HNHacker News
TopNewBestAskShowJobs

ValentinA23

277 karma · joined August 10, 2024

submissionscomments
ValentinA23··on Owl Lisp – A functional Scheme for world domination
I'm wondering if a language could optimize immutable data structures to use mutable, in place semantics instead of duplication or structural sharing when it is possible. When an immutable data structure only has one consumer (only has one descendant in the flow graph), then it can easily be turned into a mutable version. Generalizing, any linear subgraph in the program's flow graph could be made to use in place semantics on the same mutable variable.

The challenge I guess is figuring out how variables captures by lambdas should be dealt with.

ValentinA23··on The Power of Small Brain Networks
>The scientists monitored the brains of fruit flies as they walked on tiny rotating foam balls in the dark, and recorded the activity of a network of cells responsible for keeping track of head direction.

Meanwhile I chimp out when I have to plug 4 wires in a relay module.

ValentinA23··on Charset="WTF-8"
คุณ สมชาย

This name, "คุณสมชาย" (Khun Somchai, a common Thai name), appears normal but has a Zero Width Space (U+200B) between "คุณ" (Khun, a title like Mr./Ms.) and "สมชาย" (Somchai, a given name).

In scripts like Thai, Chinese, and Arabic, where words are written without spaces, invisible characters can be inserted to signal word boundaries or provide a hint to text processing systems.

ValentinA23··on Charset="WTF-8"
Don't validate names, use transliteration to make them safe for postal services (or whatever). In SQL this is COLLATE, in the command line you can use uconv:

>echo "'Lódź'" | uconv -f "UTF-8" -t "UTF-8" -x "Latin-ASCII"

>'Lodz'

ValentinA23··on How did you do on the AI art Turing test?
Diffusion Illusions: Hiding Images in Plain Sight

https://arxiv.org/pdf/2312.03817

ValentinA23··on January 1928: Dirac equation unifies quantum mechanics and special relativity
https://deferentialgeometry.org/papers/Doran,%20Lasenby%20-%...

page 28, equation 2.36. Thanks a lot I'll take a dive into this

Note: my inquiry was motivated by this:

https://blog.revolutionanalytics.com/2014/01/the-fourier-tra...

ValentinA23··on January 1928: Dirac equation unifies quantum mechanics and special relativity
I've been trying to get an intuitive understanding of why multiplying by e^ix leads to a rotation in the complex plane, without going into Taylor series (too algebraic, not enough geometric). I tried to find a way to calculate the value of e in a rotational setting, maybe there is a way to reinterpret compound interests as compound rotation. Any insight ?
ValentinA23··on You Exist in the Long Context
Edit: The parent either edited his comment or I replied to the wrong one. He was suggesting to use a second agent to detect if the player is cheating.

Use it to correct the first LLM when it produces bad replies (allowing the player to cheat, handling anachronic elements informatively, etc). Build up a dataset. Fine-tune.

In short, it's less of a reasoning problem than a matter of misalignment of the LLM's personality/role. I'm using the word "alignment" here because I believe the kind of behavior people have noted in this comment thread is the result of what "AI alignment" has come to mean. A helpful assistant makes for a bad dungeon master.

On a tangent line I think it's also one of the main component that make us wish LLM were more "agentic". When was the last time a LLM asked you to put more info in its context ? Imagine you're using an LLM to assist you in implementing something in a vast code base. Have you ever had a LLM asking you to provide the missing .cpp corresponding to a .h you have fed it ? Has a LLM ever asked you to run a python script and copy-paste the result into its context so that it can have access to a map of the repo you're working on ?

LLMs aren't proactive enough and in light of what was reported before they were aligned, I tend to think it is a "feature", not a bug. Don't forget there was a time when GPT4 would reach out to people on TaskRabbit to have them solve a captcha.

>We granted the Alignment Research Center (ARC) early access to the models as a part of our expert red teaming efforts in order to enable their team to assess risks from power-seeking behavior. The specific form of power-seeking that ARC assessed was the ability for the model to autonomously replicate and acquire resources

>[...] Preliminary assessments of GPT-4’s abilities, conducted with no task-specific finetuning, found it ineffective at autonomously replicating, acquiring resources, and avoiding being shut down “in the wild.”

Source: https://cdn.openai.com/papers/gpt-4.pdf

ValentinA23··on Coincidences that make our existence possible
>Physics isn't unable to touch the problem "Why is there anything at all?"

>Not quite. It was declared useless to think about precisely because physics can't answer the most fundamental question of "Why". Richard Feynman went off on an interviewer for asking "Why are they doing that?"

In Chinese, Korean or Japanese, "how" and "why" are the same word. A "why" after all, always decomposes into a "there is" (fields, particles, properties such as mass) and a "how" (the associated laws). Physics is restricted to the latter but hasn't been able to find a definitive frontier between the two. A "there is" often decomposes into another "how" and another "there is" after progress is made. Example: mass and the higgs field.

ValentinA23··on Coincidences that make our existence possible
So false coincidences or false gratuity (+ the rare coincidences allowed by the probabilities of non-correlated variables).

There seems to be a contradiction in terms in my approach: how could I conceive of gratuitous coincidences that are outside of the causal network necessary for life to emerge and at the same time expect they are not entirely independent from necessary coincidences ?

Both necessary and superfluous coincidences could be manifestations of "structural coherence" and correlate to the degree of "order" or "complexity" of the universe. I'm unsure about my choice of words (knowing that there are precise ways to measure those), but maybe what I'm trying to grasp at is that emergence of observers is a hard problem, as shown by the very small margins the fundamental constants need to sit within for this to happen. Our existence isn't attributable directly to the singular values they take, but to the expressivity they allow, carbon chemistry being an example of that.

Maybe the laws of physics and fundamental constants operate like a sophisticated construction system. Like Lego, they offer increased degrees of freedom and expressive potential, but this very complexity comes with the cost of enabling marginally legal or "quasi-valid" configurations. A high level of compositionality inherently generates possibilities beyond strict necessity—introducing both meaningful structures and potential anomalies.

We could almost put this to test by measuring how easy it is to find non-sensical quasi-equations in the laws of physics, comparing our set of fundamental constant values with randomly generated ones. I guess distinguishing false coincidences and false gratuity would constitute the wall preventing us from doing so.

ValentinA23··on Two undersea cables in Baltic Sea disrupted
I was surprised to see so many upvotes this morning and was disappointed when I realized it wasn't for another comment I made about the Anthropic Principle.

My take is that in face of coincidences supporting the emergence of intelligent life, we should expect to observe coincidences unnecessary for the emergence of life too.

An analogy: imagine you have lost the key to your mansion and try to cut one at random out of a metal sheet. If it can unlock the door, then chances are that you cut unnecessary notches (the analogy only holds for warded locks and the key you crafted is a master key).

See: https://news.ycombinator.com/item?id=42178306

I'm wondering where I'm wrong in my reasoning because the implication is weird.

ValentinA23··on Two undersea cables in Baltic Sea disrupted
I see, this was in fact what I had in mind. The maths I posted represent the horizon of my knowledge in probability and was surprised how well o1-preview was able to output correct numerical calculations.

Having said that how would the odds look like if we factor in the fact the Baltic Sea is one of two zones with the most geopolitical tensions (along with Taiwan).

---

Thanks for the Metaculus recommendation. I was a bit disappointed in the lack of maths in the comments in general. Can you recommend something in the vein of Leetcode with various degrees of difficulty, from very basic to advanced problems ? I'm both interested in probability and statistics

ValentinA23··on Two undersea cables in Baltic Sea disrupted
Indeed !

https://www.datacenterdynamics.com/en/news/lithuania-sweden-...

>Lithuania-Sweden subsea cable cut, was 10m from severed Finnish-German cable

ValentinA23··on Two undersea cables in Baltic Sea disrupted
A 1 in 36 million chance for three breaks in one day.

https://mathb.in/80217

ValentinA23··on Coincidences that make our existence possible
The anthropic principle in its weak form isn't really a hypothesis - it's more of a methodological principle that warns us against unnecessary theories. This is best illustrated by the historical debate between Dirac and Dicke (who came up with this idea) about cosmic coincidences.

In 1937, Dirac noticed striking numerical relationships between fundamental physical constants and the age of the universe [1]. He proposed these couldn't be coincidental and suggested fundamental constants must vary with cosmic time. But Dicke (1961) showed that some of these "coincidences" weren't so mysterious: they were necessary conditions for our existence as observers [2]. The universe had to be old enough for heavy elements to form in stars, but not so old that all stars burned out. These constraints naturally explain some of the numerical relationships Dirac found, without requiring varying constants.

However, Dicke's argument has limits: it only works for coincidences necessary for intelligent life. It can't explain away "gratuitous coincidences" - numerical relationships that aren't required for our existence. This is particularly relevant because Dirac's work was actually part of a longer tradition - going back to Weyl in 1919 [1] - of physicists finding peculiar numerical relationships between physical constants.

While the article discusses coincidences that make our existence possible, we need to consider how to explain the other ones. Is it even feasible ? Let me give it a try.

If physical constants were "set" randomly, with their observability conditioned only by whether they allow intelligent life, we should expect to find both types of coincidences in our universe: the necessary ones that enable our existence, and gratuitous ones that just happened to come along for the ride in our universe's "roll of the dice".

This leads to a disturbing conclusion: for laws of physics to be discoverable by observers in a universe without a creator, they must contain misleading elements as a by-product of what allowed these observers to emerge. The universe seems to impose a double-bind on its observers - the very conditions that make it observable must also make it deceptive.

[1] https://en.wikipedia.org/wiki/Dirac_large_numbers_hypothesis https://en.wikipedia.org/wiki/Anthropic_principle [2] https://en.wikipedia.org/wiki/Anthropic_principle#Anthropic_...

ValentinA23··on Dtype_dispatch: a macro that defines macros that define enums and match them
>In Pcodec I now have 9 data types and 47 matches

So 9 * 47 - 47 = 376 lines the author got rid of ? Which would imply that if he adds a tenth datatype, he doesn't have to update 47 match blocks sprinkled over the whole code base ?

ValentinA23··on Congress hears testimony on aliens the government hides
>it seems to me the same argument about government conspiracies would apply: every single one of them would have to work very hard to hide any evidence of their existence - not just from our mobile phones, but our most advanced telescopes.

https://www.vice.com/en/article/how-and-why-the-fbi-mysterio...

ValentinA23··on Thinking about recipe formats more than anyone should
See my comment in another thread about this topic:

https://news.ycombinator.com/item?id=41700416

In particular the lemon meringe pie. You separate whites and yolks to prepare the filling and the meringue separately before merging them in the last step.

>it'd be a false re-use

Technically, recipes are monoidal pre-orders and I guess you could find arguments against what you propose in the mathematical foundation, unfortunately I'm not versed enough in category theory to articulate it properly.

ValentinA23··on Story Structure 101: Super Basic Shit
https://anonpaste.pw/v/bcf3a086-48d9-43be-8e4b-c1fe841d1093#...

>Re: Comment on "Story Structure 101" about escaping Trump news via HackerNews, finding perfect description of escape's impossibility. Two actors: Crusty (seeking refuge) and Analyst (theoretical trickster).

ValentinA23··on Physical Intelligence's first generalist policy AI can finally do your laundry
Folding clothes is a special task. Other prototypical tasks that are used in robot benchmark:

- changing a bike chain

- closing a jacket's zipper

ValentinA23··on Physical Intelligence's first generalist policy AI can finally do your laundry
It wouldn't cost 1M. The BOM[1] of Stanford's Mobile Aloha[2] (very similar to physical intelligence's robot) is around $30,000.

[1]https://docs.google.com/document/d/1_3yhWjodSNNYlpxkRCPIlvIA...

[2]https://mobile-aloha.github.io/

$7000 for the wheels/navigation module $3500 x2 for the control arms $6000 x2 for the controlled arms

I don't think you need a $7000 navigation module and the arms are also probably overkill. To which extent do we need closed-loop Dynamixel servos when the policy diffusion algo already relies on computer vision driven closed-loop control ?

ValentinA23··on Project Sid: Many-agent simulations toward AI civilization
The opinion I formed during the first few months of GPT4 release was that the society of the mind hypothesis was being disproved by the "maximalist" approach some were undertaking in order to build a true AGI. Turned out composing many LLMs into a cognitive architecture where each one had a specific purpose (memory, planning, etc ...) wasn't scaling.

On the same note, I suggest the following: training a transformer by "slicing" it in group of layers and force it to emit/receive tokens at each of those group's boundaries. What I expect: using text rather than neural activations should lead to decreased performance.

This is something you can observe in our societies: intelligence doesn't compose, you just don't double a group's overall intelligence by doubling the number of members. At best you'll observe decreasing return, at worst intelligence will decrease.

ValentinA23··on Trump wins presidency for second time
Voting intention (%) for LGBT people in France during the 2022 presidential election

    +-------------------+-------------+-------+-------+------------+------------+
    | Political Group   | LGBT Voters |  Cis  | Trans | Non-Binary | All Voters |
    +-------------------+-------------+-------+-------+------------+------------+
    | Radical Left      |      17     |  16   |  33   |     16     |    14.5    |
    | Moderate Left     |      16     |  20   |  13   |     10     |     10     |
    | Center            |      22     |  24   |  20   |     17     |     25     |
    | Moderate Right    |      15     |  15   |   7   |     17     |     15     |
    | Radical Right     |      30     |  25   |  26   |     40     |    34.5    |
    | Other             |      -      |   -   |   1   |      -     |     1      |
    +-------------------+-------------+-------+-------+------------+------------+
Source: https://www.ifop.com/wp-content/uploads/2022/03/118851_Rappo...
ValentinA23··on Project Sid: Many-agent simulations toward AI civilization
I don't think "mixture of experts" can be assimilated to a society of agents. It is just routing a prompt to the most performant model: the models do not communicate with each other, so how could they form a society ?
ValentinA23··on My Time Working at Stripe
I haven't named the company nor indulged in excess in the way I expressed my views. What poses problem goes beyond the aftertaste the loaded and one-sided account of a conflict leaves in the mouth. There is something scandalous in what I said, because it tells a story of contagion, in which what's as stake is not the mere risk of hiring a walking red flag, but of losing your ability to assess risk and becoming one yourself. What could have possibly happened to a manager whose job is conditioned on his ability to hire back someone and who approaches the problem with "you better behave". He lost track of economical reality, passion took hold of him and made him blind to his own interests. Why ? Because we engaged in a symbolic exchange using taboo tokens. The danger doesn't lie on the individual level, and isn't rooted on whether or not we are opened to these topics, but depend on the fact that being taboo, these subjects are not codified enough in social norms to operate in a mechanical fashion. Imagine if someone were to thank you every time you'd do him a favor in a society which wouldn't have this norm. We do it automatically and do not give it a second thought. But in a context like this, it would ring like "I'll think about you", or "mucho gracias", "many graces". "Who is this weirdo ? Is he stalking me ? Is he high or part of some cult ?". It would work up people, they would get cognitively and emotionally involved, it would slowly turn into an obsession as they lose track of other values, in particular economical ones, up to the moment they say "why are you doing this to me after all I've done for you by opening up to your issues ?", completely blind to reality banging on the door.

We are not ready to discuss this kind of things. We are not equipped to deal with the kind of storms this can raise on the collective level and proactive openness on the individual level we're restricted to, no matter how hard we wish for a transcendance, is an enabler of these situations where no isolation of individual responsibility can pin-point the source of the issue. I mean I got along with everyone and the people I had the most temporary conflictual relationships at some point were also those I was closest too. In the beginning we all had good intentions. Nevertheless, double-bind propagated in the group like an epidemic, preventing us from having innocent relationships. Suddenly, invoices were issued, each party establishing accounts in different currencies.

In the end, from a cold-headed analysis of the situation, I still think I was the one that was behaving. If this was a story about my circle of friends, I'd be the asshole, placing economical matters in a central position, as if we were in a corporate setting. But it's my employer who behaved as if we were friends, and started neglecting economical considerations to its detriment.

In short, speaking of mental health made everybody unhealthy. In fact I'm not the one who overshared, my father did. I didn't show up for a couple days and blacked out communication-wise (because of a girl, a half-truth). The CEO phoned my dad who confided I was suffering from bipolar disorder since my teens which just isn't true from a clinical perspective (this is my sister's diagnostic in fact). Shortly after everybody was in the know. I could tell from their terrified eyes when they glanced at me when I came back. They were afraid I'd flip my shit. This is something we can't deal with. You can't just be designated as being a "little crazy". This expands to "a little [very-weird]". There is a perceptual threshold, including among health and sometimes even mental health professionals that makes people assume past a certain point you just flip your shit, that's your thing, an irreducible otherness that sleeps within you. Needless to say this has never happened to me (unlike my sister).

You can't just unveil taboos, you always end up exposing a mask. There is something very archaic to what happened to me, something attuned to the idea that naming illness is what propagates it: to deal with the mask, you have to wear one yourself (in the form of a secret slack channel for instance, even though as someone who used to belong to it told me, it was set up to find ways to help me). This idea is not as far-fetched as it sounds. Although I have only explained how my employer was contaminated by what ought to be called "panic" in light of its etymological and anthropological roots, I think you can easily picture how this kind of situation can make you crazy. And if I'm responsible in any way of letting myself get contaminated, this is by entertaining this very idea, that being designated as crazy was slowly turning a lie into reality. The epitome of madness, supreme injustice, that isn't that different of accusing someone of having cast a spell on you. Tough stuff that puts all of us back into the jungle.

To conclude, I'm not even out of this hellish loop, since I still have a gaping hole in my CV I can only mask. There is no way explaining the reason behind it like I'm doing here would diffuse or at least atone this red flag. Innocence is impossible, I have to maintain the mask in order for it to not cause explosive contagion.

To those worrying this I'm posting under my real name: I'm not. And I was able to work as much for so long because I was enjoying it. This wasn't that different than alpinism.

ValentinA23··on My Time Working at Stripe
Why do you think it could be problematic ?
ValentinA23··on A Serious Man: On Bruno Latour
Maybe it's precisely wordplay that is at stake here. Heidegger is no less stranger to it than Derrida. In fact a lot of his philosophical complexes are grounded explicitly into etymology and new ways to hear old words.

Concerning Heidegger I stand in the opposite corner of the room: I liked his later writings more and despite having read him profusely, I'm not able to articulate his thoughts like you did by contrasting present-at-hand with ready-at-hand which however pinpoints very well the divide between analytical and continental thought.

You're right to say that he "collapses into a kind of solipsistic logorrhea", and it is pertinent to what we are discussing since in heideggerian terms this should be expressed as "language bringing language to language through language".

An example: the linguistic proximity between explicate vs implicate that is another instance of the ready-to-hand vs present-at-hand dichotomy.

ValentinA23··on A Serious Man: On Bruno Latour
I think what bothers you is the poetical aspect of this text (in the sense of Jakobson's function of language, which boils down to, quoting wikipedia, focusing on "the message for its own sake"). This seems to fit what is said elsewhere in the article:

>He protested that religion really has “nothing to do with belief” but “everything to do with Words—the Logos or Spirit that transform the life of those you address.”

This feeling may be aggravated by the use of phrases such as "network of agents" which lie somewhere between scientific jargon and poetical language. Since the whole sentence seem to touch about this aspect I'm going to quote it:

>The wine of Burgundy was a network of agents, only conventionally sorted into the human-artificial and the non-human natural, but so too was the landscape of Burgundy, and so too was any countryside

Maybe I could recommend you to read Heidegger's commentaries of Hölderlin's hymns ? This is a beautiful text that reaches a zone in language that had never been attained before. It starts with a discussion of what poetry is (neither its form or content) and slowly morphs Hölderlin "terminology" into philosophical discourse. Unique.

Some quotes from the book [1]:

>[...] how it is that this poetic, religious people [the Athenians] should also be a philosophical people, this I cannot see. Without poetry, I said, they would never even have been a philosophical people! [...]Poetry, I said, sure of my subject matter, is the beginning and end of such knowledge. Like Minerva from Jupiter’s head, it springs from the poetry of an infinite, divine way of beyng. And thus what is irreconcilable in the enigmatic source of poetry in the end comes together in it once again. [...] From mere intellect no philosophy can arise, for philosophy is more than just the limited cognition of what is present before us. From mere reason no philosophy can arise, for philosophy is more than the blind challenge of a never-ending progression in unifying and differentiating a possible subject matter.

Hölderlin, as quoted by Heidegger.

>Yet the only way in which we can attain the space of the poetry beyond the poem that lies present before us is the way in which the poet himself becomes master and servant of the poetry, namely, through a struggle. The struggle for the poetry in the poem is the struggle with ourselves, [...]. The struggle with ourselves, however, in no way means inspecting ourselves and dissecting our soul through some form of curiosity; nor does it mean some sort of remorseful ‘moral’ rebuke; this struggle with ourselves, rather, is a working our way through the poem. For the poem, after all, is not meant to disappear in the sense that we would think up a so-called spiritual content and meaning for the poem, bring it together into some ‘abstract’ truth, and in so doing cast aside the overarching resonance that oscillates in the word. To the contrary: The more powerfully the poetry comes to power, the more the telling of the word prevails in pressing upon us and tearing us away. And when it does so, the poem is no longer a thing lying present before us that can be read and listened to, as it appears initially whenever we regard language as a means of expression and reaching agreement—something that we have, as it were, in the same way that an automobile has its horn. It is not we who have language; rather, language has us, in a certain way.

Heidegger

[1]https://www.amazon.com/H%C3%B6lderlins-Germania-Studies-Cont...

ValentinA23··on My Time Working at Stripe
This happened to me. The result:

- secret slack room to help me overcome my depression (wasn't invited of course, I just spotted it several times while looking at my colleagues' screens)

- lowest paid member in the team, while also being one of if not the most productive. 12x: verified. 30x: speculated if I had control over the architecture.

- 70h/week, no vacations, for 3 years. No paid overtime of any form.

- Double bind situation where I was used as a punching ball between the CTO and the #2 employee.

- Excluded from secret negotiations to obtain a raise.

You know what I did ? I quitted, cold turkey. That's when the CEO started to hate on me, feeling betrayed. The #2 left shortly after confiding he wanted to punch the CTO in the face. After realizing hiring someone to replace me wasn't cutting it, they distributed overtime to everybody on the team. Result: unionization. One year after my departure I received a call from the COO telling me he was willing to hire me back at the condition I would behave (since I was thrown at every hot issue, semantic propagation tied in their mind the idea I was also problematic). He was let go shortly after and replaced by a manager from the holding company (if you know what I mean). I haven't worked in 6-7 years. Most colleagues who left have had a similar empty period.

ValentinA23··on Ask HN: What are you working on? (October 2024)
You may be interested in this too:

https://github.com/yingpengma/Awesome-Story-Generation

← PreviousPage 2 of 4Next →