xpe

"A witty saying proves nothing." ― Voltaire 1767

Tytler's quote is trying to say too much. It might be acceptable as historical commentary, but it carries little weight to me; it seems overly confident about what the future might hold.*

Tytler died in 1813. We have learned much since then: much about human nature, institutions, experimentation, statistics, evidence, constructing good theories, and governance.** Sure, the quote is worth some reflection; it has grains of truth, but it should not be given undue weight.

* I am not saying "we can predict nothing"! Far from it. I am ok with predictions (even bold ones) to the extent they are deeply rooted in the best understandings and models we have available.

** I'm talking about what motivated people figure out through careful reasoning and evidence, not simply how the median person funnels information from their ears to their mouth. And I'm certainly not commending the effort and thought that the median person puts into stewarding their democracy (if they have one). While we (in the USA, for the time being?) have something like one.

CEL looks interesting and useful, though it isn't common nor familiar imo (not for me at least). Quoting from https://github.com/google/cel-spec

    # Common Expression Language

    The Common Expression Language (CEL) implements common
    semantics for expression evaluation, enabling different
    applications to more easily interoperate.

    ## Key Applications

    - Security policy: organizations have complex infrastructure
      and need common tooling to reason about the system as a whole
    - Protocols: expressions are a useful data type and require
      interoperability across programming languages and platforms.

Not "new" but hopefully useful. Some earlier discussions:

    https://news.ycombinator.com/item?id=28259104 (2021)
    https://news.ycombinator.com/item?id=29363054 (2021)
    https://news.ycombinator.com/item?id=38039616 (2023)
    https://news.ycombinator.com/item?id=42550706 (2024)
    https://news.ycombinator.com/item?id=46744354 (2026)

> Fun fact: Palantir is powered entirely by Claude

Charitably, this is ambiguous. What does the commenter mean exactly by "entirely"?

> All things equal, you’d be better off not exposing yourself to risk of financial harm or other punitive measures.

This isn't necessarily true. This is a complex decision; the logic above frames the decision narrowly, with a short-term time horizon. This kind of decision calls for game theory, not merely an individualistic calculus. Appeasing Trump isn't a winning strategy in the long-run. History shows that cooperation (e.g. pushing back) against authoritarianism is often a better strategy. Consumers may reward companies that behave well. Bottom line: you have to game it out -- no one commenting here has done that, I'll bet. So until someone has ... stay agnostic analytically.

> and incompatible with grep syntax, which makes it useless to most system admins

I wonder how much the above reflects a dated and/or stereotyped view? Who here works with "sysadmins"? I mean... devops-all-the-places now, right? :P Share your experiences?: I'm curious.

If I were a sysadmin, I'd have some kind of "sanity check" script if I had to manage a fleet of disparate systems. I'd use it automatically when I logon to a system. Checking things like: Linux vs BSD, what tools are installed, load, any weirdnesses, etc. Heck, maybe even create aliases that abstract over all of it, as much as possible. Maybe copy over a helix binary for editing too, if that was kosher.

> we're working on a new terminal sequence specification that would allow terminal programs to expose any of their actions directly in the command palette (e.g. imagine Neovim commands being fully available in the command palette).

/mic-drop

Not just macOS: "The command palette is bound by default to ctrl+shift+p on GTK and cmd+shift+p on macOS."

> The command palette exposes almost every available keybind.

> This has some immediate benefits, namely that you can access keybind actions even if they aren't bound to a keybind. This is useful for infrequently used actions. ... For example, I personally find myself using the move_tab action via the command palette frequently, but not frequently enough to justify binding it.

Definitely a wild read. Bill has a certain style of writing; here is my favorite part:

  > So I gave up and sent mail to Amir saying - where is
  > this Moviemaker download? Does it exist?
  >
  > So they told me that using the download page to 
  > download something was not something they anticipated.
  >
  > They told me to go to the main page search button
  > and type movie maker (not moviemaker!).
  >
  > I tried that. The site was pathetically slow but after
  > 6 seconds of waiting up it came.
  >
  > I thought for sure now I would see a button to just go
  > do the download.
  >
  > In fact it is more like a puzzle that you get to
  > solve. It told me to go to Windows Update and do a
  > bunch of incantations.

That's just the beginning.

Do you think the design direction of “chat first” is compatible with editor first? I don’t know if any tools do both well. Seems like a fork in the road, design wise.

If Bill is reading this: Another way to commit to quality would be (a) fund a $200M+ open source foundation to migrate/port/rewrite all of Windows (drivers, utilities, applications) to Linux; (b) give the foundation full IP rights to do so; (c) put all Microsoft Windows employees on the effort too.

I would broaden the list of risks:

- Pandas is interwoven into downstream projects. So it will be here to stay for a long time. This is good for maintenance and stability. Advantage: Pandas.

- OTOH, the Pandas experience is awful; this was obvious to many from the outset, and yet it persisted. I haven't tracked the history. But my guess would be the competition from Polars was a key pressure for improvement. Edge: Polars.

- Lots of Python projects are moving to Rust-backed tooling: uv, Polars, etc. Front-end users get the convenience of Python and tool-developers get the confidence & capabilities of Rust. Edge: Polars.

- Pandas has a governance structure not tied to one company. Polars does not. (comment above said this) Advantage: Pandas.

But this could change. Polars users could (and may already be?) pressing for company-independent governance.

> [Polars] is much more logical and has far fewer surprises compared to Pandas

A kind understatement imo. For me, the following experiences are highly coupled in my brain: "I'm using Pandas" + "I'm feeling a weird combination of confusion and pain" + "This is a dumpster fire".

What kinds of actionable plans ever result from blaming people (as a category)? Where will it get you? Expecting some people to behave differently... just "because"? What kinds of plans flow downstream from blaming human nature? What's the plan? Does it help you somehow, practically? Or is it mostly about feeling better somehow?

If the plan is persuasion, putting blame aside goes a long way.

If you want to make change based in the real world, you could do worse that reading and absorbing "Thinking in Systems: A Primer" by Donella Meadows.

> Will humans take this to heart and actually do the right thing? Sadly, probably not.

Don’t blame the people, blame the system.

Identifying the problem is just the first step. Building consensus and finding pragmatic solutions is hard. In my opinion, a lot of technical people struggle with the second sentence. So much of the ethos in our community is “I see a problem, and I can fix it on my own by building [X].” I think people are starting to realize this doesn’t scale. (Applying the scaling metaphor to people problems might itself be a blindspot.)

Well said:

> Before LLMs, [high quality code contribution] was easier to sense because you were limited to communicating what you understood. With LLMs, it’s much easier to communicate a sense of understanding to the reviewer, but the reviewer doesn’t know if you actually understood it.

Now my twist on this: This same spirit is why local politics at the administrative level feels more functional than identity politics at the national level. The people that take the time to get involved with quotidian issues (e.g. for their school district) get their hands dirty and appreciate the specific constraints and tradeoffs. The very act of digging in changes you.

Public service announcement to hopefully reduce unnecessary knife fights*:

There are two compatible and important (but different) questions in play:

1. Is a program correct relative to a formal specification?

2. Is the formal specification what we mean/want?

*: Worth asking: “What that other person necessarily wrong? Or perhaps they are discussing a different aspect or framing?” AKA: “be curious and charitable” I’m not going to link to the specific threads, but they are happened / are happening. Le Sigh.

Yes “open weights” conveys the reality more clearly: merely having the parameters is very different than able to run a process that creates them. Without openness of the full process start to finish, much is hidden.*

Remember, language is what we make it. Dictionaries are useful catalogs of usage but we make the judgment calls.

* Even with the process, much is not well understood! / The ethics of releasing an open weights model at some capability level is a separate discussion.

> I don't know why you created a new account for this

What value does this add to the conversation? I’m not seeing it: am I missing something? It comes across as a kind of insult.

They made a good point in my opinion! (The “Uhm no” part got it off on the wrong foot, I will admit.) But even if you felt annoyed or didn’t agree with the point, it was substantive and moved the conversation forward. I’m here for the (genuine) questions and (constructive) debate and (civil) pushback.

I like to welcome new users before they take too much of a beating. That can come later when they are too invested to leave and/or when morale needs improving.

So welcome! Bring a helmet, and don’t stop disagreeing.

The main point I want to make now is: running different experiments based on different policies and approaches is valuable. Different online communities navigate this space in different ways.

> Travel is fatal to prejudice, bigotry, and narrow-mindedness. - Mark Twain

Travel comes in many forms. The best kind, imo, involves getting out of your comfort zone.

The balls-from-the-sky sieve-style animation* showing classifications literally falling out of the decision tree is my favorite part. I haven't seen this anywhere else (yet); this visualization technique deserves more percolation (pun intended). (#1)

Not even to mention the fact that the animation is controlled by scrolling, which gives an intuitive control over play, pause, rewind, fast-forward, etc. Elegant and brilliant. (#2)

Stunningly good also in the sense that it advances the story so people don't just drool at the pretty animation and stop engaging. Thus putting the "dark arts" in the service of learning. (#3)

All three ideas warrant emulation in other contexts!

* Find it towards the bottom under the "Making predictions" heading.

Hello and welcome. I appreciate your philosophical background; we need more of that around here imo. In a totally unrelated question /s, have you seen the movie Get Out by Jordan Peele? :P For philosophical discussions of AI, I much prefer the Alignment Forum. For thoughtful, critical, charitable discussion, I recommend LessWrong by leaps and bounds, as long as one doesn't demand brevity. Also, the bar for participation can feel higher other there. I'm ok with that because it encourages people to build up a lot of shared foundations for how we communicate with each other.

First, please don't take this as an endorsement of minimum-effort posting (of any kind, whether LLM-assisted or not). I feel the need to say this because people seem to be on hair-trigger alert for anything that seems in any way to denigrate the importance of human-written comments. I want people to "be human" here while also being mindful of how to contribute to the culture and conversation. What that looks like and what that entails is certainly up for discussion. / Ok, with that out of the way, I have four major points that build on each other, leading to a more direct response to the comment above.

1. Reasonable people may disagree in meaningful ways about what "respecting one's audience" means. There is significant variation in what qualifies as a "good faith participant" in a conversation.

In my case, I strive to seek truth, do research, be thoughtful, and write clearly. Do I hope others share these goals? Yeah, I think it would be nice and helpful for all of us, but I don't realistically expect it to happen very often. Do other people share these goals? Do they even see my writing as striving in those directions? These are really hard questions to answer.

2. It helps to recognize the nature of human communication. It a sloppy, messy, ill-defined not-even-protocol. The communication channel is a multi-layered mess. Participants bring who-knows-what purposes and goals. (One person might care about AI-assisted coding; another might be weary and sick of their employer pushing AI into their workflow; another might be seeing their lifelong profession being degraded; etc.)

3. What do the other participant(s) have in common? Background knowledge? Values? Goals? Norms and expectations? Part of communication is figuring out these "out-of-band" aspects. How do you do it? Hoping to do this "in-band" feels like building an airplane while flying it!

4. How does communication work, when it sort of works at all? Why? Individual interactions (i.e. bilateral ones) often work better when repeated over time. These scale better with the help of group norms. Norms make more sense and are more durable in the context of shared values.

So, with the above in mind, you might start to reframe how you think about:

> It shows a clear prioritization of the author's time over the readers', which fine, you're entitled to valuing your own time more than mine, but if you do, I'll receive that prioritization as inherently disrespectful of my time.

The reframing won't suddenly make the communication a better use of one's time. But it does shed light on the mindset and motives of others. In other words, communication breakdowns happen all the time without malicious intent or disrespect.

It seems like ^ and ^^ agree to me. Am I missing something?

> but proving correctness is one of those basically impossible tasks.

To aim for a meeting of the minds... Would you help me out and unpack what you mean so there is less ambiguity? This might be minor terminological confusion. It is possible we have different takes, though -- that's what I'm trying to figure out.

There are at least two senses of 'correctness' that people sometimes mean: (a) correctness relative to a formal spec: this is expensive but doable*; (b) confidence that a spec matches human intent: IMO, usually a messy decision involving governance, organizational priorities, and resource constraints.

Sometimes people refer to software correctness problems in a very general sense, but I find it hard to parse those. I'm familiar with particular theoretical results such as Rice's theorem and the halting problem that pertain to arbitrary programs.

* With tools like {Lean, Dafny, Verus, Coq} and in projects like {CompCert, sel4}.

I want to point out two conversational disconnects and offer some feedback, person to person. I edited my post a bit, so maybe you replied to a previous draft of mine. Anyhow, in terms of what we can see now, I want to clear up a few things:

---

>>> aB: The prompt & any follow-ups do have notable effects, but IMO this just means that most of actual meaning you wanted to convey is in those prompts.

>> xpe: If you mean in the sense of differentiating meaning from the base model, I take your point.

(I clarified; seems like we agree on this.)

> aB: That’s not [my] point.

(Conversational disconnect #1)

---

>>> aB: If I was your interlocutor, I'd understand you & your ideas better if you posted your prompts as well as (or instead of) whatever the LLM generated.

>> xpe: Indeed, yes, this is a good practice for intellectual honesty when citing an LLM.

(I clarified; seems like we agree on this.)

> aB: Post your prompts.

(Conversational disconnect #2)

---

> Post your prompts.

This feels abrasive. In another comment you repeat this line pretty much verbatim several times.

It is unclear if you are accusing me of using an LLM. I'm not.

---

> If you believe that LLM conversation is better, that’s great.

I hope you recognize that is not what I said, nor how I would say it, nor representative of what I mean.

> I’m sure there’s a social media network out there featuring LLMs talking to other LLMs. It’s just not this one.

This doesn't reply substantively to what I wrote; it feels like a caricature of it.

> That’s not the point.

This is kinder to the reader if you say "That's not my point". Otherwise it can sound like that you get to decide what the point is.

Overall, in total, we agree on many things. But somehow that got lost. Also, the tone of the comment above (and its grandparent too) feels a bit brusque and condescending.

Thanks for sharing. / Speaking of keystroke analysis, have you read "Fall" by Neal Stephenson? A fun read; there is a generalization of this idea therein.

> Anyone with a level of proficiency sufficient for reading the comments here can manage writing English at a passable level.

I'm an English speaker with some Spanish education and practice. My experience is that reading, writing, listening, and speaking can be quite uneven. Uneven enough to matter.

In the long-run, yes, learning a language is better, assuming your goal is to learn the language. I'm not trying to be snarky: sometimes people simply want to communicate an idea quickly in the short-run and/or don't prioritize deepening a language skill.

I would rephrase the comment above as a question: "Given the set of tools available (in person tutoring, online tutoring, AI-tooling, etc) and what we know about learning from cognitive science, for a given budget and time investment, what combination of techniques work better and worse for deepening various language skills?"

I agree with the above comment on a broad normative (what is good) take: on a forum for humans, yes, please, bring your human self. But there is a lot of room for variety, choice, even self-expression in the be your human self part! Some might prefer using the Encyclopedia Brittanica to supplement an imperfect memory. Others DuckDuckGo. Some might bounce their ideas off friends. Or (gasp) an LLM. Do any of these make the person less human? Nope.

Of course, there are many ways to be more and less intellectually honest, and there is a lot to read on this, such as [1].

Now, on the descriptive / positive claims (what exists), I want to weigh in:

> LLMs are an autocomplete engine.

Like all metaphors, we should ask the "what is the metaphor useful for?" rather than arguing the metaphor itself, which can easily degenerate into a definitional morass. Instead, we should discuss the behavior, something we can observe.

> [LLMs] aren't curious.

Defined how? If put aside questions of consciousness and focus on measuring what we can observe, what do we see? (Think Turing [2], not Chalmers [3].) To what degree are the outputs of modern AI systems distinguishable from the outputs of a human typing on a keyboard?

> LLMs CANNOT provide unique objectivity...

Compared to what? Humans? The phrasing unique objectivity would need to be pinned down more first. In any case, modern researchers aren't interested in vanilla LLMs; they are interested in hybrid systems and/or what comes next.

Intelligence is the core concept here. As I implied in the previous paragraph, intelligence (once we pick a working definition) is something we can measure. Intelligence does not have to be human or even biological. There is no physics-based reason an AI can't one day match and exceed human intelligence.*

> or offer unknown arguments ...

This is the kind of statement that humans are really good at wiggling out of. We move the goalposts. So I'll give one goalpost: modern AI systems have indeed made novel contributions to mathematics. [4]

> because they can only use their own training data, based on existing objectivity and arguments, to write a response.

Yes, when any ML system operates outside of its training distribution, we lose formal guarantees of performance; this becomes sort of an empirical question. It is a fascinating complicated area to research.

Personally, I wouldn't bet against LLMs as being a valuable and capable component in hybrid AI systems for many years. Experts have interesting guesses on where the next "big" innovations are likely to come from.

[1]: Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and Biases: Biases in judgments reveal some heuristics of thinking under uncertainty. science, 185(4157), 1124-1131.

[2]: The Turing Test : Stanford Encyclopedia of Philosophy : https://plato.stanford.edu/entries/turing-test/

[3]: The Hard Problem of Consciousness : Internet Encyclopedia of Philosophy : https://iep.utm.edu/hard-problem-of-conciousness/

[4]: FunSearch: Making new discoveries in mathematical sciences using Large Language Models : Alhussein Fawzi and Bernardino Romera Paredes : https://deepmind.google/blog/funsearch-making-new-discoverie...

* Taking materialism as a given.

> The prompt & any follow-ups do have notable effects, but IMO this just means that most of actual meaning you wanted to convey is in those prompts.

If you mean in the sense of differentiating meaning from the base model, I take your point. But in another sense, using GPT-OSS 120b as example where the weights are around 60 GB and my prompt + conversation are e.g. under 10K, what can we say? One central question seems to be: how many of the model's weights were used to answer the question? (This is an interesting research question.)

> If I was your interlocutor, I'd understand you & your ideas better if you posted your prompts as well as (or instead of) whatever the LLM generated.

Indeed, yes, this is a good practice for intellectual honesty when citing an LLM. It does make me wonder though: are we willing to hold human accounts to the same standard? Some fields and publications encourage authors to disclose conflicts of interest and even their expected results before running the experiments, in the hopes of creating a culture of full disclosure.

I enjoy real human connection much more than LLM text exchanges. But when it comes to specialized questions, I seek any sources of intelligence that can help: people, LLMs, search engines, etc. I view it as a continuum that people can navigate thoughtfully.