HNHacker News
TopNewBestAskShowJobs

YeGoblynQueenne

24,940 karma · joined September 26, 2015

This a common question on this board:

What is reasoning?

In computer science and AI when we say "reasoning" we mean that we have a theory and we can derive the consequences of the theory by application of some inference procedure.

A theory is a set of facts and rules about some environment of interest: the real world, mathematics, language, etc. Facts are things we know (or assume) to be true: they can be direct observations, or implied, guesses. Rules are conditionally true and so most easily understood as implications: if we know some facts are true we can conclude that some other facts must also be true. An inference procedure is some system of rules, separate from the theory, that tells us how we can combine the rules and facts of the theory to squeeze out new facts, or new rules.

There are three types of reasoning, what we may call modes of inference: deduction, induction and abduction. Informally, deduction means that we start with a set of rules and derive new unobserved facts, implied by the rules; induction means that we start with a set of rules and some observations and derive new rules that imply the observations; and abduction means that we start with some rules and some observations and derive new unobserved facts that imply the observations.

It's easier to understand all this with examples.

One example of deductive reasoning is planning, or automated planning and scheduling, a field of classical AI research. Planning is the "model-based approach to autonomous behaviour", according to the textbook on planning by Geffner and Bonnet. An autonomous agent starts with a "model" that describes the environment in which the agent is to operate as a set of entities with discrete states, and a set of actions that the agent can take to change those states. The agent is given a goal, an instance of its model, and it must find a sequence of actions, that we call a "plan", to take the entities in the model from their current state to the state in the goal. This is usually achieved by casting the planning problem as pathfinding over a graph with a search algorithm like A*. Here, the agent's model is a theory, the search algorithm is the inference procedure, and the plan is a consequence of the theory. Deductive reasoning can be sound, as long as the facts and rules in the theory are correct: from correct premises we can deduce correct conclusions. We know of sound deductive inference rules, e.g. A*, and Resolution, used in automated theorem proving and SAT-Solving, are sound.

The classic example of inductive reasoning is inferring the colour of swans. Most swans are white (apparently) so if we have only seen white swans we have no reason to believe there are any other colours: we are forced to infer that all swans are white. We may only be disabused of our fallacy if we happen to observe a swan that is not white, e.g. a black swan. But who is to say when such a magnificent creature will grace us with its presence, outside of Tchaikovsky's ballets? Induction is thus revealed to be unsound: even given true premises we can still arrive at the wrong conclusions. Another example is the scientific method: imagine an idealised scientist, perfectly spherical, in a frictionless vacuum. She starts with a scientific theory, then goes out into the world and makes new observations about a phenomenon not described by her theory. She constructs a hypothesis to extend her theory so as to explain the new observations. The hypothesis is a set of rules, where the premises are the consequences of the rules in her initial theory. Then, being an idealised scientist, she goes looking for new observations to refute her hypothesis. Science only gives us the tools to know when we're wrong.

Abductive reasoning is the mode of inference exemplified by Sherlock Holmes. We can imagine Sherlock and Watson standing outside a tavern in London, watching as a gentleman of interest steps out of the tavern with egg on his lapel. "Ah, my dear Watson, what can we conclude from this observation?". "Why my dear Holmes, we can conclude that the man had eggs for breakfast". Holmes and Watson can arrive at this conclusion, about a fact that they have not directly observed, because they have a theory with a rule that says "if one eats eggs, one may get some on one's lapels". Working backwards from this rule, and their observation of egg on the man's lapels, they can guess that he had eggs even if they didn't directly observe him doing so. Abduction is also unsound: the man may have swapped coats with an accomplice, who was the one who had eggs for breakfast instead.

And now you know what "reasoning" means. So the next time someone asks: "what is reasoning?", you can let them know and turn the discussion to more interesting, more productive directions.

submissionscomments
YeGoblynQueenne··on More questions about whether researchers can trust OpenAI with unpublished math
>> Internal OpenAI models are reportedly solving open problems at a surprisingly fast rate[2]

Maybe I'm failing to read that graph properly but the y axis says "pass rate" and it only goes up to 0.5. That would mean every single problem is at most half-solved.

I don't know what that means though. What is "0.5 pass rate" in the context of "open math problems" (as in the graph title)?

YeGoblynQueenne··on More questions about whether researchers can trust OpenAI with unpublished math
The brute-forcing is a good, old-fashioned generate-and-test approach like in Simon and Newell's Logic Theorist, which was presented in the Dartmouth convention in 1956, where AI was named by John McCarthy. Logic Theorist caused a big stir by (re) proving several of the theorems in Principia Mathematica by Russel and Whitehead.

There was much excitement, then, as now, for this kind of approach and there were several systems that followed along the same lines, e.g. Automated Mathematician by Doug Lenat.

Eventually it became clear that this approach is limited by what it can generate: you may have a sound and complete verifier, but if the generator, i.e. the first step in the generate-and-test pipeline, is incomplete, then the entire thing will run out of steam sooner or later.

The difference with LLMs is that they are... well, large. They are the most powerful generators ever created. That means their limits are not in sight and it will probably take us a very long time to find them.

Which is all to say that, yes of course, automatic verification is indispensable. But without an LLM generating an unprecedentedly large number of plausible theorems, there would be no AI mathematics, or in any case AI mathematics wouldn't have gone as far as it has.

YeGoblynQueenne··on Growing proof that autonomous cars save lives
Thanks. Well, that doesn't sound independent at all.
YeGoblynQueenne··on Growing proof that autonomous cars save lives
>> In March, Waymo’s independent study showed 92 percent fewer fatal or serious-injury crashes, a 13-fold reduction versus human drivers in comparable city environments.

I can't progress beyond this point in the article. How is Waymo's study "independent"? I don't understand.

YeGoblynQueenne··on Navier-Stokes – Tristan Buckmaster [pdf]
You're also free to plagiarise anyone you want. There are no laws against it on most jurisdictions.

Also brain raping* is not illegal in most jurisdictions.

But they're both deeply disturbing.

_________

* https://youtu.be/JlwwVuSUUfc?si=uWl4-LCHAeI7qtb3

YeGoblynQueenne··on Navier-Stokes – Tristan Buckmaster [pdf]
Your comment was greyed out when I saw it earlier, maybe you missed some downvotes?

About the plagiarism issue, I model it as OpenAI being an advisor and their AI a PhD student. If the advisor puts their name on a paper behind that of their PhD and it turns out the PhD copied the text of the paper from somewhere else the advisor is also responsible of plagiarism, not just the student. The least the advisor can do is withdraw their authorship from the paper.

But, yeah, point well made: it could be much worse than that. Like an advisor instructing a student to copy someone else's paper.

YeGoblynQueenne··on Navier-Stokes – Tristan Buckmaster [pdf]
>> Does openAI train on user conversations in general? I assume so. But so fast as that? That seems unlikely in general. I expect OpenAI will come out denying this.

How "fast" does it have to be? Buckmaster and Alpoge have been working on this for just a day short of a year. See Alpoge's tweet announcing his collaboration with Bukmaster dated 9/19/25:

https://x.com/__alpoge__/status/2097206973418611054

It takes a few months to train a model these days but not a whole year. OpenAI had all the time to train on Buckmaster and Alpoge's results of just a few months earlier at which point they must have been well on the path to their result.

YeGoblynQueenne··on Navier-Stokes – Tristan Buckmaster [pdf]
Duh. There are supposed to be limits to what OpenAI is allowed to access with respect to logs and user interactions but there is no technical limitation.

It's a bit like sending unencrypted messages through a messaging app and the developer having a TOS that says they don't look at your messages. They might not, but they are fully capable of doing so. If they have a reason to do it, they will. Nobody's stopping them.

YeGoblynQueenne··on Navier-Stokes – Tristan Buckmaster [pdf]
That's absolutely right. Why the downvotes? If OpenAI are using but not acknowledging the work of others that's plagiarism. If they don't know for sure, but aren't performing due dilligence to make sure they aren't, that's also plagiarism.
YeGoblynQueenne··on Navier-Stokes – Tristan Buckmaster [pdf]
>> I asked whether the model had been trained on, or had access to, our sessions in Codex, into which we had been putting all our drafts for the whole of this project. I was told the model did not look up user data. I asked again, about training, and I did not get an answer.

This sounds like a very big coincidence and it looks really bad for OpenAI but there is an alternative explanation that I can only state as a conjecture.

Suppose that the ability of LLMs to generate mathematical proofs is like a quiver full of arrows: each arrow, one proof. The same quiver is shared between all instances of one model and substantially similar models share substantial subsets of the arrows in the same quiver.

That would allow two independent teams to converge on the same LLM-aided solutions to the same problems. Even more likely so if the quivers were small and finite and their arrows were specific to a distinct class of problems (without being able to suggest a particular class from what we've seen so far).

This would explain the kind of LLM-mediated results we've seen so far that tend to be ... sparse. By which I mean that every time there's a new model release we get some new results and then they seem to dry out, until the next release.

It would also explain how OpenAI was about to prove the same result as Buckmaster and Alpoge, while absolving OpenAI of any misconduct. And this is one reason to prefer this explanation: one should not favour accusations of misconduct as long as there are conceivable alternatives.

But, that's just a conjecture that I can't prove.

YeGoblynQueenne··on A Tesla ran a stop sign and killed a man, Full Self-Driving/Autopilot was on
@dang @tomhow can you please de-flag?

Also can you please check whether articles about Tesla with FSD or Autopilot tend to get flagged a lot? It seems so to me but I can't tell for sure.

YeGoblynQueenne··on A Tesla ran a stop sign and killed a man, Full Self-Driving/Autopilot was on
>> Consent. We've consented to people learning to drive. Everyone is aware of those risks and they act accordingly. Often, to make other drivers aware of the driver's learning status, it's customary to affix a sticker to the car.

It's not juts customary. Some jurisdictions (probably outside the US) require a special sign meaning "new driver" to be displayed in the car for a period.

E.g. in Greece where I stay an "N" (for "νέος" or "new") must be displayed for 12 months after a driver first obtains their license provided they're 18 or over:

https://www.magentainsurance.gr/car/odigos-asfalisis/sima-n-...

(link in Greek)

Drivers who are 17 years old must display a different sign (a "Σ").

YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
That's right. Kefir is essentially a mesophilic yogurt. I don't pretend to understand why but it seems the difference in texture has to do with the different organisms that produce one vs. the other (although the thermophilic bacteria used in yogurt-making -Lactobacilus bulgaricus and Streptococcus thermophilus I think- have also been found in kefir; but I don't know if they are very active given the lower temperature).

To be fair, back when I used to make kefir it often came out quite firm to the point that I could lift it out of a bowl with a perforated laddle and eat it with a spoon. What seemed to make the difference was using non-homogenized milk rather than homogenized. Using unhomogenized milk from Jersey cows also seemed to do the trick but the result was a bit too fatty.

YeGoblynQueenne··on Babylonian Lamb Stew with Beets (1750–1730 BCE)
>> The bra went in the bin.

XD My friend got her skirt munched. While wearing it.

I totally think it's like you say: the goats have never learned to be milked and now they'll never learned. The new generation might, but who's going to teach them? Not me because I'm only there half the year (when it's hot and I don't want to move).

>> As an aside, you probably want to get yourself a copy of "Self-Sufficiency" by John Seymour, and have a read of that.

Thanks. I'll have a look.

YeGoblynQueenne··on Babylonian Lamb Stew with Beets (1750–1730 BCE)
I'd say that in Greece the dominant meats are pork and chicken. That's the main meat in all the grill houses and the meat in gyros and souvlaki. Lamb is beloved but eaten more rarely, particularly in religious feasts, especially Easter Sunday where the tradition is to put a lamb on a spit and eat everything off it down to the bones (except the entrails that are made in a special soup to end Lent, or various sweet meat skewers like kokoretsi, gardoumba, splinandero, etc). It's also more expensive than pork and chicken so it tends to be left for special dinners at home.

But, honestly, nothing beats a nice pilafi with lamb, with garlic wedges and slowly cooked until the meat falls off the bone.

YeGoblynQueenne··on Babylonian Lamb Stew with Beets (1750–1730 BCE)
Wait 'till you hear about melon - "μήλον" i.e. "apple" in Greek.

Melon, watermelon, pomelo, etc. The ancient Greeks themselves started that with the "golden apples of the Hesperides" a.k.a. oranges.

I guess any fruit people didn't know how to call was ... some kind of apple.

YeGoblynQueenne··on Babylonian Lamb Stew with Beets (1750–1730 BCE)
I thought animal farming in general is an arcane science.

My friends have goats but they aren't farmers and they haven't got a clue how to care for them. The females sometimes are left with milk in their breasts after their young are weaned and one of my friends will get anxious about that and ask us to help her milk them. It's impossible. You can't milk a goat that doesn't want to be milked. If you push them too hard they lie down so you can't reach their teats and that's that unless you want to have three people lifting the goat up and another one trying to milk her while she squirms and kicks. I don't trust the milk either. Once in a while they succeed in milking a goat and when I tried the milk (after boiling it) it was sort of gritty, like it was mixed with sand. It was not impurities from the environment because I filtered it through a muslin so I think the animal may have had some infection like a mastitis or something that affected the milk's texture.

I bet farmers know exactly what to do about all that or probably they know how to get the animals used to milking when they're still young so that the won't resist it that much. I've watched videos of goats being milked and it looks almost effortless when a pump is used. The goats just line up in their feeding trays and then someone walks up to them from behind, pumps out their tits and moves to the next one, bish bash bosh. My friends don't even have a dedicated structure to feed the goats, we just throw hay for them on the ground so they end up fighting.

My friends really have no idea how to care for the goats, it's rather sad. The owner of the farm has even said he wants to get "wild" goats that don't need to be cared for. By that he means a herd of goats that have gone feral and roam the hills in a village nearby. He thinks because nobody actively cares for them and the herd persists that must mean the goats are somehow self-sufficient. I think that's a big misunderstanding. Domesticated animals need a lot of care one way or another.

I think I'd prefer to have sheep though, if I had animals. Sheep look much nicer than goats.

YeGoblynQueenne··on “Next-token predictor” is the wrong mental model for LLMs
You're right of course and my bad but in my defense even Sutton & Barto lapse into using the term interchangeably (I caught them at it while reading the book earlier).

>> And, the expected return from a state is often estimated rather than an explicit trajectory run out all the way. A value function can estimate expected return without explicitly predicting the future states or individual rewards that make up that return.

I think rather that's the job of the value function, to inform the policy of the, well, value of a state, or state-action pair. Somehow the policy needs to know what a good action is to take in the current state, or what is the best action to take. And of course it doesn't have to run the entire episode before it does that: but that's what I mean that it predicts the next move and not just the outcome of the entire game.

YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
Sharing is caring.
YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
Yes that was my argument and it stands. The Kalona yogurt you showed is not UPF but I think the honey might be only in the bottom. The lafermiere yogurts are UPF.
YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
Nobody said that pasteurization makes anything ultra-processed.

Pasteurised milk is classified as Group 1 (Unprocessed or minimally processed food) in NOVA. Pasteurised cream is the same thing [1].

Pasteurisation doesn't have any stabilising effect.

I argued that the pasteurised cream added to the ingredients of the lafermiere yogurts is a stabiliser. The stabilisation effect comes from the increased viscosity of milk with added fat that helps the emulsion of milk and flavourings keep stable for longer than normal while milk sets during yogurt making.

I did not argue that the stabilisation effect comes from the fact that cream is pasteurised.

I argued that the lafermiere yogurt is a UPF because of the addition of pasteurised cream as a stabiliser.

I did not argue that the lafermiere yogurt is a UPF because of the pasteurisation of the added cream; or the milk itself for that matter.

And now I have to ask: in what sense could anything I said before be taken as a claim that pasteurisation is ultra-processing or that the lafermiere yogurts are UPFs because of pasteurisation?

_____________________

[1] If anything it has the opposite in that pasteurised milk tends to produce looser, softer curds than raw milk because it weakens protein bonds and causes calcium to leak out (soorta), hence the addition of calcium chloride to milk when making cheese.

Cream is usually made by centrifuging milk to separate the fat out, which should count as minimal processing and then pasteurisation itself is minimal processing on top. I doubt pasteurised cream counts as a UPF.

To be fair that doesn't really matter in yogurt making because milk is boiled anyway, which denatures milk proteins and we don't care if they leak calcium or not because we're not trying to make curd, as such, but yogurt. Not withstanding the fact that people in the Indian subcontinent call yogurt, "curd". Just a linguistic quirk.

The fact that lafermiere list milk and cream as "pasteurised" in their ingredients is probably something to do with marketing, to make it clear they aren't using raw milk, for whatever reason that might be necessary to clarify.

My fresh Rodopi ewe's milk yogurt lists as ingredient "full organic ewe's milk". They say nothing about pasteurisation and I doubt they even bother to pasteurise it since they're going to blast it with considerably more heat than for pasteurisation anyway. OtoH the difference may not be marketing but the fact that Rodopi get their milk directly from producers while lafermiere buy it second hand from dairy companies that sell it already-pasteurised. I mean the pasteurisation doesn't hurt. It's just unnecessary. So it's a bit weird that they clarify it in the list of ingredients.

YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
I think the confusion is that you're loosening up the relation: food can be unhealthy without it being UPF. The relation is an implication, not an equivalence: UPF implies unhealthy. Unhealthy doesn't imply UPF.

So e.g. food with too many calories as in your example can be unhealthy because it can make you fat without it having to be a UPF.

Are you saying that you want a classification that flags up every kind of unhealthy food?

YeGoblynQueenne··on “Next-token predictor” is the wrong mental model for LLMs
Hey man, I'm not crazy. Yes, the goal is to maximize the cumulative reward like you say but to do that a policy (agent) has to take the actions that maximize its expected return in each time step. That's what the discount factor applies to, the expected return.

To maximize its expected return it has to predict that taking action α in state s in time step t will produce state s' and reward r in time step t+1. Which btw it can predict by estimating a value function that sums over the expected return from time step t onwards. The value function essentially tells the policy what state it wants to be in and what action to take to get to that state, in order to eventually achieve its goal (represented by max cumulative reward at the end of an episode).

So, yes, like you say, the whole policy is a game result predictor but the decision making process is a next-move predictor. The policy must make the locally optimal decisions to get to the globally optimal result, if I may be so bold. It's the same problem with a different solution as in heuristic search, except you have a reward function instead of a heuristic cost function.

Sutton & Barto 2ed is free here btw:

https://web.stanford.edu/class/psych209/Readings/SuttonBarto...

The value function stuff is in Section 3.7. Now if someone could explain the grid world example in Figure 3.5 to me that would be great because I'm very confused about the fact that any action at state A or B takes the agent to state A' or B' respectively, since those are not even adjacent to A and B. I'll go ask chat I suppose.

YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
It's late and I have to go to sleep so answering here.

>> No. Pasteurization isn't a UPF technique. See the problem?

My argument is that the extra fat is used as a stabilizer that wouldn't need to be included if the maker wasn't trying to force their yogurt to hold other stuff in it than milk. The use of a stabilizer makes it a UPF, not the fact that the cream is pasteurised- that makes no difference as you say.

YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
I don't get that. AFAICT UPF is about the way foods are made, not how many calories they have.
YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
So, I haven't made any yogurt in a while but from memory it takes about half an hour for milk to set, depending on what milk and what yogurt starter etc. No way you can make an emulsion of e.g. milk, sugar and vanilla that is stable for that long. So either you have to use stabilizers, or you have to accept that your flavourings are not going to be evenly distributed throughout your yogurt.

I can't know the latter since I can't buy that brand in Greece (for rather obvious reasons it's not imported here). If I look at the list of ingredients for e.g. the Vanilla Bean yogurt I see "Pasteurized Whole Milk, Pasteurized Cream..." and that's your stabilizer there.

How much is there? The "Nutrition Facts" label on top lists "Total fat 12g" per serving which is "1 container" of 140 grams. That's ~9% fat, so about 5% more than plain cow's milk (at 4g per 100g). The total fat is more than the fat in ewe's milk (about 6%). I ... guess the extra fat makes the milk and vanilla bean powder emulsion last a bit longer so it gets better mixed in during the fermentation time?

Dunno. The packaging says "Traces of vanilla seeds and marbling on top of the yogurt are the results of a natural infusion process". The other flavours don't have a similar disclaimer but they all have extra fat in the form of pasteurized cream.

So anyway you get both a stabilizer and uneven distribution at least for the vanilla. If you have a strong enough flavour I reckon that's not a problem though.

EDIT: (The fruit-at-the-bottom variants are presumably UPFs).

They don't have to. If you're just mixing some fruit in milk and then fermenting it to make yogurt, that's just processed (going by NOVA AFAIU). But looking at the ingredients list I see e.g. "Corn starch, Natural Flavor, Lemon Juice Concentrate, Pectin" etc. Ew. No, that's a UPF. You don't need to put that stuff in yogurt. Not if the point is to enjoy it.

YeGoblynQueenne··on Can AI design circuit boards yet?
I guess I walked right into that one.
YeGoblynQueenne··on Pushdown automata in statistical machine translation (2014)
Oh absolutely. I'm just curious because we never see articles like that on HN, so I wondered if something happened that made it newsworthy.
YeGoblynQueenne··on “Next-token predictor” is the wrong mental model for LLMs
Alright, I'll have to check up on that. Thanks for being nice about it.
YeGoblynQueenne··on How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025)
Ewe's milk yogurt is more caloric than plain cow's milk yogurt (because of higher fat content). Does that make ewe's milk yogurt "worse for you than UPF yogurt"?
← PreviousPage 5 of 34Next →