OpenRouter is valuable because it has a large catalogue of competing providers for a fungible service. This does the opposite (lock in with a specific vendor).
pyenv has been standard tooling for far longer than uv. depending on package manager supplied Python packages only makes sense if you’re running rhel or Debian or something and your application is packaged/deployed/the maintenance path uses dnf/apt. Otherwise you should always use a venv and use an out of package manager update mechanism. Like, in a broader sense, vendoring dependencies only makes sense if you’re shipping an application, not on a dev box.
That's what the GP means as well, you can use pyenv and uv to install multiple versions of Python and create envs with whichever version you want to use.
Might be an indication that your task unit is too unstructured or your code base is a mess.
Is your actual code doing something complex or is this incidental complexity?
Relying on the model’s “intelligence” to patch over these issues hasn’t proven to be a reliable strategy for me. Of course, this might not apply to you, just my 2 paisa.
A well structured architecture, repo, and requirements document with concrete small deliverables can be competently delivered with cheap Chinese models. We rely on frontier models so that we don't have to spend several days/weeks on planning/architecture/documentation. That's their value proposition: superior intelligence. If they can't deliver that, they're useless at the current price.
Sometimes it is a cause vs causation. Sometimes the scientist didn’t adjust for a variable that clearly would impact both fields they were measuring. To make such a claim, I think it’s appropriate to name that hypothetical third variable. Otherwise the comment is so general it applies to all statistical studies.
Considering the importance of the subject, it still seems important to bring this out, especially if there's conclusions that if you have irregular sleep you are significantly increasing your odds of dying earlier. I think intuitively we can know how important sleep is and I desperately want to sleep better.
It's fine to speak of it being predictor etc, and all that, then they end the abstract with "Sleep regularity may be a simple, effective target for improving general health and survival.", which is technically fine since they use "may", although later articles based on the study will likely make the statement more confident.
But I guess what might be slightly triggering is claiming that it's a "simple" target. Don't I wish I could sleep on command and better?
So a clear question is - why do people choose to sleep or why do they naturally sleep irregularly?
Because for that there must be a logical cause in the first place. They say they control for mental health and all that, but is it then that ultimately it comes down to preference in their mind? I'd think most people want to sleep in healthy way.
Basically - if they were able to control for all possible confounding variables, what exactly was the cause of irregular sleep?
Anecdotally I can say that I sleep more irregularly the more stress there is, and stress could easily affect health, but if they controlled for stress, what then?
I guess ultimately they are saying it's a desirable target to measure, so it's fine in that sense. They are not really saying that choosing to sleep irregularly is what is causing the issues.
But you can just apply this to anything. I feel unlike unless you're an insider with skin in the game, your criticism doesn't land other than a generic surface one.
To be clear, I apply an equal deep skepticism to most fields that aren't math (in the sense of a priority) or physics (in the sense that you aren't trying to study the entire world, but a specific set of phenomena that you can reliably control enough + repeat to run intervention on), whether the results agree with me or not. Maybe a bit of intellectual closed-mindedness. But then that means that me, personally, I can't in good faith use the criticism as a proper 'debunk' argument - at best it's a heuristic to avoid spending cycles to evaluate it (which is 'rational' behavior, as much as I hate that word, IMO).
People love nitpicking about scientific articles, it kind of exploded with the pandemic. Especially when the conclusions of the paper don't align with the expectations of the reader.
The way I see these is "these persons invested a lot of time putting this together and all I have to counter it are personal vibes", so unless my LLM of choice can find plenty of conflicting papers, I tend to assume it's reasonably valid work.
Would love to see a causal model [0] to help better understand all of the mediators considered as well as confounders. I'm close to finishing up an interesting read from Judea Pearl/Dana Mackenzie - The Book of Why: The New Science of Cause and Effect [1]. Talks alot about Causal Models, Causal Inference, the 3 ladders of causation, etc. I liked the graphical approach to help outline exactly how one thinks about direct and indirect effects and how it facilitates counterfactual analysis and causal mediation analysis.
Do you even bother to click on links before asking a question?
There is a whole paragraph on "Statistical analysis" that even provides five supplementary methods (S1.5, S2.2 and S2.4-6) if you want detailed information.
What I wonder is even if they did the perfect adjustments, what would have been driver for different sleep regularity.
E.g. considering some common causes like work stress - if they did the causation and compared people who did the same type of work, and they controlled for stress levels then why did one group of people have sleeping regularity issues more than compared to others?
Like there has to be some other driver then that they didn't control for, as in personality, environmental or physical difference?
Most people do want to have healthy sleep, the ones who don't usually have something causing those issues.
Yes, but I always found that objection a bit silly. It's like pointing out that real cows are obviously not perfect spheres nor do they live in a vacuum.
> [...] if prices perfectly reflected available information, there is no profit to gathering information, in which case there would be little reason to trade and markets would eventually collapse.[2]
That's a stupid way to formulate this. Markets wouldn't "collapse". They would get slightly less efficient until equilibrium is restored to where arbitragers can make enough money to keep prices at that level of efficiency.
Maybe not "collapse" in a the sense of going to zero but if there was no profit to trading, then the quant trading industry would not exist, trading profits would collapse.
Meanwhile Two Sigma is hiring alpha quants to be AI research scientists at $250k starting salary + bonuses.
Even if we're just talking about the HFT/sell-side, there clearly exist various anomalous inefficiencies that can be exploited.
As I said, if we woke up this morning and prices were magically efficient in an idealised sense, at most a few quants would go home and retire early, and tomorrow we'd be back at the level (in-) efficiency that allows people to be market makers.
How can prices reflect all available information if there's no profit to collecting the information and there are no informed quant traders? Who is collecting the information exactly so that prices can reflect it and what is their incentive for doing so? Efficiency doesn't happen magically or automatically - traders create it. It's like a kaggle contest* to process information, with the incentive being profit.
You don't believe in the existence of residual return orthogonal to priced cross sectional risk factors (alpha)? E.g. Trends, momentum, volatility clustering, etc. many easily demonstrable inefficiencies. VPIN and order flow toxicity are highly predictive features. Most HFT MM especially in crypto involves hybrid alpha in addition to the (visible) bid-ask spread, which it itself an "inefficiency" to compensate market makers like Jane Street and other successful firms that operate on the assumption that weak form EMH is not accurate.
I would have hoped that by now it was obvious that we are talking about a _specific_ weak form of the EMH that takes friction into account?
What is your whole first paragraph about? Who are you trying to convince? Where's the strawman that claimed that the strongest version of EMH that you can imagine is literally true?
There's no single weak form of EMH that could be accurate or inaccurate: there are many versions of the EMH in various strengths and dimensions (that can be accurate or inaccurate).
To be more specific: Jane Street believes (or acts lie they believe) that markets are at least efficient enough that it takes a lot of effort for them to make money. As a very, very weak form: someone doing chart astrology, eh, I mean technical analysis, on S&P 500 stocks won't beat the market. But even much stronger versions than this are defensible.
The real strong forms that say that all information is preciously reflected in profits is a simplifying assumption you can sometimes make to make your life easier. Just like you sometimes neglect friction in physics. But when you want to decide how long your train needs to emergency brake, you kinda need to take friction into account. Similarly, when trying to make money in the market or trying to understand how others like Jane Street make money, the strongest EMH is not a good guide.
Question is about EMH and how you expect efficiency to be achieved absent profit for collecting the information.
There are 3 accepted forms of EMH. I'm talking about weak form - just price history and nothing else. E.g. formulaic alpha have demonstrable predictive value in modeling.
All that to say you believe trading profits are real. Maybe you just need to learn more about what a buy side alpha quant at two sigma does for a living. Trading models can be robust and exploit real inefficiencies. Weak form EMH is demonstrably false on it's face, as you agree.
You're deliberately misunderstanding that you linked an article to EMH as informative and true, and then don't want to defend it. How do markets become efficient and reflect (any) information if nobody can profit by collecting information and trading on it? EMH states it's impossible to beat the market and that all available information is priced in. How, magically?
The way you're speaking about trading in terms of technical analysis implies you have retail trading exposure and have no idea what institutional alpha quants do.
I'm not sure most people are that naïve that they can't differentiate between "any computer that acts smartly" (how the term "AI" is used) and the word chatbot. Of course, LLM is even more precise
Tangential question: what do you call transformers-based models that generate images or videos? Are they LLMs? They're not really "language" models. But there's not really an easy term for them. Maybe "image models" and "video models"?
I'd call them video/image generators because I don't know the word for the underlying technology. I can't imagine they're LLMs because, indeed, the second L is for language
Today I learned, Stockfish moved to neural network on 2023. I knew that it was just a minmax with alpha beta pruning and a really good eval function. Now its not.
> I knew that it was just a minmax with alpha beta pruning and a really good eval function. Now its not.
It is still "just" a minimax with alpha beta pruning, except the eval function is now a neural network. NNUE, to be more specific.
I highly advise anyone who is curious about chess engines, but hasn't heard about NNUE to read about it. I find this technology absolutely fascinating.
The key idea is that a neural network is structured in a way that makes it very cheap to calculate scores for similar positions. This means that during a tree search, each time you advance or backtrack you can update the score efficiently instead of recalculating it from scratch.
I mean, it still is. Now it just has a really good neural net-based eval function. Don't be fooled: it's not that stockfish just has "a really good eval function", and that's the only thing that makes it as strong as it is. The actual tree search is _incredibly_ sophisticated, with boatloads of heuristics, optimizations, and pruning methods on top of alpha-beta.
