Things to think through

(Obvious as it sounds, I stop and consciously make myself do this sometimes and it often works.)

https://twitter.com/sigfpe/status/1125857019884269569

One realization was that my parents are also not best at every possible reshare and made mistakes and bad decisions

https://ru.wikipedia.org/wiki/%D0%9A%D0%B2%D0%B0%D0%BD%D1%82%D0%BE%D0%B2%D0%BE%D0%B5_%D1%81%D0%B0%D0%BC%D0%BE%D1%83%D0%B1%D0%B8%D0%B9%D1%81%D1%82%D0%B2%D0%BE

Риск того, что в результате эксперимента ружьё выстрелит и участник умрёт, составляет 50 %. Если копенгагенская интерпретация верна, то ружьё в конечном итоге выстрелит, и участник умрёт.
Если же верна многомировая интерпретация Эверетта, то в результате каждого проведенного эксперимента вселенная расщепляется на две вселенных, в одной из которых участник остается жив, а в другой погибает.
В мирах, где участник умирает, он перестает существовать. Напротив, с точки зрения неумершего участника, эксперимент будет продолжаться, не приводя к исчезновению участника.
Это происходит потому, что в любом ответвлении участник способен наблюдать результат эксперимента лишь в том мире, в котором он выживает. И если многомировая интерпретация верна, то участник может заметить, что он никогда не погибнет в ходе эксперимента.

бля

https://lobste.rs/s/rpqzek/simulating_physical_reality_with

Asiking since I feel like this particular example doesn’t answer yet “why bother using quantum computing for that”? In essence, here we just multiply initial state by the evolution matrix, which is a trivial task for classical computers (when the size isn’t exponential).
For instance, when justifying use for Shor’s or Grover’s algorithm it’s easy to understand the speedup since the problems are fairly abstract and it’s clear how compexity grows with size. Wonder if in a similar manner, it’s possible to come up with some made-up system of N electrons (or other particles, if it would make Hamiltonian simpler/closer to reality), that would be relatively easy to solve analytically (or just reason about qualitatevely), but straighforward construction of Hamiltonian matrix would be infeasible? And then once could map the Hamiltonian onto quantum gates and that demonstrate that the results predicated by quantum computer match the analytical solution.

https://longitudinal.blog/co2-series-part-1-review-of-basics/
I could try and make this interactive?

• text aspect (pipes)
  
• spreadsheet aspect (??? map/reduce/sql)
  
• timeseries aspect
  
• tree aspect (xpath queries)
  

e.g. you put it in an integer and you can see how it propagates and impacts throught the systtem
or, you put it in a document, and you can see the further changes, also where it's published and how it spreads
kinda like CT scan

e.g. datahoarder subreddit – people literally collect data without much recollection for what they are gonna do with it
or degoogle – people are willing to get off google… to apple etc

every time you change something, it gradually propagates through VFS
unclear how to make it PL independent

Eventually you'd get extremely fat just by pure chance?

Clemens Nylandsted Klokmose
jaredwindover
    @karlicoss I think this talk is highly relevant: https://www.youtube.com/watch?v=ntaudUum06E especially the part about digital substrates. I think we have the same issue as soon as we step outside the browser, and it's a fundamental problem in the way we have imagined interfaces should be created, distributed and used.
For desktop apps check out this paper by my friend James https://perso.telecom-paristech.fr/eagan/media/papers/eagan11.pdf

Re: “macroscopic” properties, this seems to me to be closely related to a concept in the natural sciences call “renormalization group flow”, and possibly some ideas from cybernetics.
One could imagine the space of states “flowing” through an app, and the macroscopic ones are those which are the most stable under perturbations (and therefore most ideal for getting strong understanding and building structure on top of). IMHO, the problem with conventional CS is how much it’s digital i.e. conditionals and cases and expects exact states, whereas good systems (to model the real world) might look much more analog in how they can marshal a variety of states into a set of well-defined ones.

An interesting example is the brush. A brush is "smart": it can help you clean a surface without you having to care about the precise shape of the surface.

http://play.quantumgame.io

nothing obvious on github
tool in study.org

Two cases: there is greater mind that will stop nature from falling apart. Two cases: we intervene or we're not

If there is a greater mind: we're fine anyway, they will save us.
It there is no: we're at risk of nature and us extincting if we don't do anything.
So it makes more sense to do?

https://hyp.is/VJzbZCAyEemveyukcnlBBQ/cognitivemedium.com/emm/emm.html

@kenpex Technically not a visualization but I was very happy with outputting Geiger counter sounds in proportion to how many memory allocations were made. Gave great feedback on when an application was doing something dumb when it comes to allocations

https://twitter.com/laserallan/status/1159571592332087296

https://opensource.com/article/19/1/open-source-voice-interfaces

Mycroft AI uses two intent parsers. The first, Adapt, uses a keyword-matching approach to determine a confidence score, then passes control to the skill, or command, with the highest confidence. Padatious takes a different approach, where examples of entities are provided so it can learn to recognize an entity within an utterance.

huh, could use that for my org mode parsing?

https://nglgzz.com/

What is "build"?
Taking some raw materials and creating something with them. Yes, I consider bytes raw materials.

Users weighted by presence in federated timeline
Stats on toot likes/mutes

api allows retrieving messages before certain timestamp, from newer to older
can you prove that it's not possible to assemble data in some memory efficient and consistent way?

e.g. if we had oldest to newest, it'd be possible to 'extend' both ways, thus persisting middle bit
with one way we have to keep track of and merge 'segments'

I'd never imagine I'd be coming up with theoretical puzzles involving APIs… but here I am.

Say you've got an API serving messages (say, <= 10 at a time) from newest to oldest.

In addition you've got optional 'before' parameter:

• fetch() – gives 10 latest messages (TODO this one isn't even necessary?)
  
• fetch(before='31 Dec 2019 23:59:59') – gives 10 latest (ish) messages in 2019
  

Now imagine you want to persist all messages in the database and TODO properties

• assumptions on resiliency – say you experience cosmic particles every minute that reboots your computer. So you can't keep messages in RAM/database journal for too long
  
• assumptions on data transfer?
  
• assumptions on RAM – not sure if necessary, basically bounded by ram/s ans seconds of uptimes?
  
• assumptions on catching up – i.e. you want to progress in terms of ??? e.g. for each n, amount of messages <n unavailable decreases with time?
  

If API was different (TODO maybe even mentioning newet to oldest isn't important). can always sort.

• fetch(after=XXX)
  
• fetch(before=XXX)
  

allows for fairly easy

[oldest … newest] – persistent in db

loop:
oldest, newest = query_db()
extra_old = fetch(before=oldest)
with db.transaction():
db.insert(extra_old)
extra_new = fetch(after=newest)
with db.transaction():
db.insert(extra_new)

TODO in fact, even easier. if we only had after (assuming we could use 0 timestamp), then

loop:
newest = query_db() # always assumes
if newest is None:
newest = 0
extra_new = fetch(after=newest)
with db.transaction():
db.insert(extra_new)

the question is – what's fundamentally wrong with 'before' api that makes it so hard? am I onto something deeper?

it collects properties and force evaluates them

class Error(Exception):
    def __init__(self, raw: Dict) -> None:
        super().__init__(f'error while processing {raw}')
        self.raw = raw

    @property
    def uid(self) -> str:
        return self.raw['hash']


# def get_entries() -> List[Result]:
#     return list(sort_res_by(iter_entries(), key=lambda e: (e.created, e.url)))
#
#
# def get_ok_entries() -> List[Bookmark]:
#     logger = get_logger()
#     results = []
#     for x in get_entries():
#         try:
#             res = unwrap(x)
#         except Exception as e:
#             logger.exception(e)
#         else:
#             results.append(res)
#     return results

# TODO motivation for having historic backups: can keep track of changes (if you're into that sort of stats)
# TODO why data backups are hard: defensive parsing so it wouldn't require your attention immediately?
# TODO error attribute?


# alternatives:
# fill fields with dummy ids/etc/ and pass eror=Exception
# error: Optional[]
# might still break some invariants (e.g TODO friends??)
# benefit is that you don't have to do anything special and users code wouldn't fail
# downside is that it's easy to miss errors?

# returning Union[Result, Exception]
# downside: no standard method of processing such things in python
# if user forgets to handle Exception, they would end up with more exceptions which is arguably more annoying?
# on the other hand, if you got exception and
# the only minor annoyance is mypy?
# annoying to force user to han
# upside is that static checkers assist you with that (e.g. isinstance(x, Exception))
# TODO could also export Exception/Error type?

# best of two worlds?
# 'strict' -- throw all errors? requires assistance from TODO; error=None
# 'defensive' -- sets up error=attribute?
# 'return' -- TODO return exceptions?

# TODO defensive

Run custom command so it could support grep, ripgrep or whatever

@hillelogram had similar thoughts lately; on the other hand implementing right bindings/keyboard navigation patterns is quite hard from the developer's perspective. I wish there was something like html DOM for desktop/phone applications, then 3rd party devs could enhance existing software

https://twitter.com/karlicoss/status/1174839964472872961

E.g. you could compensate with code for something you couldn't do in dsl

e.g. strip off unnecessary images and javascript , make CSS denser and more informative, etc

maybe there is no sunk cost here.

• writing up something: X days
  
• polishing for othe people: 2X days (feels like comparable number of days!! if not more!)
  
• documenting: X days
  

def split(it: Iterator[X], predicate: X -> bool) -> (Iterator[X], Iterator[X]):
passed = slot()
failed = slot()
for i in it:
if predicate(x):
yield into passed
else:
yield into failed
return passed, failed

not sure if that's very self consistent…
e.g. when we iterate over returned passed – how to iteratoe over failed as well
although could just treat
but how to transform this syntactically?

Try to find an answer why? Why do I want to remove all this physical routine from my life? What do I do instead?

The physical objects around us have aspects, e.g. size, weight, temperature, fragility, stiffness.
They are very macroscopic, continuous and stable properties, so easy to experiment with and reason about.
That's also why it is somewhat flexible and possible combine and use in novel ways.

Grimes
Kedr

After hearing about friends having great experiences treating GPT-3 as a therapist I tried it myself and 5 minutes in oh my god it goes straight for the throat I feel so seen

(1 0) – spin up, (0 1) – spin down
quantum gates – a thing can only evolve according to schrodinger's equation
(analogy to ordinary gates?)
δ t – gate only acts over Hamiltonian (???), it means inherent error?

https://twitter.com/karlicoss/status/1195858695692980225

chrome addon to display div ids?

• dependent types/property testing for verifying lenses
  
• supply javascript code for migration, and send as part of data. execute in a sandbox. kind of like event streams??
  
• even if you can't squeeze everything into that paradigm, maybe 80% would be a great improvement.
  kind of like ADTs are awesome and even though not everything can be ADT, representing most of your data as such is a massive improvement
  

from functools import lru_cache
class lazy:
    def __init__(self, factory):
        self.factory = factory

    @lru_cache(1)
    def _inst(self):
        return self.factory()

    def __getattribute__(self, key):
        print(key)
        # return getattr(self._inst(), key, value)


l = lazy(lambda: {'a': 'xxx'})

l.test
len(l)

print(getattr(l, '__iter__'))

for x in l:
    print(x)

simulation bugs must be fucking annoying – in real programs program states are often very hacky and arbitrary even from the outside, let alone how hard would it be to grasp from within inside

is it moral to jump on a spacecraft near the speed of light and come back when infinite life extension is possible?
the dilemma is – someone got to. if everyone jumps on the ships to cheat death, they'd all have the same flow of time relative to each other
in a way, fastforwarding time is a privilege?

hmm so I guess your body could be 10^9m long and it wouldn't matter because you can only see past signals anyway (and you'll cross the event horizon faster?) basically your body is your lightcone, so the side doesn't matter, and the mind is pointlike (in GR you either have the same observer, or different, and it's kind of discrete in that sense)

I wonder what it means for a distributed system of neurons as it's passing through the events horizon though? Although it'll have a local update limit (irrelevant to speed of light, more of signal propagation rate), wonder what it would sense
https://twitter.com/karlicoss/status/1313417602618789888

would be sort of cool for 'parallel' processing of heterogenous functions etc, dunno

wonder if this could be achieved via ML?
training set – plaintext, result is a tree, without any special meaning, just with tokens? or transformed tokens

https://twitter.com/karlicoss/status/1211457892592820224

so I wonder if it should be an external tool. Kinda like "tail -f" + watch, but aware of certain keywords (e.g. DEBUG/INFO/ERROR in my case)

you still can make a computer and try to program it, but you can't achieve anything without a community

Could check it on examples from this dragon slaying game

Portia uses Scrapy project and enable users to make annotations in webpages. Then users can do crawlers for these webpages.

hmm, interesting, could use that for RSS extractor thingy? although common ancestor thing might be easier:
basically go up until you meet

Hm. I guess money, to an extent, is a motivation for unpleasant work?
E.g. once we solve basic needs with UBI, someone will have to for example, assemble computers.. or program them (also kind of unpleasant work?)

Agree it would be way better off, but I'm not sure that 'profit motive' will go away. Money is not just luxury, it's also a distributed computation of the utility (somewhat flawed, sometimes very flawed, but still), and also a means of getting resources.

people who want open, interoperable and malleable worlds while potentially trading some security
people who want (need?) to be locked in and nannied by the big corporations etc