for another 3 orders or so of compute (100–1000×) before crossing over and hitting harder diminishing returns, the cross-entropy loss will drop to ~1.24 (2.57 × (3.64 × (103 × 103))−0.048).
-
-
Cross-page: “Killing Rabbits” (whole page)
GPT-3 roleplaying (to a specific
<span>)
Link Icons
Link icons should follow the priority of: file type < domain < organization.
For example, a DeepMind paper hosted on Gwern.net should be assigned the DeepMind icon, not the Gwern.net ‘𝔊’ icon nor a PDF icon. This is to make link icons as informative as possible: knowing a link is by or about DeepMind is far more informative—the reader instantly knows it involves deep learning, and almost certainly deep reinforcement learning, of DeepMind’s particular house-style—than knowing that knowing it’s a file which happens to be hosted on some domain (gwern.net), which is itself (usually) more informative than knowing it’s just some sort of PDF document (at who knows where by who knows whom about who knows what).
To display all link colors simultaneously, paste into JS browser console: Array.from(document.styleSheets).forEach(styleSheet => { Array.from(styleSheet.cssRules).forEach(rule => { if (rule.selectorText > "" && rule.selectorText.includes('[data-link-icon-color]')) rule.selectorText = rule.selectorText.replaceAll(":hover", ""); }); });
Most recently-added link-icon test-cases first:
Color-Only
Icon + Color
-
IDEAS/RePEc (econ preprints like NBER)
-
ABC (American Broadcasting Company)/ABC (Australian Broadcasting Company)
-
Fandom.com (formerly: Wikia)
-
Stack Exchange:
-
SSC:
-
NGE:
-
Mailing lists:
-
Google; Google Brain: et al 2017, et al 2020; “Profiling a warehouse-scale computer”, et al 2015
-
Facebook:
-
DeepMind:
-
Boorus: Danbooru, Derpibooru, Safebooru (brown)
-
Amazon:
-
Video/Music:
Icon-Only
-
Gwern.net:
-
TXDNE:
-
Wikipedia:
-
PLOS:
-
OpenAI:
-
Internet archives:
-
archive.*: skipped because they are targeted for removal
File Type
-
Document:
-
Image:
-
Code:
-
Archive/Binary:
-
Video:
-
Audio:
WP
WP interwiki links:
-
Should popup (as dynamic content using WP API): Small caps, Talk:Small caps, User:Gwern, User talk:Gwern, Help:Authority control, Help talk:Authority control, MediaWiki:Citethispage-content, Portal:Current events; slash tests: Bouba/kiki_effect, Fate/stay_night, Fate/stay night: Unlimited Blade Works (film); section link: Chernobyl § History
-
Should popup (as a live link?): Category:Buddhism and sports, File:NASA Worm logo.svg, Wikipedia:Wikipedia Signpost, Wikipedia talk:Wikipedia Signpost
-
Should not popup at all: Special:Random, Special:BookSources/0-8054-2836-4
Examples of heavy WP link usage:
Unfortunately, I know of no particularly comprehensive lists of examples of mis-specified rewards/unexpectedly bad proxy objective functions/“reward hacking”/“wireheading”/“perverse instantiation”. Getting into more general economic, behavioral, or human situations would be going too far afield, but the relevant analogues are “principal-agent problem”, “perverse incentives”, “law of unintended consequences”, “Lucas critique”, “Goodhart’s law”, or “Campbell’s law”; such alignment problems are only partially dealt with by having ground-truth evolutionary ‘outer’ losses, and avoiding reward hacking remains an open problem (even in theory). Speedrun gaming communities frequently provide examples of reward-hacking, particularly when games are finished faster by exploiting bugs to sequence break; particularly esoteric techniques require outright hacking the “weird machines” present in many games/devices—for example, pannenkoek2012’s ‘parallel universes’ Super Mario 64 hack which avoids using any jumps by exploiting an integer overflow bug & modulo wraparound to accelerate Mario to near-infinite speed, passing through the entire map multiple times, in order to stop at the right place. Perhaps people can make suggestions, but a few examples I have found or recall include…
Guerrilla warfare’s effectiveness is its own topic; we can note that many of the same cognitive biases like the availability heuristic that skew our beliefs on terrorism also apply to guerrilla warfare as well—everyone remembers the successful American Revolution, but who ever invokes the scores or hundreds of other revolts & failed revolutions in the British Empire which involved guerrilla tactics? (Or in the American empire, for that matter—eg. Shays’ Rebellion, the Whiskey rebellion, or Nat Turner? How well did the American South succeed in seceding, in a conflict with quite as many irregular forces as the American Revolution?) Does a close examination of the Vietnam War, where the much-heralded Vietcong were destroyed after the Tet Offensive and before the North Vietnamese army crushed the ARVN and conquered South Vietnam, reveal it to have been more effective than conventional warfare? A cursory look through any somewhat comprehensive list of guerrilla movements does not reveal it to be a list of luminaries. “Nobody likes a loser”, least of all in war. But to return to terrorism.
Many configuration or special-purpose languages or tools or complicated games turn out to violate the Rule of least power & be “accidentally Turing-complete”, like Sendmail1, MediaWiki templates, sed or repeated regexp/find-replace commands in an editor (any form of string substitution or templating or compile-time computation is highly likely to be TC on its own or when iterated since they often turn out to support a lambda calculus or a term-rewriting language or tag system eg. esolangs “/// ” or Thue), XSLT, Infinite Minesweeper, Dwarf Fortress2, Starcraft, Minecraft, Ant, Transport Tycoon & Cities: Skyline, C++ templates & Java generics, RNA/DNA computing etc are TC but these are not surprising either: many games support scripting (ie. TC-ness) to make their development easier and enable fan modifications, so games’ TC may be as simple as including syntax for calling out to a better-known language like Perl, or it may just be an obscure part of a standard format (most people these days are probably unaware that TrueType & many fonts are PostScript programs based on stack machines, similar to DWARF debugging and ELF metadata, or that some music formats go beyond MIDI in providing scripting capabilities and must be interpreted to be displayed; once one knows this, then fonts being TC are no more surprising than TeX documents being TC, leading of course, to many severe & fascinating font or media security vulnerabilities such as the BLEND vulnerability or SNES & NES code exploiting Linux systems. Other formats, like PDF, are simply appalling.).
Live Link Testcases
Automatically-generated testcases for new live link popups:
-
The baroque complexity of Sendmail possibly contributed to its infamous reputation for insecurity—it was one of the exploit vectors of the Morris worm, and for years shipped with a remote root backdoor (
WIZ).↩︎ -
Dwarf Fortress provides clockwork mechanisms, so TC is unsurprising; but the water is implemented as a simple cellular automation, so there might be more ways of getting TC in DF! The DF wiki currently lists 4 potential ways of creating logic gates: the fluids, the clockwork mechanisms, mine-carts, and creature/animal logic gates involving doors+pressure-sensors.↩︎