photonthug

Meh. For a long time people have been saying stuff like "devops is dead, long live the platform engineer" exactly because the figuring-out-what-works phase of wild experimentation is over and there are unambiguous "winners" for technology in most every niche. You've listed lots of commercial vendors and alternate choices for backends/front-ends here as if to illustrate a lack of standardization in tools/frameworks, but is it really?

Whereas churn in web-dev seems self-inflected.. devops practitioners don't actually create vendor/platform fragmentation, they just deal with it after someone else wants the new trendy thing. Devops is remarkably standardized anyway even in the face of that, as evidenced by the fact that there's going to be a terraform'y way to work with almost all of the vendors/platforms you mentioned. And mentioning 20 ways to use kubernetes glosses over the fact that.. it's all just kubernetes! Another amazing example of standardization and a clear "winner" for tech stack in niche.

An loanword in English, many will recognize it from Bach if nothing else https://en.wikipedia.org/wiki/The_Well-Tempered_Clavier

The silly thing about this is that context is everything. I bet it's extremely easy to be a top-tier figure-skater in, say, a small tropical island nation? In a similar way, I very much doubt that you'd really need to be in the top 0.2% of the population to complete a phd. Do you need to be in the top 0.2% of people to compete as a contributor with absolutely everyone else in the whole world at the same time? Well yeah, but at that point the statement is so obviously true that it doesn't mean much.

> One thing that it doesn't really cover is the rest of German society and how those thugs managed to get power.

Just to clarify for other folks, there are many episodes re: Nazis, but it also covers everything from Khmer Rouge to more modern coverage that's truly the more banal kind of evil, covering the worst and most destructive grifters. So while it's definitely kinda preoccupied with fascism, there's another through-line with dis/misinformation, etc etc.

I do agree with your basic criticism though, fair to say the general show format for dictators is 1st part bio which is frequently unremarkable, then the 2nd part is appalling crimes. How society was complicit/tolerant enough to allow the decline to happen is usually sidelined. On the other hand though, it's kind of always the same and pretty simple. To the extent it's not simply hidden or covered up, it works like this. After things are definitely very shitty, whatever misguided optimism folks can muster is usually all about "harming the out-group will help somehow!". (It doesn't.)

But the astute dictator (or their advisors) can rely on and exploit that kind of tribalism. Common sense, static value-systems, or any sensitivity to blatantly hypocritical statements/behaviour etc just are not things that the common person can really hang on to once they are angry/impoverished/aggrieved/hungry

As a big believer in documentation and communication in general, there's this inevitable double-bind that people hate whatever you give them and also hate it if you give them nothing. LLMs have made this worse.

No emojis and any effort to be comprehensive? Everyone complains "what is this wall of text", or "this is industry not grad school so cut it out with the fancy stuff" or "no one spends that much time on anything and it must be AI generated". (Frequently just a way of saying that they hate to read, and naively believe that even irreducibly complex stuff is actually simple).

Stuff that's got emojis, a friendly casual tone and isn't information dense? Well that's very chatty and cute, it also has to be AI and can't be valuable.

Since you can't win with docs, the best approach is to produce high quality diagrams that are simultaneously useful for a wide audience from novice to expert. The only problem is that even producing high quality diagrams at a ratio of 1 diagram per 1k lines of code is still very time consuming to produce if you're putting lots of thought into it, double so if you're fighting the diagramming tools, or if you want something that's easy for multiple stakeholders with potentially very different job descriptions to take in. Everyone will call it inadequate, ask why it took so long, and ask for the missing docs that they will hate anyway!

On the bright side, LLMs are pretty great at generating mermaid, either from code, or natural language descriptions of data-flows. Diagrams-as-code without needing a whole application UI or one of a limited number of your orgs lucid-chart licenses is making "Don't like it? Submit a PR" a pretty small ask. Skin in the game helps to curbs endless bike-shedding criticism

To know the answers to all of these questions, you should really check out the Behind the Bastards podcast because that is the whole premise. Covering the lead-up to horrible situations and the inevitable slide in fascism. It's insanely detailed about covering many, many stupid fascist bastards and a few smart ones.

To know the answers to all of these questions, you should really check out the Bbehind the Bastards podcast because that is the whole premise. Covering the lead-up to horrible situations and the inevitable slide in fascism. It's insanely detailed about covering many, many stupid fascist bastards and a few smart ones.

Love the premise and I see several years are posted. I like your philosophy of science section especially, because while most would neglect that area, it's probably got good predictive/foreshadowing juice in general (although not necessarily for any given year).

Skimming https://en.wikipedia.org/wiki/1875_in_science the twin studies and behavioural genetics is interesting. The challenger-deep thing too since it's earlier than I would have guessed, but IMHO it would be more exciting/appropriate to categorize as "exploration" than science. Did they publish a "paper" about stuff like that back then, or just tell the royal society, tell the newspapers and call it good?

A pointless but fun question to think about is, how to decide the most important thing that happened in a given year? Sometimes a discovery, sometimes an idea, sometimes a project, election, or war. But for a slow year.. maybe it's just that someone who will have that idea or start that project later was born.

> There is a certain amount of regular work that I don't want to automate away, even though maybe I can. That regular work keeps me in the domain. It leads to epiphany's in regards to the hard problems. It adds time and something to do in between the hard problems.

Exactly, some kinds of refactors are like this for me. Pretty mindless, kind of relaxing, almost algebraic. It's a pleasant way to wander around the code base just cleaning and improving things while you walk down a data or control flow. If you're following a thread then you don't even make decisions really, but you also get better acquainted with parts you don't know, and subconsciously get the practice holding some kind of gestalt in your head.

This kind of almost dream-like "grooming" seems important and useful, because it preps you for working with design problems later. Definitely formatting and style type trivia should absolutely be automated, and real architecture/design work requires active engagement. But there's a sweet spot in the middle.

Even before LLMs maybe you could automate some of these refactors with tools for manipulating ASTs or CSTs, if your language of choice had those tools. But automating everything that can be automated won't necessarily pay off if you're losing fluency that you might need later.

A related topic you might want to look into here is called nucleus sampling. Similar to temperature but also different.. it's been surprising to me that people don't talk about it more often, and that lots of systems won't expose the knobs for it.

> where is the problem?

Honestly, if you have to ask then you'll probably not understand the answer, but here's some related questions to ponder. What's the problem with having money in politics as much as possible? What's the problem with eliminating all leadership with relevant domain-expertise, replacing it with people who know how to "play the game"? What's the problem with class-based societies in general? What's the problem with ignoring all fundamentals, denying expertise can even exist, and just full on embracing superficial optics everywhere? We've been in the fuck-around phase for a while now, but we're moving closer to the find-out phase.

> Are we just invoking the spectre of fallen comrades in battle as an emotional plea?

No. The emotional plea would be that IT and SWE actually created upward mobility for a lot of talented people who otherwise would not have been able to buy their way into the American middle class without, for example, joining the army to risk death for the benefit of elites. It will be sad to see backwards movement on that for sure, but we don't even need to invoke this argument.

The more rational argument is simply that meritocracy works better than classism. Even if you're fine with feeding people into the meat grinder on the off chance you get some personal glory, it's not just bad for the victims, it's bad for general morale, the army, the country involved, etc. Substitute these words with money/markets/shareholders/industry or whatever if it helps you to understand.

After we fix the all the simple specious reasoning of stuff like Alexander-the-great and agree to out-source certain problems to appropriate tools, the high-dimensional analogs of stuff like Datasaurus[0] and Simpson's paradox[1] etc are still going to be a thing. But we'll be so disconnected from the representation of the problems that we're trying to solve that we won't even be aware of the possibility of any danger, much less able to actually spot it.

My take-away re: chain-of-thought specifically is this. If the answer to "LLMs can't reason" is "use more LLMs", and then the answer to problems with that is to run the same process in parallel N times and vote/retry/etc, it just feels like a scam aimed at burning through more tokens.

Hopefully chain-of-code[2] is better in that it's at least trying to force LLMs into emulating a more deterministic abstract machine instead of rolling dice. Trying to eliminate things like code, formal representations, and explicit world-models in favor of implicit representations and inscrutable oracles might be good business but it's bad engineering

[0] https://en.wikipedia.org/wiki/Datasaurus_dozen [1] https://towardsdatascience.com/how-metrics-and-llms-can-tric... [2] https://icml.cc/media/icml-2024/Slides/32784.pdf

From the paper abstract,

> (4) we derive the optimal chain-of-thought length as [..math..] with explicit constants

I know we probably have to dive into math and abandon metaphor and analogy, but the whole structure of a claim like this just strikes me as bizarre.

Chain-of-thought always makes me think of that old joke. Alexander the great was a great general. Great generals are forewarned. Forewarned is forearmed. Four is an odd number of arms to have. Four is also an even number. And the only number that is both odd and even is infinity. Therefore, Alexander, the great general, had an infinite number of arms.

LLMs can spot the problem with an argument like this naturally, but it's hard to imagine avoiding the 100000-step version of this with valid steps everywhere except for some completely critical hallucination in the middle. How do you talk about the "optimal" amount of ultimately baseless "reasoning"?

> There are two camps emerging, and the difference isn't really about skill level or experience.

Assuming everything else the author believes is true, the real camps are "money" and "less money". Those camps already determine the success of businesses to a large extent. But especially in SWE where we traditionally cared less about degrees and more about skill, it's a new thing that "skill" and "experience" are directly cash related, something you can buy and out-source.

Looking for work and need a better github portfolio? Just up your Claude spend. Find yourself needing a promotion at work and in possession of some disposable income? Just pay out of pocket for the AI instead of expecting overtime from your employer or working nights and weekends, because you know you'll make up the difference when you're in charge of your department.

There is some historical precedent for this sort of thing; just read up on the buying and selling of army commissions. That works as well as you might expect, because when "expertise" is purchased like this it turns out that the officers you get are incompetent, and they mostly just fed soldiers into a meat-grinder. https://en.wikipedia.org/wiki/Purchase_of_commissions_in_the...

About those rules. Any high-profile court case is by definition crime/politics, and all over TV news. Have you recently mentioned the name of any famous person(s) in your comments, offering opinions and critiques perhaps? But that's another way of saying "celebrities". So I'd say you're unambiguously failing on 4 of these criteria. In terms of forming teams and bitching about the refs decisions, not much difference between court-cases and sports either, neither are big on inspiring curiosity. But hey.. rules are for other people right boss?

And stop telling people to "look". You look. Because listen, I know that phrases like this one are well-loved by a certain type of person. Shows who's the adult in the room, and also frightens subordinates into silence, right? But understand me now when I say that it's much too transparent when used too often. Realize that there are other adults in the room, and when you toss out too many imperatives too fast then it's easy to see how much you want to control people as well as the topics under discussion.

> General intelligence may not be SAT/SMT solving but it has to be able to do it, hence, backtracking.

Just to add some more color to this. For problems that completely reduce to formal methods or have significant subcomponents that involve it, combinatorial explosion in state-space is a notorious problem and N variables is going to stick you with 2^N at least. It really doesn't matter whether you think you're directly looking at solving SAT/search, because it's too basic to really be avoided in general.

When people talk optimistically about hallucinations not being a problem, they generally mean something like "not a problem in the final step" because they hope they can evaluate/validate something there, but what about errors somewhere in the large middle? So even with a very tiny chance of hallucinations in general, we're talking about an exponential number of opportunities in implicit state-transitions to trigger those low-probability errors.

The answer to stuff like this is supposed to be "get LLMs to call out to SAT solvers". Fine, definitely moving from state-space to program-space is helpful, but it also kinda just pushes the problem around as long as the unconstrained code generation is still prone to hallucination.. what happens when it validates, runs, and answers.. but the spec was wrong?

Personally I'm most excited about projects like AlphaEvolve that seem fearless about hybrid symbolics / LLMs and embracing the good parts of GOFAI that LLMs can make tractable for the first time. Instead of the "reasoning is dead, long live messy incomprehensible vibes", those guys are talking about how to leverage earlier work, including things like genetic algorithms and things like knowledge-bases.[0] Especially with genuinely new knowledge-discovery from systems like this, I really don't get all the people who are still staunchly in either an old-school / new-school camp on this kind of thing.

[0]: MLST on the subject: https://www.youtube.com/watch?v=vC9nAosXrJw

[flagged]

What? I'm not arguing that people have the right to ignore important safety-related notices, especially if it's related to public safety.

I'm arguing that we should have the right to reliably separated channels for safety/operational notifications vs commercial content / outright scams. We don't have that though, which effectively erodes the safety you are saying you want to protect. If you're serious about safety, you should agree that using airplane PAs for emergencies instead of ads is a good idea. You should also be onboard with the idea that "service required" should actually mean "service required", not just that it's time to pay what amounts to a subscription fee to the vendor. Once a signal has degraded into pure noise, people get used to ignoring it.

The situation is mostly the same with software updates.. no way for end-users to reliably separate updates that help them vs ones that are only going to hurt them. Serious about security? Don't get too comfortable blasting your users with immaterial "news and updates" trash, or of course they want to ignore you

You're acting like this is related to necessary maintenance / safety, why give corporate the benefit of the doubt without knowing about the vehicle involved?

Even things like washing machines and coffee makers will soft-brick themselves these days asking you to "start self clean-cycle with Foo(tm) substance" and then won't perform their function without some kind of forced reset. That part of the airplane ride where the PA is blasting some kind of "join our miles club" literally at a captive audience with no choice but to listen? It's not safety related either. My headphones that I would like to use to drown out the PA advertisement literally stop working if they detect speech, and the only way to disable this "feature" is to download their app.

This is just growth-hacking "zero-cost advertisements to a targeted audience" stuff that's extremely disrespectful at best, and kinda looks like it's edging closer to threats and extortion.

Open-source / noncommercial isn't exactly a cure for the steady drip of "like and subscribe" type of harassment these days. Probably just because commercial interests have pushed user-harassment so hard for so long that the window has shifted and now users expect harassment, and because at least some makers actually feel their work is less professional if they don't engage in harassment.

Open your laptop, dismiss ubuntu wanting to update stuff; open firefox and have your adblock extension popup to tell you how many ads it blocked and to give you a helpful advertisement for updating your adblock; open a web page, any webpage, dismiss at least 3 popups for cookies, decline to signin in google/facebook, decline the newsletter. Get another browser extension to solve these problems, it will probably have pop-ups to tell you about "what's new". Open github to look at some code, get asked to star the project above the fold in the documentation. Forget what you even wanted to do with a laptop, close it, dive into much more productive work by figuring out the best way to feed your laptop into the kitchen garbage disposal in small pieces. Just another Tuesday

This misses the point completely? Wanting to know that you're not alone and wanting to be left alone are completely different, despite the word choice. Surely you realize this. Saying that "I prefer to avoid unsolicited harassment from corporate interests" is hardly the same as saying "I'm an unapologetic misanthrope".

I'm also reading TFA's intent as less about big-brother and info-hygiene stuff, and more about standard enshittification.

Since we're talking about "doing more things with a piece of software than [users] could do without it", well, living with 10% of the features at 1000x of the price would typically still fit those criteria. Since this is exactly where most companies would like to go after establishing market-dominance, yeah, I think we do want to be protected.

Whether any dev or group of devs could realistically push back against the forces at work in the org or the wider economy here is a separate question of course.

> https://github.com/pydantic/pydantic-ai/issues/2405

Thanks, this is a very interesting thread on multiple levels. It does seem related to my problem and I also learned about field docstrings :) I'll try moving my dependency closer to the bleeding edge

Thanks for the reply. Native output is indeed what I'm shooting for. I can't share the model directly right now, but putting together a min-repro and moving towards and actual bug report is something on todo list.

One thing I can say though.. my models differ from the docs examples mostly in that they are not "flat" with simple top-level data structures. They have lots of nested models-as-fields.

I wanted to love pydantic AI as much as I love pydantic but the killer feature is pydantic-model-completion and weirdly.. it has always seemed to work better for me when I naively build it from scratch without pydantic AI.

I haven't looked deeply into pydantic's implementation but this might be related to tool-usage vs completion [0], the backend LLM model, etc. All I know is that with the same LLM models, `openai.client.chat.completions` + a custom prompt to pass in the pydantic JSON schema + post-processing to instantiate SomePydanticModel(*json) creates objects successfully whereas vanilla pydantic-ai rarely does, regardless of the number of retries.

I went with what works in my code, but didn't remove the pydantic-ai dependency completely because I'm hoping something changes. I'd say that getting dynamic prompt context by leveraging JSON schemas, model-and-field docs from pydantic, plus maybe other results from runtime-inspection (like the actual source-code) is obviously a very good idea. Many people want something like "fuzzy compilers" with structured output, not magical oracles that might return anything.

Documentation is context, and even very fuzzy context is becoming a force multiplier. Similarly languages/frameworks with good support for runtime-inspection/reflection and have an ecosystem with strong tools for things like ASTs really should be the best things to pair with AI and agents.

[0]: https://github.com/pydantic/pydantic-ai/issues/582

Ok this nerd-sniped me pretty good, never seen this before and assumed it would be quickly connected to the Ulam spiral mentioned elsewhere in the the thread. That particular rabbit hole kinda bottoms out in polynomial residues and the very mysterious-sounding "Conjecture F" [0].

This parallax primes thing though led to the linked page [1] which has lots of background and other connections, including the most satisfying part, which turned out more geometric [2]

[0] https://en.wikipedia.org/wiki/Ulam_spiral#Explanation [1] https://www.novaspivack.com/science/we-have-discovered-a-new... [2] https://www.cut-the-knot.org/Curriculum/Arithmetic/PrimesFro...

As linked elsewhere in this thread, see Uri Berliner on the subject https://www.thefp.com/p/npr-editor-how-npr-lost-americas-tru...

> logics philosophers use .. aren't very "modular" and can't easily be mixed

Not sure if the model-checking communities would agree with you there. For example CTL-star [0] mixes tree-logic and linear-temporal, then PCTL adds probability on top. Knowledge, belief, and strategy-logics are also mixed pretty freely in at least some model checkers. Using mixed combinations of different-flavored logic does seem to be going OK in practice, but I guess this works best when those diverse logics can all be reduced towards the same primitive data structures that you want to actually crunch (like binary decision diagrams, or whatever).

If no primitive/fast/generic structure can really be shared between logics, then you may be stuck with some irreconcilable continuous-vs-discrete or deterministic-vs-probabilistic disconnect, and then require multiple model-checkers for different pieces of one problem. So even if mixing different flavors of logics is already routine.. there's lots of improvements to hope for if practically everything can be directly represented in one place like lean. Just like mathematicians don't worry much about switching back and forth from geometry/algebra, less friction between representations would be great.

Speaking of CTL, shout out to Emerson[1], who won a Turing award. If he hadn't died recently, I think he'd be surprised to hear anyone suggest he was a philosopher instead of a computer scientist ;)

[0]: https://en.wikipedia.org/wiki/CTL* [1]: https://en.wikipedia.org/wiki/E._Allen_Emerson

> why are we ignoring the loud hints from ML solutions that this is a limiting heuristic?

This comes up a lot and always strikes me as rather anti-science, even anti-rationality in general. To speed run the typical progression of this argument, someone says alchemy and astrology occasionally "work" too if you're determined to ignore the failures. This point is then shot down by a recap about the success of QM despite Einstein's objections, success of the standard model even with lots of quasi-empiricism etc, etc.

Structurally though.. if you want to claim that the universe is fundamentally weird and unknowable, it's very easy to argue this, because you can always ignore the success of past theory and formalisms by saying that "it was nice while it lasted but we've squeezed all the juice out of that and are in a new regime now". Next you challenge your detractors to go ahead and produce a clean beautiful symmetric theory of everything to prove you wrong. That's just rhetoric though, and arguments from model/information/complexity theory etc about fundamental limits on what's computable and decidable and compressible would be much more satisfying and convincing. When does finding a complicated thing that works actually rule out a simpler model that you've missed? https://en.wikipedia.org/wiki/Minimum_description_length#MDL...

Huh? I defended your proposal of asking strangers to read stuff against the detractors here, because even though it's a bit obnoxious, I think it's a more interesting conversational gambit than discussing sports or something. And now I'm the weirdo? You attack your allies here in a way that raises questions about your own ability to read well.