3
points
parentheses
Joined 1,038 karma
- This is different because now the cats out of the bag: AI is big money!
I don't expect AGI or Super intelligence to take that long but I do think it'll happen in private labs now. There's an AI business model (pay per token) that folks can use also.
- I prefer using the same feature to have an extremely warm (almost red) tone. I think it's much more pleasing than b/w and results in less blue light for me.
- Not to mention the saturation of training data
- If it's < 100M, with vectors of 1024 size, you could fit all of that in ~100G of memory. So, maybe storing it in memory is an easy way to go about it. This ignores a lot of "database problems". If the docs are changing constantly, or uou have other scalability concerns, you may be better off using a "proper" vector db. There have been HN postings which indicate vector db choice matters. Do your research there.
- I think this article is funny. Python's STL is way more useful and contains myriad useful things that Ruby lacks out of the box.
difflib is probably my favorite one to cite.
Go see for yourself: https://docs.python.org/3/library/index.html
The benefit there is that their quality, security, completeness and documentation are all great!
- This looks really cool! Next diagram is getting this treatment!
- Comments like these are unwarranted.
While the parent comment indicates that a child is possibly overstepping, your comment is a greater overstep.
- This is pretty cool!
I realize that you (the 13 year old) probably did not walk this path alone and I'm guessing you had a lot of help. Not sure how much help, but I don't mind giving you the benefit of the doubt today.
Great to see that you're being creative and (at least) participating in making things!
I know that it took a lot of work for my parents to encourage my passions and for me to actually pursue them in earnest. I could have easily gotten lost in being a consumer of media and programs. Instead, I make them and it is very fulfilling. I now have a lot to thank them for and I don't think I'd be here without all their hard work and hoisting me up in my youth. It sounds like you have parents that care that much and for that, you should be grateful and participate in your growth as they do.
- I think a large issue at play here is post training. Pre training models the original distribution of input data. RL techniques tweak the models to "behave". This step changes how the models "think" in a fundamental way .
- Yes. Maybe a better way to put it would be, "all models guess every time because they are stochastic in nature. However, we only want the answers with high confidence."
- Typing not being a natively ruby thing makes all efforts to type the language in place seem second class. Ruby needs a typed variant that compiles down to type less Ruby.
- Clearly the question is meant to spur speculation. If your response is going to be "online opinions are not going to give you good predictions", you're better off not responding. It's a non answer and counter productive.
- They probably do lots of tricks like using quantized or distilled models during times of high load. They also have a sizeable number of free users, who will be the first to get rate limited.
- I think the answer is obvious so maybe you're looking for something different.
More perf means more attempts in parallel with some sort of arbiter model deciding what to pick. This can happen at the token, prompt, or agent level or all of them.
- Note to Zed: I prefer paid products to enshittened ones.
Please please please, get paid rather than holding on too tightly to making things free forcing future enshittening.
- That's kind of the point. These methods are never meant to be called directly. They're used to desugar.
I think it's fairly short sighted to criticize these. FWIW, I also did that the first time I wrote Python. Other languages that do similar things provide a useful transparency.
- Because, like engineers, their work requires intelligence and would benefit from highly adaptable software.
Finance and engineering both have a degree of verifiably. Building evals around finance is easier than, e.g., marketing work.
- The idea here (I think) is that it (1) can be a program with code used to generate the XML and (2) is typed.
Just by virtue of being a go program it enables even more sophisticated validation and automation if you want to implement it.
- Humanity's abilities are always enhanced by their tools. This simply changes the judgement of art in the face of easier execution.
Let's say I used a custom power saw to carve a statue faster than ever before and more precisely. Would that reduce my influence and my application of taste? No. I would in fact be able to produce a piece faster and have more room for making more attempts.
Neural network based art tools are all giving us the same thing - easier execution. This means greater production and the ability to try most possibilities. The fact that creating art is more accessible to the public means that more creatives can be in the arena, making for more competition.
Any creator grapples with this change over time. Woodworkers of old prefer their techniques to modern power tools, painters prefer physical media, carvers prefer real blocks of marble/whatever. All of these things have modern digital equivalents, but the establishment of existing artists refuse to leave their posts. They hold their ground that the medium is critical to the art.
Art moves and changes slowly because of this human bias against new solutions. Go to any museum of modern art and you'll find that most of it could have been executed as such 20+ years ago. It's just that art takes time to accept a new way of doing something.
- The fundamental thing that try/catch and similar language structures give us is an error _boundary_. I feel that not having one is a weakness of the language. The nesting and syntax all serve the purpose of handling errors neatly. I think it works very well and Go has unsolved the problem.
I am not saying that the mechanism is perfect but it is more useful if we have it than not. IMO it's only weakness is that you never know if a new exception type is thrown by a nested function. This is a weakness for which we really don't have a solid solution - Java tried this with checked exceptions.
Go not using such a paradigm to me is bonkers. Practically every language has such a construct and how we use it best is pretty much convention these days.
- OOC, what is the benefit of having a "library" that requires such manual labor to maintain and upgrade?
You'd miss out on CVEs because you don't use the common dependency paradigm.
You'd also miss out on bug fixes if you are not detecting the bug itself.
Help me understand because I'm with you on less dependencies but this does feel a bit extreme.
- I think one-shotting a full feature is a losing battle today. You have to figure out the best step size for the situation and prompt using components of that size.
My prompts usually resemble actions I could tell a college student — they just have a better understanding of concepts and professional lingo.
The benefit of this approach is that you know the code fairly well. You are staying with the LLM in developing a deeper understanding of the code you'll ultimately create a PR for. Then when there's an incident, you have enough deep knowledge of the code that you can be tactical.
I have found that until I trust AI to develop the code unsupervised, I have to have an equally good mental model of everything AI makes.
- https://www.focusmate.com/
I used this during the early pandemic and found it very useful
- Love this zine! Great work. Glad I saw it.
- I recall using your product years ago and it was fantastic. It helped me work solo on a web app with ease. Not having CSS and JS skills like a lot of front end devs, it was nice knowing all viewports were handled.
- It's just a matter of time. Your statement assumes AI won't help to develop robotics.
Robotics is the big unlock of AI since the world is continuous and messy; not discrete. Training a massively complex equation to handle this is actually a really good approach.
- I feel that raw code is simpler to work with, more flexible and more ubiquitous (at least for bash).
- My criticism is aimed at the glacial and mostly mishandled infrastructure projects. That is one of the big reasons for zoning changes taking so long.
I hope the US gets its act together and learns from exemplar infrastructure projects around the world.
I really dislike that this is always the argument that's being attacked. It's not even what most people are thinking when they respond.
It's clear that the exchange is privacy for effort. If I want to self host, I need to pay time and money to get it all working, then continue to maintain it forever.