HN user
I don't know the exact details, but I have heard (on C++ Weekly, I believe) that it offers some advantages when linking code compiled with different compiler versions. That said, I normally avoid it and use fmtlib to avoid the extra compile time. So it isn't clear if it is a win to me. Header-only libraries are great on small projects, but on large codebases with 1000's of files, it really hits you.
My personal experience is that the increase in ads has encouraged me to subscribe to creators I like via Patreon and view content on there. If many people are doing this, I wonder if it skews the view statistics and, therefore, lowers the number of recommendations for the best channels. In turn, this makes it less likely for good channels to be discovered. The increase in YouTube ads also makes me much less interested in browsing there, and I am finding other things to do instead.
I opted for sorting by project. We live in a world where you can get most things the next day. Then, I keep a few part-sample books for common components like capacitors and resistors for modding. Even if I need a specialist component, I know what projects I have used it in before
My take is that it means founders are sharing the risk by investing in the company and by taking a smaller salary than they could otherwise. This means there is a clear motivation to get a successful exit for them as well as the VC's
It saves writing lots of boilerplate. If you implement it for a type, you automatically get: <, >, <=, >=, ==, !=
From what I understand, it can learn and execute the algorithm fairly reliably, though it won't be 100%. When the LLM generates text, it is randomised a little, as well as some tricks that prevent repetition, which would likely cause problems with numbers containing all the same digit.
Because often, the tokens are broken up as random groups of numbers. For example, let's say 1984 appears quite a few times in the source text, this will become a single token. Given that these many different, semi-random groups of digits it is hard for the LLM to learn any consistent rules. I believe there are papers showing that if you structure numbers more consistently LLMs have no problem with this kind of arithmetic.
Too your first point, I am not sure what constitutes new or novel here. I've seen the AI do exactly the type of borrowing from different sources you describe and producing something that seems very new to me. Though I admit, often these are under human direction with some carefully chosen sentences. Is that enough to generate something really new?
To your second point, most of the work I've seen generated from these models were done for free. It could be argued that these tools add to an artists toolbox rather than take something away. I can see for example where one poor artist could create a computer game of the same quality that it takes a triple-AAA game company to do today. Is that good or bad?
From what I've seen in the art world it is common practice to build on the work of others without attribution. Is this really all that different? That said, I think these tools still implicitly depend on people to pick out good images that resonate with people, as if people don't like the image it creates it won't get shared.
From what I've experimented with so far the biggest barrier to using rust was having a mature ecosystem of libraries to use. This is changing though!
I am wondering if after changing career, would they still read this site, or is it part of what they wanted to get away from?
I find picking one calorific food and cutting it out, but otherwise eating the same works really well. For example I used to have small bottle of orange juice with lunch, which I cut out and it put my weight on a downward path. It is easier than worrying about the whole of your diet.
That could be part of the reason why 85% drop in power consumption only leads to twice the perfomance.
It is interesting to wonder what this might look like. There are still limits to what is physically possible, we only have finite resources. I also personally value having a diversity of people and views. I wonder if at least from the points of view of some of the inhabitants the world will be unfair. That said, we should always try and do better.
The upside of that maybe there is more awareness of what is going on. If there is no centralization, could we end up in a world there no consensus on what is acceptable or right?
I've read applications like little snitch are popular enough that some malware changes their behavior if it detects it. It must be a non-trivial number. I do personally out of curiosity to see what connections are being made.
Location: Surrey UK
Remote: Yes
Willing to relocate: Possibly
Technologies: Machine learning and AI (20 years), Computer Vision, Face Recognition, C++, Python, Robotics, some electronic design. Startup and CTO experience.
Résumé/CV: https://www.linkedin.com/in/charlesgalambos/
Email: craftit@gmail.com
Though if you think about the number of hours of use and testing javascript and browser technology in general has had, maybe it does make sense.
After coming across this, I wonder just how much of the brain is required to exhibit intelligence. It seems like some people can get away with a lot less that the 'normal' amount.
As I understand it, Go also give some guarantees about the maximum amount of time the garbage collection will stall the program. An issue where ever you are trying to achieve consistent performance. As I understand it most of the java benchmarks postpone clearing up memory until it can be done trivially at exit, something you just can't do in a whole class of programs.
It is a fair point, if you want to keep the playing field level and you want to compete on the basis of coding skill then you need to manage the code on the server. Otherwise simple network latency to the server could skew the playing field, assuming it is real-time.
I must admit I just want to experiment with different AI algorithms like deep learning, genetic programming, rule induction and see how they do. I have much of the code already but I am unlikely to get the time to rewrite it, even if could get it to fit within the game environment. Especially as some of the training is best done on a GPU.
It would be great to be able to remotely connect to the world, so plays can use the language of their choice. Personally I like web sockets and JSON, as they are well supported by most programming languages. In any case it would set it apart from some the other online AI arena's I've seen so far.
But the code path is first pass + optimisation gives correct code. That doesn't guarantee the first pass generates correct code. I believe there were bugs along those lines at one stage, though I admit they are rare. I've not had to track down a compiler bug in quite a few years.
Interesting they are using it without optimisations. The compilers is used far more with optimisations, so those code paths are actually far better tested. I have heard of cases where bugs only appear in unoptimised code.
I suspect setting up motivations for the AI is going to be a big research issue before too long. If you can write a simulator for the task you want to it solve, you should be able to train it. Often writing the simulator is much easier than solving a task itself. For example, the atlas robot, then can simulate it but struggle to control it.
Keeping someone entertained in a game, the longer they play the more reward you give it. I was also thinking of hooking up a face recognition system, so it can be rewarded when the user smiles.
Its not all that different, we take multiple asynchronous streams of messages integrate them into a coherent predictive model, and use that to feed the reinforcement learning. The messages can contain images, a simple case can be seen in the demo with a 1d vision sensor.
It is an amazingly powerful technique. We've been working on a service which lets you do this kind of learning with any JSON stream. You can see a demo here:
All sorts, we have a machine learning service which lets you do something similar with JSON messages.
Definitely, but at the cost of requiring more memory and processing power. If its goal is to be compact and lightweight I think it would be helpful to have something in protocol itself.
Looking at the spec though, it does allow numeric keys in maps so you could use id's and provide the definitions elsewhere.