HNHacker News
TopNewBestAskShowJobs

robot-wrangler

639 karma · joined September 10, 2025

submissionscomments
robot-wrangler··on Google's 200M-parameter time-series foundation model with 16k context
> Because many of these have the same underlying causal structures - humans doing things, weather correlations, holidays.

Or, you know, maybe they aren't. Thermometers and photon counts are related to weather sometimes, but not holidays. Holidays are related to traffic sensors and to markets, but not Geiger counters.

> Well studied behavioral stuff like "the stock market takes the stairs up and the elevator down" which is not really captured by "traditional" modelling tools.

Prices are the opposite, up like a shot during shocks, falling slowly like a feather. So that particular pattern seems like a great example of over-fitting danger and why you wouldn't expect mixing series of different types to be work very well.

robot-wrangler··on Google's 200M-parameter time-series foundation model with 16k context
This seemed like a good answer at first. But on further thought, images on the whole really do seem to have quite a bit more standard structure / "grammar" to exploit compared to arbitrary time-series. Many images are of the world, where there is gravity so you might see preponderance of blobs at the bottom, or the repetitive types like people, animals, faces, eyes. Wildly abstract images still have some continuity, pixels in a neighborhood are likely to be similar.

Time series in general have none of this kind of structure that's strictly necessary. I'm sure that many real-world sensors typically have some gaussian distribution aspects + noise and/or smoothness and locality types of assumptions that are pretty safe, but presumably that simple stuff is exactly what traditional time-series modelling was exploiting.

Maybe the real question is just what kind of time-series are in the training data, and why do we think whatever implicit structure that is there actually generalizes? I mean, you can see how any training that mixes pictures of dogs and cats with picturing of people could maybe improve drawing hair, detecting hair, or let you draw people AND dogs. It's less clear to me how mixing sensor data / financial data / anything else together could be helpful.

robot-wrangler··on So where are all the AI apps?
> Heh, I got a solid five seconds with the chart until the paywall popped up.

Relevant! If the maximalist interpretation of AI capabilities were close to real, and if people tend to point their new super powers at their biggest pain points.. wouldn't it be a big blow for all things advertising / attention economy? "Create a driver or wrapper app that skips all ads on Youtube/Spotify" or "Make a browser plugin that de-emphasizes and unlinks all attention-grabbing references to pay-walled content".

If we're supposed to be in awe of how AI can do anything, and we notice repeatedly that nope, it isn't really empowering users yet, then we may need to reconsider the premise.

robot-wrangler··on A pig's brain has been frozen with its cellular activity locked in place
Herbert West requires extremely fresh specimens
robot-wrangler··on EsoLang-Bench: Evaluating Genuine Reasoning in LLMs via Esoteric Languages
> Stay tuned

Never heard that before! But ok, it seems like this entity is affiliated with the paper, I'm interested..

> A little harness engineering was enough

Enough for what? It's not enough to crush the benchmark if that just means showing it is feasible to generate esolang code. No one cares about that if we're using it as a proxy to investigate general reasoning. Given validation/execution feedback loops, and 1000 retries for hello-world where we succeed with trial and error, the case for reasoning still wouldn't look great.

Suppose it's way better than that though; maybe trials are few and show clear logical progression. Well, we needed a harness, and that's still damning for whether and to what extent models can reason. But with harnesses at least we have a way to do general reasoning well enough on novel problems, right?

> mimic how humans would learn to solve problems in esoteric languages

Well hold on, does the harness do that, or does it enable models to do reasoning? We've retreated back towards solving that thing we weren't actually interested in..

robot-wrangler··on EsoLang-Bench: Evaluating Genuine Reasoning in LLMs via Esoteric Languages
> Frontier models score ~90% on Python but only 3.8% on esoteric languages, exposing how current code generation relies on training data memorization rather than genuine programming reasoning.

Finally! This is a really obvious test-case that I've wondered about myself, and have seen many casual skeptics and cautiously optimistic people independently raising for several years now. When megacorp is not crowing about such a test, the silence is deafening, and it was practically guaranteed that they tested, didn't like the results, and didn't publish.

I'm still surprised it took this long for academics to try it, and skimming cites, I don't see anything similar. Anyone know if this is the first paper to try this kind of thing, or just the first paper to put together a especially good suite of reusable benchies?

If this benchmark becomes popular, then presumably to avoid such embarrassments synthetic data is eventually added to training sets to make sure even esolangs are somewhat more in-distro, and then we gradually run out of esolangs to do honest testing with. SAT is a whole different animal admittedly, but comparable honest tests might involve just forcing models to use randomly generated but easily checked EBNF grammar? I don't have a quick link to the relevant papers, but afaik benchmarks of strict adherence to non-simple JSON schemas is also still pretty bad, and we're just working around it with lots of retries/tokens. "But look how well it works for 10k lines of kubernetes manifests!" Well yeah, maybe, but it barely needs to really follow a schema since that is more stuff that's in the training set..

robot-wrangler··on Explore 19th Century Scientific Correspondence
Found nothing I was actually looking for, but also found neat/weird stuff by accident almost immediately. For example here's some incomplete record of Darwin and Lewis Carrol[1] passing notes. I'm sure much of the corpus is trivial like that, but always fascinating to see legends humanized.

[1]: https://epsilon.ac.uk/search?keyword=dodgson&f1-addressee=Do...

robot-wrangler··on Warranty Void If Regenerated
Can't speak for them but FWIW it does not sound like OP is necessarily aware of the genre at all. They asked Claude to explain something via fiction, and then perhaps Claude made the "creative decision" based simply on the availability of the material.
robot-wrangler··on Warranty Void If Regenerated
I think this passes the sniff test only if you're not too familiar with this neighborhood of the training set. Not that the writing is bad but it's just derivative. I listen to stuff like "Lost Scifi" podcast almost daily for example, but there are many similar ones which are focused on reading classic stuff from the golden-age zines because it's all public domain.

The premise/structure/flavor of TFA is an almost pitch-perfect imitation of that kind of voice, to the point that I immediately flagged it as probably generated. I actually think a modern person would have some difficulty even in consciously mimicking it. There's an "aw shucks" yokel-thrown-into-the-future aspect to it. Plot-wise you have rural bicycle repair shop that expands operations to support nuclear reactors and that sort of thing. Substitute any of the more atomic-age stuff for AI stuff and you're mostly there. If you have some Amazing Stories from the 1920s on your shelf then you kind of know what I mean.

robot-wrangler··on Why AI systems don't learn – On autonomous learning from cognitive science
> Curious how you'd handle the reward signal for deciding when to switch between observation and active exploration without it collapsing into one mode.

If you like biomimetic approaches to computer science, there's evidence that we want something besides neural networks. Whether we call such secondary systems emotions, hormones, or whatnot doesn't really matter much if the dynamics are useful. It seems at least possible that studying alignment-related topics is going to get us closer than any perspective that's purely focused on learning. Coincidentally quanta is on some related topics today: https://www.quantamagazine.org/once-thought-to-support-neuro...

robot-wrangler··on Language Model Teams as Distrbuted Systems
Regardless of whether it's framed as old-school MAS or new-school agentic AI, it seems like it's an area that's inherently multi-disciplinary where it's good to be humble. You do see some research that's interested in leveraging the strengths of both (e.g. https://www.nature.com/articles/s41467-025-63804-5.pdf) but even if news of that kind of cross pollination was more common, we should go further. Pleased to see TFA connecting agentic AI to amdahls law for example.. but we should be aggressively stealing formalisms from economics, game theory, etc and anywhere else we can get them. Somewhat related here is the camel AI mission and white papers: https://www.camel-ai.org/
robot-wrangler··on Language model teams as distributed systems
> Next up, LLMs as actors & processes in π-calculus.

You jest, but agents are of course already useful and fairly formal primitives. Distinct from actors, agents can have things like goals/strategies. There's a whole body of research on multi-agent systems that already exists and is even implemented in some model-checkers. It's surprising how little interest that creates in most LLM / AI / ML enthusiasts, who don't seem that motivated to use the prior art to propose / study / implement topologies and interaction protocols for the new wave of "agentic".

robot-wrangler··on Apideck CLI – An AI-agent interface with much lower context consumption than MCP
> I'm getting tired of everyone saying "MCP is dead, use CLIs!".

The people saying this and attacking it should first agree about the question.

Are you combining a few tools in the training set into a logical unit to make a cohesive tool-suite, say for reverse engineering or network-debugging? Low stakes for errors, not much on-going development? Great, you just need a thin layer of intelligence on top of stack-overflow and blog-posts, and CLI will probably do it.

Are you trying to weld together basically an AI front-end for an existing internal library or service? Is it something complex enough that you need to scale out and have modular access to? Is it already something you need to deploy/develop/test independently? Oops, there's nothing quite like that in the training set, and you probably want some guarantees. You need a schema, obviously. You can sort of jam that into prompts and prayers, hope for the best with skills, skip validation and risk annotations being ignored, trust that future opaque model-change will be backwards compatible with how skills are even selected/dispatched. Or.. you can use MCP.

Advocating really hard for one or the other in general is just kind of naive.

robot-wrangler··on Apideck CLI – An AI-agent interface with much lower context consumption than MCP
> Limit integrations → agent can only talk to a few services

The idea that people see this as one horn of a trilemma instead of just good practice is a bit strange. Who would complain that every import isn't a star-import? Bring in what you need at first, then load new things dynamically with good semantics for cascade / drill-down. Let's maybe abandon simple classics like namespacing and the unix philsophy for the kitchen-sink approach after the kitchen-sink thing is shown to work.

robot-wrangler··on Verified Spec-Driven Development (VSDD)
> Try speccing out a flux capacitor. I'll wait.

One way to spec that is presumably something like "X% more efficient than current best-in-class", "made of Y,Z with no exotic materials", "takes no longer than T days to create" and so on.

Anyway, being "anti" spec isn't even wrong because it's just a completely incoherent position. There's always a spec.. including any informal prompt you kick off your agents with. Call it a "structured prompt" if that soothes you and your agents, then let's move on to the interesting part where we decide how much structure is optimal

robot-wrangler··on Verified Spec-Driven Development (VSDD)
That seems barely related and settles nothing? Bottom line is simple, saying "you can't spec out something you have no clue how to build" is saying you cannot desire coldness unless you understand how to build a refrigerator. It's just the difference between what and how. If you don't know the difference between implementation and specifications, just try a whole day of answering "what" and "why" questions with "how" answers and see how it goes.
robot-wrangler··on Elixir: Lazy BDDs with eager literal intersections
Not an expert, and not sure how but lazy BDD's fit in specifically, but I thought BDD's are already the gold-standard data structures in most model checkers and assumed they were at the limit of possibilities for performance. Can anyone comment whether that's true, whether it means this kind of implementation is already standard in other programming languages for sets/bools? What about infrastructure with fast set operations like redis?
robot-wrangler··on What is happening to writing? Cognitive debt, Claude Code, the space around AI
Doesn't matter when post-truth started because it's now over, and it's more accurate to characterize this era as "post-rationality". Most people do seem to understand this, but we are in different stages of grief about it.
robot-wrangler··on Kolakoski Sequence
A few hops away, the non-mathematical part of this story is interesting[0] and Conway's related weird world of audio-active decay is another recurring favorite[1]. Both are even more fun with esolangs[2]. I'm sure I've heard audio-encodings of Kolakoski before, but couldn't find any with a quick search

[0] https://en.wikipedia.org/wiki/William_Kolakoski#Personal_sig... [1] https://en.wikipedia.org/wiki/Look-and-say_sequence [2] https://esolangs.org/wiki/Kolakoski_sequence

robot-wrangler··on LLM Structured Outputs Handbook
> strongly believe that structured outputs are one of the most underrated features in LLM engines

Structured output is really the whole foundation of lots of our hopes and dreams. The JSONSchemaBench paper is fairly preoccupied with performance, but where it talks about quality/compliance, the "LM only" scores in the tables are pretty bad. This post highlights the on-going difficulty and confusion around doing a simple, necessary, and very routine task well.

Massaging small inputs into structured formats isn't really the point. It's about all the nontrivial cases central to MCP, tool-use, local or custom APIs. My favorite example of this is every tool-use tutorial that's pretending that "ping" accepts 2 arguments, but, it's actually more like 20 arguments with subtle gotchas. Do the tool-use demos that correctly work with 2 arguments actually work with 20? How many more retries might that take, and what does this change about the hardware and models we need for "basic" stuff?

If you had a JSON schema correctly and completely describing legal input for say ffmpeg, then the size and complexity of it would be approaching that of kubernetes schemas (where JSONBench compliance is only at .56). Can you maybe yolo generate a correct ffmpeg command without consulting any schema with SOTA models? Of course!, but that works well because ffmpeg is a well-documented tool with decades of examples floating around in the wild. What's the arg-count and type-complexity for that one important function/class in your in-house code base? For a less well-known use case or tool, if you want hallucination free and correct output, then you need structured output that works, because the alternative is rolling your own model trained on your stuff.

robot-wrangler··on Every country should set 16 as the minimum age for social media accounts
Is it? If kids grab the top result of a massively A/B tested algorithmic feed trying hook people and/or maximize the controversy / engagement, then arbitrage it onto another simpler less gamified platform... is that really human curated? There's some truth to the idea that, as long as there is any social media, everything is social media.
robot-wrangler··on 1000 Blank White Cards
Nothing really specific.. just some kind of of relatively tidy insight in classical terms. Game-trees, clearly articulated dominant strategies, surreal numbers.

You could model most games with anything simple that's convenient (trees, state machines, term-rewriting systems). Meta-games, dynamic protocols, and multi-agent systems are broadly related but also different animals where you might need sims, full-blown process calculi, weird new kinds of logic. Depending on where you land a natural model candidate might be messy, maybe you have to give up things like completeness or decidability. Maybe the closest fit for a formalism here is dynamic deontic logic: https://www.cse.chalmers.se/~gersch/jlap2012.pdf

robot-wrangler··on 1000 Blank White Cards
> anyone can at any time create a card that says "I win the game" but that is no fun [..] The metagame of 1kbwc allows the deck to become its own story

Yep exploring this question collaboratively is of course the real activity. Depending on your perspective it's barely recognizable as a game, or it's the ultimate / only game. Also kinda related here is Carse on finite and infinite games and Wittgenstein on language games[1,2]. It is "only" philosophy, but also feels ripe for more rigorous treatment

Presumably a good theoretical treatment would try to look at how games and their meta's are related: how the number and stability of rules changes the richness of interaction, enjoyment, flexibility in strategy, average duration and tolerable length of game-play, etc

[1] https://openlibrary.org/books/OL22379733M/Finite_and_infinit... [2] https://en.wikipedia.org/wiki/Language_game_(philosophy)

robot-wrangler··on 1000 Blank White Cards
This is a meta-game. I got curious about related topics in game theory once and found out about [1,2]. There are also a few papers directly trying to study calvinball and so-called minimal-nomic. It's pretty crazy how little we know theoretically about this stuff, considering how relevant games with dynamic rules actually are for daily life.

Of course, there's probably no clean solutions in this space short of lots of sims. Regardless of whether new agentic stuff works for everything else in AI.. agent-based modeling seems likely to benefit from some kind of renaissance and that should be really interesting.

[1] https://en.wikipedia.org/wiki/Constitutional_economics [2] https://en.wikipedia.org/wiki/Mechanism_design

robot-wrangler··on Don't fall into the anti-AI hype
> you CAN get good results for problems that can be reduced to a robust conformance suite.

If that's what is shown then why doesn't it work on anything that has a sufficiently large test-suite, presumably scaling linearly in time with size? Why should we be selective, and based on what?

robot-wrangler··on Don't fall into the anti-AI hype
AI Engineer to Software Engineer: Humans writing code is a waste of time, you can only hope to add value by designing agentic workflows

Prompt Engineer to AI Engineer: Designing agentic workflows is a waste of time, just pre/postfix whatever input you'd normally give to the agentic system with the request to "build or simulate an appropriate agentic workflow for this problem"

robot-wrangler··on Don't fall into the anti-AI hype
So extrapolating from here and assuming applications are as easy as libraries, operating systems are as easy as applications.. at this rate with a few people in a weekend you can convert anything to anything else, and the differences between different programming languages are very nearly effectively erased. Nice!

And yet it doesn't feel true yet, otherwise we'd see it. Why do you think that is?

robot-wrangler··on Don't fall into the anti-AI hype
> How would we measure the effects of AI coding tool taking over manual coding ?

Instead of asking "where are the AI-generated projects" we could ask about the easier problem of "where are the AI-generated ports". Why is it still hard to take an existing fully concrete specification, and an existing test suite, and dump out a working feature-complete port of huge, old, and popular projects? Lots of stuff like this will even be in the training set, so the fact that this isn't easy yet must mean something.

According to claude, wordpress is still 43% of all the websites on the internet and PHP has been despised by many people for many years and many reasons. Why no python or ruby portage? Harder but similar, throw in drupal, mediawiki, and wonder when can we automatically port the linux kernel to rust, etc.

robot-wrangler··on Don't fall into the anti-AI hype
I agree about skills actually, but it's also obvious that parent is making a very real point that you cannot just dismiss. For several years now and far short of wild AGI promises, the answer to literally every issue with casual or production AI has been something like "but the rate of model improvement.." or "but the tools and ecosystem will evolve.."

If you believe that uncritically about everything else, then you have to answer why agentic workflows or MCP or whatever is the one thing that it can't evolve to do for us. There's a logical contradiction here where you really can't have it both ways.

robot-wrangler··on Don't fall into the anti-AI hype
Let's maybe avoid all the hype, whether it is for or against, and just have thoughtful and measured stances on things? Fairly high points for that on this piece, despite the title. It has the obligatory remark that manually writing code is pointless now but also the obligatory caveat that it depends on the kind of code you're writing.
← PreviousPage 3 of 8Next →