Show more
Sundar

ChatGPT stuff 

tl;dr: it seems likely ChatGPT took advice meant for Python or MATLAB, and substituted Julia’s name in their place because it considers them “close enough”.

1 isn’t wrong, but makes no mention of type stability, which is at least as important as its (more generic, applicable many languages) suggestion.

2 is also generic advice that’s good to have, but isn’t specific (but I consider this one a success)

3 rates as kinda okayish advice. One could argue that for a beginner, the base functions offer a solid place to start, if we assume they’re gonna write badly optimized code (but then teaching them how not to do that - as the prompt asked - is a better way to solve that). But since this is Julia, it’s not uncommon that simple custom code you write beats the obvious ways using built-ins. So I’d consider this potentially misleading.

4 is likely the smoking gun here - the repeated mentions of vectorization and asking to use techniques to “vectorize your code” seems to suggest that this whole thing was taken from guides written for Python or MATLAB, which are at this point more numerous than those for Julia, and then the Language Model substituted Julia in the language name’s place because it considers them similar. (A previous answer said “there are a number of programming languages that are commonly used for numerical and scientific computing, including Python, Julia, MATLAB, R, and others” - so it knows they’re in the same category.)

Show thread
1
Sundar

ChatGPT stuff 

[ChatGPT’s full response below:]

There is no single secret to making Julia code fast, as the performance of Julia code can depend on a number of factors. However, there are a few general strategies that can help to optimize the performance of Julia code and make it run faster, such as:

  • Choosing the right data types and structures: Julia is a dynamically-typed language, which means that the type of a variable can change at runtime. This can make it difficult to optimize the performance of your code, because the compiler cannot always determine the best way to represent and manipulate your data. To make your code run faster, you should choose the appropriate data types and structures for your data, based on the operations you want to perform and the performance characteristics of the different types.
  • Avoiding unnecessary computations: Inefficient code often contains unnecessary computations, such as calculations that are performed multiple times or operations that do not affect the result of the program. To make your code run faster, you should avoid performing unnecessary computations, and try to structure your code in a way that minimizes the amount of work that the computer has to do.
  • Using built-in functions and libraries: Julia has a rich set of built-in functions and libraries that can help you perform common tasks efficiently. Rather than implementing your own algorithms or functions, you should use the built-in functions and libraries whenever possible, as they are often highly optimized and can save you a lot of time and effort.
  • Exploiting parallelism and vectorization: Julia is designed to support parallel computing and vectorization, which can help you to make your code run faster by utilizing multiple cores or processors. To make the most of these features, you should try to structure your code in a way that allows it to be executed in parallel or vectorized, and use the appropriate techniques and tools to parallelize or vectorize your code.
Show thread
1
Sundar

ChatGPT stuff 

I asked “What is the secret to make Julia code fast?” (just to see if it picks up anything from Jakob Nissen’s excellent guide viralinstruction.com/posts/opt or the many other optimization guides online).

The answer turned out to be good example of the model’s ability to be subtly misleading while not being entirely wrong.

(Thread, with ChatGPT’s reply as the next post.)

1
Sundar

Todo.txt with the Vim plugin gives me most of these except for the first two.

+Projects don’t really do the job for me with regard to subtasks, not nearly flexible enough.

Someone recommended TaskWarrior, but it looks like that doesn’t have subtasks either?

Have been thinking about extending the Todo.txt format in some way to support subtasks, not sure yet what the best way to go about it is. Sometimes I also want the subtasks to be strictly ordered (“show me only the next available subtask in order, not all of them”), sometimes they’re fine unordered. Hmmm…

Show thread
1
Sundar

Thinking out loud:

What I want from a todo system are

  • ability to break down a task at any point into subtasks (and automatically track completion)
  • dependence between tasks i.e. ability to say Task X depends on Task N which depends on Tasks C and D
  • easy keyboard manipulation, no mouse dependence
  • plaintext or near-plaintext storage
  • priorities
  • contexts
  • being FOSS and local
  • (nice to haves:) tags, recurring tasks

1
Sundar boosted
Void Abyss

What I like the most about the is the randomness in the
social interactions among the people from different instances and the diverse niche content you find.

1
Sundar boosted
🎓 Dr. Freemo :jpf: 🇳🇱

So I'm watching the 1976 "the boy in the plastic bubble" and it got me thinking.. there is a scene where he says hje doesnt take showers and doesnt smell because there are no germs...

So if people who are in sterile tooms all their life are truely germ free... that must mean all sorts of things... no BO, you never smell, do you even need to brush your teeth, you wouldnt get plaque or anything if there are no germs to consume the food on your teeth... Does it also mean they have no gut bacteria..

and aside from all those dise effects, which im curious about... how do you even steralize a human in the first place, even if they are born that way and go in from birth, surely your born with bacteria on you.. how do you steralize a human and if you cant how do they not die from this...

1+
Sundar boosted
Scott McIndoe

Another term of allowing students to ask me questions anonymously during class. >800 Q in 66 Chem101 lectures, and always a different flavour to the ones I get via traditional hand-raising (the ones below aren't at all representative, just some fun examples)

1
Sundar boosted
Raghav Agrawal

ChatGPT on What are some non-european contemporary philosophers and social theorists who write about imagination

ChatGPT on the same question without "non-european"

1+
Sundar boosted
Shark Blackle

jesus christ wolfram alpha, I'm sorry I asked

1+
Sundar boosted
Ambulocetus

They have a Roomba that can clean stairs now, which sucks on so many levels.

1
Sundar boosted
nomi

Checklists dwindling
To do list is running dry
Is this happiness?

0
Sundar boosted
Miguel Guhlin 🦉 🇵🇦 🇺🇲

Revised document attached. Removed Metatext and replaced it with Mastoot per someone’s suggestion. :-)

Show thread
1+
Sundar boosted
Ada Palmer

Me in class: “Looks like no one managed to finish the reading, that’s on me, clearly I assigned too much today. Let’s discuss [X] instead.”

Students: 😮🤯🥹

It’s really powerful for students when the role model at the front of the classroom takes responsibility & shows imperfection. The more I #Teach the more I realize showing weakness teaches humaneness & combats impostor syndrome, & is one of the things my students most often say they loved when a course is over. #teaching #academidon #histadon

1+
Sundar boosted
Kushal Das :python: :tor: 🇵🇸

Is there any #OpenSource #desktop application to manage equipments/parts catalog?

I want to find any equipment/parts fast enough. Please boost so that I can get more suggestions.

1
Sundar boosted
Colin Purrington

Close-up of a wax plant (Hoya carnosa) that shows the fuzzy corolla and fused male and female reproductive parts (gynostegium). Flowers emit a cinnamon/chocolate scent, especially at night, and are pollinated by moths. #botany #Apocynaceae #flowers #pollination

0
Sundar boosted
TILvids

I've often wondered why open-source orgs like @mozilla @nextcloud #Canonical and others don't take a more active role in the fediverse. All of those orgs should not only be actively running #Mastodon accounts, but should also be posting content to #PeerTube and honestly even running their own instances of both. In the case of Mozilla, it would make a ton of sense for them to host both Mastodon and PeerTube instances as an open alternative to Google.

Thoughts on that?

1+
Sundar boosted
Sebastian Marsching

I think that there are some misconceptions about how hashtags, and more specifically searching for or following hashtags, works on Mastodon.

Unlike Twitter (and other centralized services), you cannot “follow” or search for a hash tag and expect to see all posts using this hashtag across all instances. If you want this, you will have to use a “group account” instead of or in addition to the hashtag.

0
Sundar boosted
Eric Leamen

What excites me most about this place right now: I’m tempted to actually start posting and sharing my thoughts again. Engaging in the discourse, as they say. The other socials became intimidating places to actually converse in the last few years. Pleased that our nerdy little community is here, and growing.

Maybe we can actually do this thing.

1
Sundar

It’s concerning how much of my self-talk involves telling myself:
“We’ll catch up on sleep later, I promise, but let’s pretend for now that we’re well-rested and alert and not a grumpy sleep deprived husk at all.”

0
Show more

Sundar's choices:

julialang

13

india

4

rationality

2

diversity

2
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.

Trending now

Resources

Developers

What is Mastodon?

qoto.org

More…