TA here. Biggest changes are in the second assignment (distributed) where we added a bunch of memory, profiling and distributed tasks, as well as in the fifth assignment (alignment), where most of the RL tasks are fresh this year. Assignment 3 (scaling laws) was also completely updated, but in a way that might be difficult to run without substantial resources. I'm working on a way for external students to be able to run simulated experiments for free!
Assignment 1 (basics) has the most hours of preparation invested in it, and only minor modernization/bug fixes were necessary this year.
We have autograding for code through tests written by hand, and additionally do manual code audits if we see suspicious behavior. We also do grading the old-fashioned way for writeups.
We do indeed catch students who don't follow the honor code. It's very obvious from how the code looks, as well as the rate of progress. Since we use Modal for class submissions, we have code deltas for every time they run something on B200s. The diffs often contain something like 300 lines in 5 minutes, in which case we review and report based on how egregious/provable it looks.
At least for Nix itself, that's pretty much it except via Dutch.
> The name Nix is derived from the Dutch word niks, meaning nothing; build actions do not see anything that has not been explicitly declared as an input
Also, I think the founder's username in various places is nixnut. Which to an English-only speaker means someone crazy about Nix (Nix fan). However in Dutch 'niksnut' or 'nietsnut' loosely translates to 'bum'.
"gar" is a useful amplifying prefix in German that can be used in all kinds of situations and I think it lacks a direct equivalent in English. Unlike totally, for example, gar can only stand alone in very specific contexts and usually is used more like an intensifying prefix.
I don't know about that. Even Harvard has a big grade inflation problem. And non-elite colleges are trying to make it as effortless as possible to get a degree.
grade inflation is the right thing to do as long as employers and post graduate schools keep looking at grades or gpa. if you do strict "fair" evaluation you put your students at a disadvantage compared to same level students at other schools that grade more relaxed. grades should be feedback not something to compare with others instead we should set up a standard state exam (pass/fail, unlimited cheap retakes) to decide if you get a degree. but until that happens keep on inflating
My worst interview was bringing my interviewer to a state of shock and hyperventilation from pointing out an error in their math formula they confidently showcased after telling me my prior answer was wrong. 10 minutes of intense breathing with a shocked expression on their face appeared, I was worried the person would collapse in front of me and didn't know what to do. In the end the interviewer collected themselves and told me they knew from the start I was right and were just testing me. And I instantly knew I failed that interview loop.
Not as bad as it is now. All I see are suggested posts from people I never connected with and those are full of instagramesque self-promoting banal vibes.
Paradoxically, nicotine has some medical use in e.g. displacing viral debris and autoantibodies from nAChR (nicotinic acetylcholine receptors) due to having highest affinity to these receptors, which seems to help with (long) Covid; "smoker paradox" in lower covid-related hospitalizations.
A molecule can be socially and medically associated with a very harmful delivery mechanism, while still having specific biochemical effects that are worth studying on their own. Ant that's interesting
I was expecting to see a longer list of medical uses, but the wiki says that nicotine has performance impacts on cognition, improving fine motor motion and memory.
They changed how they make chewing tobacco (aka moist snuff) about 20 years ago and it has less of the cancer causing stuff (Nitrosamines) in it, its now closer chemically to snus - I’ll point out that Scandinavian countries have some of the highest use of oral tobacco in the world, yet last I looked some of the lowest incidences of oral cancer per capita.
The function of if tobacco causes cancer has as much to do with processing (it used to be cured by wood fire at a higher temperature which is where much of the carcinogenic properties came from) and the byproducts that processing creates, particularly Nitrosamine, its now cured differently in a process which is closer to snus, and somewhat safer.
Nicotine addiction (which I have) should be about harm reduction first, cigarettes are the only product that I can think of if used as commonly used will kill you or dramatically shorten your life, and it probably wont be cancer, it’ll be COPD, heart disease, or other cardiovascular issues - which are the same issues firefighters get from repeated smoke exposure. Breathing the byproducts of combustion is what’s really awful (and deadly).
It's a serious concern and switching to synthetic nicotine products may prolong your life. All tobacco products are highly carcinogenic. Contrary to what was said earlier it is not really about the smoking (though of course that makes things worse).
Nicotine products aren't safe; they are highly addictive and may exacerbate tumors that are already there. But they're far less addictive than tobacco products and they probably won't kill you.
I agree with you that tobacco is uniquely harmful, but smoking by itself is also bad by itself. Even exposure to smoke from campfires, if chronic, will elevate your risk for COPD, cardiovascular disease, lung cancer, etc.
Smoking/smoke inhalation is not safe, it leads to all the bad outcomes you mention, and I encourage everyone to seek out alternatives like oral nicotine pouches and cannabis gummies. They will almost certainly improve your healthspan and may well improve your lifespan. I personally stopped smoking because my wife told me she wanted me around longer. I am all aboard the "smoking is bad for you" train.
But the role of the smoke itself is overemphasized which leads to a false sense of security. Switching from cigarettes to dip does lead to a significant improvement in mortality. But the real step change is moving to oral nicotine pouches, gum, patches, etc.
I'd also point out the risk from campfires and from cigarettes is not at all comparable. They are several orders of magnitude apart. Even smoking marijuana isn't nearly as dangerous as smoking tobacco. (Smoking marijuana is not safe, that just goes to show how ridiculously dangerous cigarettes are.)
Becoming addicted to a substance is harm, so being addictive is a risk.
Nicotine is not harmless. It is a teratogen, it may exacerbate cancers you already have, it can harm brain development in young users, it can cause high blood pressure, etc. And as stated previously - it is not good to be addicted to something! That is a bad health outcome in and of itself!
I occasionally use synthetic nicotine products, I don't judge people for using them, but let's not misrepresent what this is. It is a drug. If you take a look at the risks and decide it's worth it more power to you. But don't tell people it's harmless, that is dangerous misinformation.
I don’t think this really holds up, for example helium itself is chemically inert and not toxic. The main risk from inhaling helium is probably oxygen displacement at a push.
Millions of people have been using inhalers to control asthma too, this well studied and agreed to be safe. This is just off the top of my head.
The HN guidelines say "Please respond to the strongest plausible interpretation of what someone says". The way I understand it is that different people on the thread often have different ways of thinking about the topic, and we shouldn't dismiss something because it's not what "we" were talking about. In this case, it was obvious to you that the parent was talking about smoking tobacco, right? So you can either engage with it, or not, but there's no need to reject someone's comment for not adhering to what you decided is the topic.
Who's trying to avoid talking about tobacco? TFA is titled "Scientists solve 200-year-old puzzle of how tobacco plants make nicotine", and my sense is that the thread is very much about which of the effects of smoking are explained by nicotine, and which are better explained by other factors of smoking tobacco.
Can it instruct DeepSeek during an LLM call to start removing old tool calls from the context instead of waiting for the LLM call to finish if the context size approaches DeepSeek's dumb zone? Claude Code can't do that, /compact can only happen after the LLM call; it's often preferable to start cleaning up context during an LLM call, especially when tool calls are huge like reading markdown files; implementation-wise all that is needed is to start removing earliest <tool call start> ... <tool call end> and replacing them just with some log entry stating this tool call was already performed, then re-running KV cache prefill (so the "online" compaction would get 0.5s latency hit every time it's performed). That way one can read 1000 files in one LLM call.
Not trying to be paranoid, but losing recorded history raw as it was originally reported could lead to quick AI-assisted rewrites in the archives of news outlets to fit whatever narrative of the "jour" is in fashion/that powerful of those times want. We are already seeing it in new editions of some old books that suddenly miss some currently controversial topics. History is written by the victors could change to history is rewritten by the (current) victors, as they see fit.
That was an option, but having decided on a true server chassis for other reasons, it made sense to use server-edition cards to take advantage of all those fans. I downclock them to 300W anyway for longevity, but it's nice to have the option to go to 600W if needed.
The server is going to live in the garage, so I'm not that concerned with noise. But I had no idea what to expect when I flipped the switch for the first time. It sounds like something out of the Book of Revelation. No way, no how could something like this be used in an inhabited area.
This is really not so clear cut as "fair use" might cover 99% of all data scrapping; you are not reproducing the originals just use them to estimate probabilistic distribution of tokens in pre-training. You are never going to get the exact book word-for-word using LLMs.
>You are never going to get the exact book word-for-word using LLM.
This is pretty much the exact claim of a NYT lawsuit against OpenAI.
"One example: Bing Chat copied all but two of the first 396 words of its 2023 article “The Secrets Hamas knew about Israel’s Military.” An exhibit showed 100 other situations in which OpenAI’s GPT was trained on and memorized articles from The Times, with word-for-word copying in red and differences in black."
Yes, LLMs fundamentally operate as a lossy compression scheme for their training data. There's been countless examples of them reproducing their training data with very high accuracy
People claim that the data isn't stored, but clearly a representation of it is encoded and reproducible. I saw chatgpt word for word plagiarise a stack overflow comment just two days ago
Well yeah, if you're making the claim that it stores a representation of the data in some form.
Does your calculator app store a representation of the answer to 1+2/2*1.1 and all other combinations of inputs or does it determine the answer from a set of rules?
It's a different case when the input contains more information than the output.
If you put "1+2/2 x 1.1" into a calculator and it spit out a verbatim copy of a New York Times article, does it necessarily contain a representation, or does it just contain some really extensive rules? I'd argue those rules necessarily are a representation of that information, given that it contains far more information than provided by the input.
Ah but can it reply snarkily and close your ticket as a duplicate that is NOT A DUPLICATE? If not it will never recreate the real stack overflow experience.
When I was in school, writing "in my own words" was never an excuse to not cite a source. It was actually something that took me a little while to understand, it's the source of the information that needs to be cited, and that's not limited to literal quotations of someone else's writing.
That's more an argument for why you can't just use LLMs as a source of truth. Conveniently, LLMs like ChatGPT do often cite their sources, especially if you prompt them to.
You can get it to reproduce content but it’s a game of cat and mouse. Were it not for the alignment to avoid direct reproduction it would taken far more often.
> RECAP consistently outperforms all other methods; as an illustration, it extracted ≈3,000 passages from the first "Harry Potter" book with Claude-3.7, compared to the 75 passages identified by the best baseline.
In Bartz v. Anthropic the judge ruled that Anthropic making a digital copy of a printed book and then discarding the physical book was not infringing when used to a train a model.
Come up with obscure topic that has few relevant results, post about to Reddit on your profile page, wait a few hours and then query Gemini/ChatGPT about that exact thing and tell me you still feel this way.
Fair use was built around human limitations. The mass scraping campaigns done by the AI giants were clearly an overreach in spirit, if not letter. Most people's intuition is that these massive operations that are valued in the trillions can't have been drawn from some untapped common resource, and they're correct. Someone, somewhere is not being properly compensated.
I have no problem with taxing AI companies so that their profit is marginal, or forcing them to provide compute for free. That seems like the correct balance of what they're harvesting from the "commons" (which is really just the totality of private IP that was exposed to their crawlers).
Fair use is the balance between creators and those that in someway use the content. Somehow it has become excuse not to compensate the creators in anyway. To me AI training part really looks something that should be treated separate and thus give the creators compensation when their works are used.
Now how much and should it be based on revenue from output is open discussion. And it might also be that there is no fair model to pay them. Which means that well too bad for LLMs...
The nature of how LLMs work makes it impossible to connect a derivative work to its source data in the training. However, the weights couldn't exist without that training data - the works of the creators were used during training - and the entity making money off the use of that training data is primarily the LLM platform owners. So they should pay.
We are trying to avoid another situation where "resource wealth" goes uncompensated, producers remain poor while processors, marketers, and merchants reap all the benefit. Unless your aim is something else, in which case you should state it.
I don’t buy this argument. The tokens are useless without their context, which provides the probability distributions needed to make them useful. Sure you MIGHT not be able to get the book word for word, but it’s impossible to make a useful model without the whole book and all of the artistry that went into it, to guide the tokens in their expected output.
Fair use generally does not cover commercial use, which this clearly is, and is dependent on the amount of the original content present in the derived work, which I would contend in this case is “all of it”
"Commercial Use" is only one part of the four prongs of the fair use test. For example, commercial Parody is generally considered Fair Use. Look at Space Balls, which is a direct transformation from Star Wars.
This is all new territory. We don't have court-settled law yet.
It's more complicated than that. Quite a bit more.
Commercial use counts _against_ a fair use defense, but is not dispositive: it's not accurate at all to say it "generally does not cover" commercial use. This is the "purpose and character" test, one of four in contemporary (United States) fair use doctrine.
Purpose and character also includes the degree to which a use is _transformative_. It's clear that the degree to which a training run mulching texts "transforms" them is very high. This counts toward a fair use finding for purpose and character.
> is dependent on the amount of the original content present in the derived work, which I would contend in this case is “all of it”
The "amount and substantiality" test. Your case for "all of it" can't possibly be sustained: the models aren't big enough. It's amount _and_ substantiality: this has come up in the publication of concordances, where a relatively large amount of a copyrighted work appears, but it's chopped up and ordered in a way which is no longer substantially the same. Courts have ruled that this kind of text is fair use, pretty consistently. It's not an LLM, of course, but those have yet to be ruled on.
Also worth knowing that courts have never accepted reading or studying a work as incorporation, and are unlikely to change course on the question. It's taken for granted that anyone is allowed to read a copyrighted work in as much detail as they wish, in the course of producing another one. Model training isn't reading either, but the question is to what degree it resembles study. I'd say, more than not.
Specifically:
> it’s impossible to make a useful model without the whole book and all of the artistry that went into it
Courts have never once accepted "it would be impossible for defendant to write his biography without reading plaintiff's" as valid, and it's been tried. The standard for plagiarism is higher than that.
"Effect upon the work's value" is probably the most interesting one. For some things, extreme, for others, negligible. I suspect this is the one courts are going to spend the most time on as all of these questions are litigated.
Ultimately, model training is highly out-of-distribution for the common law questions involving fair use. It was not anticipated by statute, to put it mildly. The best solution to that kind of dilemma is more statute, and we'll probably see that, but, I don't think you'll be happy with the result, given what I'm replying to. Just a guess on my part.
It is of course true that it is unsettled law, and that fair use is more complicated than my offhand comment suggested.
> Courts have never once accepted "it would be impossible for defendant to write his biography without reading plaintiff's" as valid, and it's been tried. The standard for plagiarism is higher than that.
This I think misses the thrust of my argument, though. Its hard to find an exact human analogy, because neither the technology nor the scale at which it operates is remotely human.
I see it less as “writing his biography without reading the plaintiff’s” and it’s more “using the same style and metaphors to make thousands of copies of very similar biographies, with certain bits tweaked,” like turning an existing work into mad lib.
I don’t know how the courts will eventually rule on it, but it certainly feels like theft to me.
It's fascinating how intuitions differ. To me, it doesn't feel like theft at all. For one thing, theft is depriving another of something, and has therefore never been a good metaphor for infringement; hackers used to be the most insistent about this principle, and it's weird to see a doctrine which was cooked up in a literal AI lab get thrown out the window for literal AI.
But pretending you said "infringement", for me it comes all the way back to the Constitution: "To promote the Progress of Science and useful Arts". I cannot possibly twist the development of large language models into something which violates the spirit of that purpose. I don't see how anyone can.
Your point about the scale is valid, and the alienness of it, sure. But you haven't made the case that the vastness of the scale should affect the conclusion.
Something I left out in the first post is that copyright is meant to protect expression, and not ideas: this is the deciding factor in the 'nature of the copyrighted work' test for fair use. More expression, more protection: more ideas, less.
I think the visual arts have a strong case that image generators directly infringe expression: I'm not convinced that authors do, and I think software should never have been protected under copyright because the ideas-to-expression ratio is all wrong for the legal structure. There's clearly no scale case to be made for ideas: "but what if it's _all_ the ideas" fails, because the ideas are not protected at all. Nor should they be, that's what patents are for, and why patents are very different from copyright.
LLMs are remarkably good at 'the facts of the matter', hallucination not withstanding. They're very poor at authorial 'voice transfer', something image generators are far too good at. It's when I start asking myself "well what even _is_ this 'expression' thing anyway?" that I conclude that we're out over our skis on the LLMs-and-IP question: precedent can't tell us enough, and that leaves legislation.
reply