@khoiiiind

it has just dawned on me

Joined July 2024
The Marie Kondo method of self-therapy: 1. Gather into a heap all the “I must”s and “I can’t”s. 2. Pick them up one by one, see if they spark joy. 3. Those that don’t: 3a. Thank them for their service. 3b. Throw them in the trash 4. The rest: 4a. Substitute “might” for “must”. 4b. Substitute “need not” for “can’t”. Result: a very bearable lightness of being.
13
1,732
k h ô i retweeted
Having said I haven’t been impressed by anything vibe coded, it’s actually a joy to have people happily share their developments here. It’s clearly enabled them to build things THEY’RE HAPPY ABOUT, & that’s an important value not to overlook.
6
2
53
1,186
Zhuangzi on denotational semantics
1
7
1
47
1,196
k h ô i retweeted
Quentin Heath and Dale Miller. 2019. A Proof Theory for Model Checking. J. Autom. Reason. 63, 4 (Dec 2019), 857–885. DOI:10.1007/s10817-018-9475-3 While model checking has often been considered as a practical alternative to building formal proofs, we argue here that the theory of sequent calculus proofs can be used to provide an appealing foundation for model checking. Since the emphasis of model checking is on establishing the truth of a property in a model, we rely on additive inference rules since these provide a natural description of truth values via inference rules. Unfortunately, using these rules alone can force the use of inference rules with an infinite number of premises. In order to accommodate more expressive and finitary inference rules, we also allow multiplicative rules but limit their use to the construction of additive synthetic inference rules: such synthetic rules are described using the proof-theoretic notions of polarization and focused proof systems. This framework provides a natural, proof-theoretic treatment of reachability and non-reachability problems, as well as tabled deduction, bisimulation, and winning strategies. arxiv.org/abs/1701.04915
1
1
16
745
For a fixed artifact, fixed semantics, fixed specification, and fixed TCB, we can prove surprisingly strong things. Rice's theorem doesn't forestall that. But deployed software is a *process*, not an artifact. We continually modify it and compose it with new things.
1
2
18
531
inverse limits are limits and direct limits are not limits but colimits, which are like limits but inverse.
Limits in category theory are also called inverse limits, projective limits, and left roots.
2
4
43
1,579
awakening is an inexact science
5
2
33
861
Yes 100x. If you run a model for weeks on a decidable open problem, you'll see it start to reach for new formalisms in absence of everything else. Like Kummer's ideal numbers in pursuit of Fermat's Last Theorem, who knows what can smart models come up with, if "allowed"?
I cannot agree more. Kevin Buzzard made so many points I agree with. But the best one is this "I thus believe that in the future we will reach a new “natural boundary” in mathematics, beyond (and perhaps way beyond) where we are now, but where machines are going to get stuck and where it is not viable to expend any more resources to make the next big leap. (...) I believe that the optimal thing to do (...) is to let the machines loose, see what happens, and then begin the journey to where they have stopped." xenaproject.wordpress.com/20…
2
6
78
7,613
People are like, “She’s selfish!” All I can say is that it’s easy to be selfish; it’s exceedingly difficult to just do as you please.
favorite reminder for creatives, by Björk
3
41
1,446
As an editor of two prominent mathematics journals and a professor in a large mathematics department at a research-intensive university, I have a bird’s-eye view of how mathematicians are reacting to AI. My view is that the damage currently being done to mathematics is not due to AI, but to mathematicians’ reaction to it. Many of these reactions are self-serving and collectivist, and demonstrate a failure to think clearly about the future. AI is a revolutionary technology unlike anything we have created before. Many claims made by AI sceptics, AI accelerationists and AI doomers alike are poorly evidenced; they pay insufficient attention to liberal epistemology and to the relationship between intelligence and power. AI could enormously expand what humans can know and do. To realize this promise, we need to be open to empirical evidence and be willing to solve difficult problems. I will have an essay coming out tomorrow that goes into some of these ideas in more detail.
27
53
4
376
39,975
tfw you just read Rilke
I need to reevaluate my entire life
12
104
8
932
33,871
A few comments on the set of recommendations: - The introductory paragraph says "we ask them to stop testing advanced mathematical problems on proprietary models." It is not going to happen. I see no strong reason to want it to happen. The authors admit that their recommendations are formulated with the assumption that this will not happen! I think this sentence should not have been there at all. - I strongly agree with the recommendation on initial release (section 2.B. Step I) except for the sentence "We strongly recommend that AI labs refrain from treating the release of mathematical results as marketing vehicles to promote their models, ignoring the substantial negative externalities that this practice inflicts on the mathematical community". - I think that Step II goes too far in parts. "One of our principles is that AI labs have a responsibility to provide support, including funding, for the development of human understanding of the AI mathematical output that they release." What does responsibility mean here? Does it equate to an ethical requirement? I don't think such a requirement exists. I would have preferred it simply stated this as "We recommend that AI labs should.." (This is one of several places where the document could have been likely improved by greater political diversity among the members -- I suspect most authors are on the economic left) -- Apart from these, the ideas on dissemination seem generally sound.
The advisory group on mathematics and artificial intelligence has just published a set of recommendations concerning the responsible release of mathematical results generated by AI companies using internal models. 1/3 agmai.org/general-sep29/
9
3
1
45
13,928
Narcissus never realizes what captivates and torments him so is but a reflection of his own self. A solipsist does: all is self, all is this self, tat tvam asi. Solipsism is narcissism enlightened.
are psychopaths and narcissists more likely to be solipsists?
3
8
529
I am running a formal methods hackathon on November 1st, with a bunch of my friends. Contestants will compete to build normie software -- video games, AI agents, travel booking tools, etc. -- but formally verified. We are going to measure to what degree these tools are usable by SWEs with no prior FM background and no FM-specific education. This is a super exciting opportunity to see where the ergonomic gaps are that we, the FM community, need to fill in order to take advantage of our moment in the zeitgeist. If you want to get involved, please DM me! In particular, I think I've got location/food/tokens/photography all covered, but I need some sponsors for the prizes. If you make a cool gizmo or product we can give away and you're down to donate one, please do so!
20
29
7
254
19,563
Maybe I’m wrong, but I believe dependent types are a great fit for low-level programming because correctness there is relatively stable and easy to express formally. High-level features are different beasts. How do you formalize something like “Send no more than three notifications per day, except for urgent events”, let alone verify it? Things close to the business end are unstable in both meaning and priority. There is a case to be made for decoupling types and properties here. We can’t capture all the invariants we want, but we can choose to put some of what we can in types, some in properties, and we can leverage many existing tools to give us sufficient assurance. This is not to say dependent types are not worth it at high levels, though. I just think they need a friend in a separate proof system, ideally built into the same language.
Replying to @statusfailed
I disagree with both of these claims. You can have what is essentially raw assembly + a type system. I have friend working on intrinsically typed / correct by construction compiler optimization and being able to write correct by construction MIPS is part of this.
10
7
2
74
4,703
Relations are lindy
1
1
5
325
k h ô i retweeted
I don’t think HoTT implies the “propositions as (some) types” perspective. In this article Shulman (I think convincingly) argues that we should consider all HoTT types as “propositions” golem.ph.utexas.edu/category…
1
2
16
678
Everyone working on lovable competitors in Summer 2024 litigated the file system vs db debate at great length. File systems won because the agents were already RLd that way. In a future where “intelligence” is abundant, cheap, local, and trainable, I think @hillelogram is correct that we’d want to revisit these debates. In particular, an AST is awfully schemafiable. Why save files as a construct at all even in the db? Why not have code native abstractions totally disconnected from human usability or the training data, and then synthetically train accordingly? I’ve written a bit before about some related ideas here lesswrong.com/posts/CeRHyeot…
Replying to @josevalim
I disagree that debuggers will be obsolete: we often want to give agents faster feedback, and letting them step through and reverse a program state seems like it would be very useful for that. In the same vein, I see a lot of opportunity for languages with configurable runtime environments. Imagine a program that could hand off all random() calls to a deterministic supervisor, or a automatically stub all HTTP requests. This would make it much easier for an agent to explore all of the program's properties. Moonshot crazy idea: if we don't need to write code by hand, there's no reason to use text files. The source code could instead be in a SQL database. What weird capabilities would that open up?
2
1
15
1,497
k h ô i retweeted
What does the rapid success of Jev say about Agent Frameworks?? Jev is a fast and cheap classifier: a type of model that outputs NOT text, but a single number, which shows not only massive cost and latency advantages, but also accuracy advantages on the subset of tasks that it can be used for (bye-bye auto-regression!). The success of Jev indicates that historical and popular 'agent frameworks' that rely on YAML or JSON objects to specify certain parameters, ranging from model choice to role prompt and tools, are increasingly obsolete. The future brings with it a diversity of specialized models that excel at specific parts of agentic work. A diversity of specialized models implies specialization in the architecture of agents, which requires flexibility in the 'agent loop'--the harness that leverages models to drive them toward success. There is no way to encode agents as YAML or JSON properties. A custom agent or harness must necessarily be written in a general-purpose programming language, which has boundless capacity for specialization and customization at every layer. Every good agent framework is going to evolve to be code-first; the rest will be gone within a year.
10
6
35
3,631
I have friends who are hunters, I have been on a deer hunt once (there was no deer,just sat for hours) I can't respect it at all. There is nothing noble about using advanced tech to kill animals in nature for pleasure.
Kristi Noem Trumps former head of Homeland Security is back killing animals again. Here she is posing with an elk she killed with a bow
9
4
38
1,552
No, but listen, you know this "put the fries in the bag, Terence Tao"'--*sniff, sniff*--already the usual replies concede too much. "If Tao is serving fries, YOU will be in the lithium mines." My God! This is our defense of human dignity? Don’t worry, the correct people will get the terrible jobs! *Pulls at shirt* You see, everything is up for radical transformation, human intelligence, consciousness, and so on and so on--except that somebody must still have a shitty afternoon so I can enjoy my lunch. But to say "these people are envious", this is too easy. It allows us, "the educated ones," our own little fantasy: we are disliked only because we are so wonderful. You know, the more interesting thing is this *sniff* peculiar identification with the machine. When the mathematician discovers something, apparently this was HIS achievement, his privilege, his little secret. When the machine discovers something, suddenly it's "Look what WE can do!" The achievement has become collective precisely at the moment when no human being can claim credit. See how this "we" functions. For the achievement, I identify with the machine. For the consequences, I identify YOU with the human. "We have surpassed you." Who are we? Well, myself and the thing that has also surpassed me. There is something almost touching here, really. To escape the humiliation of somebody absorbed in something opaque to you, you appeal to something which knows incomparably more than either of you. And this is supposed to settle the matter in YOUR favor. You know the old example of canned laughter. Here we have a similar arrangement. The machine achieves something on my behalf. The mathematician suffers the loss of human significance on my behalf. Somebody to be brilliant for me and somebody to be obsolete for me, without me having to be either, so I can remain as I was. But *pulls on shirt* there is this little difficulty that hides the essential thing: you still need a response from Tao! A correct proof, this Lean slop and so on, is not enough. You need him to certify his own defeat. You need him to say "By god, this is extraordinary! Yes, this is correct, but also this is interesting." Which means--and here comes Hegel--the person you are reducing to an obedient instrument must remain more than an obedient instrument. You need recognition from someone whose authority you are abolishing. "Your so-called expertise is worthless. Now tell me, in your expert opinion, that I, qua machine representative, was right." *Touches nose* But I would go a step further. What does this math person actually do to provoke you? In the fantasy, I mean. He need not insult you. He can be generous, mild-mannered, whatever. Something remains irritating: there is something over there which occupies him, which matters to him, and your opinion does not settle what it is worth. Perhaps he is not even looking down on you. This is the really intolerable possibility. This, you see, is closer to the Lacanian problem of the Other's desire. Not simply "What does he have that I don't?" but "What the hell is going on with him?" Perhaps the answer is: he is thinking about something else. The fries solve this beautifully. Now you know what you are for him. He needs the money. Your presence creates an obligation. And this, at last, also explains why the machine can be less threatening than the mathematician. The machine's intelligence may exceed yours unimaginably, but you imagine it arrives in response to your prompt. You can complain about the tone. You have the privileges of an angry customer and so on and so on. The human has this irritating possibility of being interested in something you did not request and cannot order. And please, this includes the person already serving the fries! *Pulls at shirt* I am almost tempted to say: the dream of unlimited intelligence conceals just this: a the wish that nobody should be doing anything you cannot immediately understand as a service.
70
220
41
1,478
107,816