eh, I don't know. Cursive is pretty rare nowadays. You also in my opinion do not really need the capital letters, and many of them look like their lowercase counterparts, but I take your point. I definitely think English can be tricky to learn, though, and feel lucky to speak it natively given that it is the lingua franca in many (perhaps most) contexts.
My guess is a lot of it. The 25% number is meaningless because they're not comparing to the proper baseline. Anecdotally, I was in Berlin last year, and there was a similar product, and locals said not that many people were taking advantage of it.
I think the use of "sir" depends on the person's background culture. If you're from the US, it's probably too formal unless you're in the military or something, but other cultures place a higher value on respect for authority. I've had people not originally from the US call me "sir", and while I make sure they know they don't have to, I also don't worry about it.
I certainly don't think you have to, and I don't think that's what Graham is saying, either. For those who are ambitious/do seek "greatness" in some form, though, I think this is a good article.
> Primarily, what I’ve learned is that Wayland is actually really well-designed. The writing is on the wall for X, and Wayland really is the future.
Hmm. The first sentence does not imply the second is true. I wonder what it would take to see widespread adoption of a new windowing system. Quite a lot, I would imagine.
Thank you for this detailed comment, which does a much better job in my opinion of critiquing the substance of the article than many others. If I may:
> I think most of the arguments the author makes against worrying about superintelligence are pretty weak, though.
Which ones did you find to have more merit?
Personally, the premise of recursive self-improvement seems most suspect to me. It is somewhat related to how the author points out that we can't define and measure intelligence precisely. Even if we can't do that, though, it's still plausible to me that recursive self-improvement is possible; I think the fundamental question is about the nature of intelligence. Regardless of whether it can be precisely defined, either by us or by an entity smarter than we, the question is: Can it be improved ad infinitum with no serious side effects? I don't know that we have the evidence to answer this question (though I am very open to learning about it).
I agree that it seems kind of invalid but only kind of. Any site that has an editorial policy and/or a ranking algorithm for what gets more views will have some forms of bias. I agree that in the case of reddit, it makes sense to evaluate each subreddit on its own. Also, if a page where users make submissions that mostly lean in one direction or another, that could be useful information to know.
What about this makes it a pain point? Is it the amount of time that it takes to deal with cheating, the inability to effectively punish the behavior, the impact it has on the professors' (and perhaps some students') morale, or something else?
Hmm; I would think a better definition would include a description of the behavior as being greater than the sum of its parts. I can imagine abilities and behaviors being in the training data but not exhibited by a smaller LLM but being present in a bigger one. My favorite example is the ability to multiply large number consistently and accurately (e.g. 23540932 x 3274, which ChatGPT-4 just failed to do correctly for me when I asked it). A better LLM would learn how to carry the process out step-by-step.
This is an interesting suggestion to me. Honestly, I consider a lot of other things like type safety and ergonomics of the language before I think about compile speed if at all.
However, that's not to say it's irrelevant. Perhaps I should be thinking about it more. I did notice that they didn't suggest any particular language with this comment. Compile speed seems like a difficult thing to compare between languages in general, since it would be difficult to make a genuine apples to apples comparison.
Is it just me, or does the PR experience with Github leave a lot to be desired? Unlike some, I don't have a problem with the concept of a PR per se, but I frequently find navigating the various conversations in my PRs more difficult than I feel it should be. If a branch is rebased, then comments on code that was changed get lost, and one has to go the Conversations tab and search for the comment there. I understand that the commit itself is gone, which is fine, but I think an easy way to navigate through the comments, even if their context is lost, would be a great improvement.
I am actually more interested in the suggestion to try VCS alternatives, though. To me, it seems like we have reached a local maximum with git where in principle, better approaches exist, but no one seems to think the cost of switching would be worth it.
> Based on my understanding of the approach behind ChatGPT, it is probably very close to a local maximum in terms of intelligence
To me that is an open question. OpenAI has not revealed a lot of the relevant stats behind GPT-4. I also haven't seen anything about their future LLM research (though I haven't looked very hard), but I think the above assertion remains to be seen.
I have tried GPT-3.5 and 4. There is a marked difference. (I have used it do what I think of as simple but nontrivial programming tasks, asked it for recommendations for various products, etc.) GPT-4 still fails for me regularly but not nearly as often as GPT-3.5. I find it useful enough in my daily life to pay $20/month for. So if you did want to try it, you might be surprised.
So when you are speeding, you are endangering people and property on the road. The amount of danger is proportional to the time you spend speeding among other things.
With the falsification of business records, the situation does not seem as clear-cut. It is wrong because it is preventing the public or other relevant parties from having access to accurate information. I don't know if one can really make a formula to calculate how much damage is done by a given falsification, but I personally do not think I would include the number of checks or (probably) the number of invoices written in that formula. The method of payment seems mostly incidental and not essential to the wrongness of the crime. (I include the number of payments as part of the method of payment.)
If this does not clear up my position, feel free to let me know. There is more I could write to flesh it out, but I don't know if that's necessary or if it'd be overkill =)
Yes, it is normal. I was talking about the method of counting. I suppose one has to pick something, and the number of checks written is as good as any other metric. I do think that it is a little disanalogous to speeding because it is the same thing being concealed each time. Also, to be clear, I was not making a definitive normative judgment. I was more pointing out the interesting nature of how what seems to me to be an ongoing act of deception gets charged, and how it conflicted somewhat with my intuitions.
Interesting that they chose to go for 34 separate counts of the same thing, falsifying business records in the first degree. It seems to me that the punishment for a crime should be proportional to the amount of damage done, and how many checks one wrote seems to me like it could be less important than, say, the total dollar amount, or the duration of time that the falsification(s) took place, or the number of things that were concealed. I am curious if anyone knows the different approaches that prosecutors have taken in various states over the years, what the statutory and case law says, etc. I imagine a lot of the judgment calls in these situations are left to the judge and/or jury.
Are you saying that you don't suffer (or at least consciously feel the effects of) cognitive fatigue? If I am interested in something, I like to focus on it for quite a while, but after enough time passes, I just can't any more. I do think that there are some people who can naturally last much longer doing cognitive work than is the norm, though, much like some athletes are natural ultramarathoners. So perhaps you are one of those people.
Suppose I write a unique and deeply technical textbook on a subject. Someone could buy that book, read it, become an expert on that subject, and then profit from that expertise. This is all well and good. One of the things that makes ChatGPT different is its ability to scale such that it can provide that expertise to a much larger number of people and at a much lower cost.
> libraries, frameworks, and paradigms evolve so rapidly in the web
Speaking of that issue, wasn't GPT-3 trained with data that doesn't go past 2021? Won't this be an issue for any AI assistant using GPT-3 to write code?
One that I have been thinking about lately is the `QEnumComboBox` [1]. It takes the concept of a combobox (i.e. drop down list/`select`) and combines it with a python `Enum` to produce a very commonly desired functionality: allowing the user to select a value and then converting it to an `Enum`. This to me seems easily reusable because it encapsulates that functionality in a clean way. I can't think of a corresponding example in a JavaScript framework, but I would love to see one. I think this might be more difficult because one can't really pass types at runtime.
I think instead of "truly reusable" I should have said something like "requires minimal configuration." I think Python's reflection capabilities allow one to supply less boilerplate when writing PyQt apps. This is just a subjective impression that I have, and unfortunately I am having trouble elaborating on this in a way that I think would be effective. That is why I asked the author what they thought, since I am wondering if anyone else has had this impression. However, I will think about this more, and I might provide more examples if I can think of them.
Cool! Do you have any Web programming experience? If so, how would you compare/contrast the experience coding in Qt vs. JavaScript? I have ~1-2 years of experience programming in each, and I tend to prefer Qt because to me, it seems easier to make components that are truly reusable. On the other hand, there are numerous advantages to Web programming, such as the fact that it's easier to deliver content and it's ubiquitous.
We are the first and only company to commercialize Molecular Rotational Resonance Spectroscopy. We recently raised a Series C round and are expanding our team. We are seeking an experienced Software Product Manager who will be responsible for managing the product definition for Brightspec’s software offering, with a focus on translating user requirements into actionable work items for the development team. This is a great opportunity to be part of something that expands the boundaries of what is possible in analytical chemistry. Please feel free to inquire more about the position by emailing me at first_name dot last_name at brightspec.com. You can also submit your cover letter and resume to careers@brightspec.com.
Agreed. It's a nice, succinct way of tying the natural language to the components of the formula. I don't know how well it would work in longer form content, but as a one-off or perhaps occasional technique, I think it works really well.
I'm working on upgrading Pyalgoviz [1] to use Python 3 and run using Pyodide so it doesn't depend on the server. It's been kind of slow going since I'm unfamiliar with some of the technologies involved, but I am really looking forward to releasing it and seeing what people think. Repo (still very much a WIP): https://github.com/stephen-h-d/pyalgoviz