KingRandomGuy
@KingRandomGuy@lemmy.world
- Comment on Nobel Prize 2024 5 weeks ago:
I work in an ML-adjacent field (CV) and I thought I’d add that AI and ML aren’t quite the same thing. You can have non-learning based methods that fall under the field of AI - for instance, tree search methods can be pretty effective algorithms to define an agent for relatively simple games like checkers, and they don’t require any learning whatsoever.
Normally, we say Deep Learning (the subfield of ML that relates to deep neural networks, including LLMs) is a subset of Machine Learning, which in turn is a subset of AI.
Like others have mentioned, AI is just a poorly defined term unfortunately, largely because intelligence isn’t a well defined term either. In my undergrad we defined an AI system as a programmed system that has the capacity to do tasks that are considered to require intelligence. Obviously, this definition gets flaky since not everyone agrees on what tasks would be considered to require intelligence. This also has the problem where when the field solves a problem, people (including those in the field) tend to think “well, if we could solve it, surely it couldn’t have really required intelligence” and then move the goal posts. We’ve seen that already with games like Chess and Go, as well as CV tasks like image recognition and object detection at super-human accuracy.
- Comment on So glad I'm ditching these fucking idiots 9 months ago:
One of the big changes in my opinion is the addition of a “Smart Dimension” tool where the system interprets and previews the constraint that you want to apply instead of requiring you to pick the specific constraint ahead of time(almost identical to SOLIDWORKS), and the ability to add constraints such as length while drawing out shapes (like Autodesk Inventor, probably also Fusion but I haven’t used that). It makes the sketcher workflow more like other CAD programs and requires a little less manual work with constraints.
- Comment on So glad I'm ditching these fucking idiots 9 months ago:
Hmm, that actually sounds like a bug. I haven’t seen that on my end (once I set the locations of options in my toolbars they stay there, even after restarts). You may want to report that on their issue tracker.
- Comment on So glad I'm ditching these fucking idiots 9 months ago:
Last I tried it, there was no fix. Their latest update on the website says:
The work on the toponaming problem is an ongoing project, and we are very grateful to the FreeCAD community for contributing a lot to that effort. But it’s not complete yet, there will be much more to say when it’s largely done. So let’s focus on the other three.
So I take it they haven’t implemented a fix. They previously said they were going to work with the FreeCAD team on mainlining a toponaming fix, using realthunder’s work as a proof of concept, but said fix has not landed in mainline FreeCAD yet. I believe that’s the major feature they’re looking to implement for FreeCAD 1.0.
Definitely excited for Ondsel though! Hopefully that fix can be integrated quickly.
- Comment on So glad I'm ditching these fucking idiots 9 months ago:
IMO the bigger problem with FreeCAD is the topological naming problem. It’s very easy to get frustrated because your model broke due to a change you made in an earlier feature.
The UI isn’t amazing though, and that unfortunately happens quite a bit with open source software. Hopefully it’ll go the way of Blender and KiCAD with an eventual major release that overhauls the UI.
- Comment on So glad I'm ditching these fucking idiots 9 months ago:
Ondel has a nicer user interface, but I personally use and recommend realthunder’s LinkStable branch of FreeCAD. Mainline FreeCAD (and by extension, Ondsel) suffer from the topological naming problem, which can be especially jarring to users coming from proprietary CAD software. realthunder put a lot of work into a solution that handles the problem pretty well, so I’m using his fork until toponaming gets mainlined.
- Comment on A crowd destroyed a driverless Waymo car in San Francisco 9 months ago:
Yep, and for good reason honestly. I work in CV and while I don’t work on autonomous vehicles, many of the folks I know have previously worked at companies or research institutes on these kinds of problems and all of them agree that in a scenario like this, you should treat the state of the vehicle as compromised and go into an error/shutdown mode.
Nobody wants to give their vehicle an override that can potentially harm the safety of those inside it or around it, and practically speaking there aren’t many options that guarantee safety other than this.
- Comment on New (Open Box) Ender 3 Blue Screen 10 months ago:
Not sure what’s going on but if the printer was flashed with Klipper, it won’t show anything on the screen until it is connected to a computer running klipper’s host software.
- Comment on Why do we still use stepper motors? 10 months ago:
That sounds about right. You can technically achieve full torque at 0 RPM, but the current required for that might be very high (and may not be practical for the small size, power limits, and cost of a 3D printer). I know this is a probably in Asian import mini-lathes - if you run the spindle too slow you won’t get much torque out because the driver can’t supply enough current.
- Comment on Why do we still use stepper motors? 10 months ago:
I think it would be a better idea to use linear encoders if you were going to add encoders to the shaft, as that way you could directly readout the position of the tool along an axis without issues like backlash that would mess up your calculated position. This is what I’ve seen on (both manual and CNC) mills and lathes.
- Comment on Why do we still use stepper motors? 10 months ago:
BLDC and AC servos maintain full torque at stop too, and have about 2-3× the torque of a stepper of similar size.
Huh, this is true about BLDCs as well? I remember seeing in a video that BLDCs tend to have very poor torque output when stopped and especially when at low speeds, whereas AC motors have a pretty much flat torque curve until they get fairly fast. I’d be interested to know if this is true.
- Comment on Europe reaches a deal on the world's first comprehensive AI rules 11 months ago:
I’m a researcher in ML and that’s not the definition that I’ve heard. Normally the way I’ve seen AI defined is any computational method with the completing tasks that are thought to require intelligence.
This definition admittedly sucks. It’s very vague, and it comes with the problem that the bar for requiring intelligence shifts every time the field solves something new. We sort of go “well, given these relatively simple methods could solve it, I guess it couldn’t have really required intelligence.”
The definition you listed is generally more in line with AGI, which is what people likely think of when they hear the term AI.
- Comment on Jailbroken AI Chatbots Can Jailbreak Other Chatbots 11 months ago:
I believe this is the referenced article:
- Comment on Google Researchers’ Attack Prompts ChatGPT to Reveal Its Training Data 11 months ago:
I think what they mean is that ML models generally don’t directly store their training data, but that they instead use it to form a compressed latent space. Some elements of the training data may be perfectly recoverable from the latent space, but most won’t be. It’s not very surprising as a result that you can get it to reproduce copyrighted material word for word.
- Comment on Google Researchers’ Attack Prompts ChatGPT to Reveal Its Training Data 11 months ago:
Not sure what other people were claiming, but normally the point being made is that it’s not possible for a network to memorize a significant portion of its training data. It can definitely memorize significant portions of individual copyrighted works (like shown here), but the whole dataset is far too large compared to the model’s weights to be memorized.
- Comment on Fusion360 free Startup option is going away! 1 year ago:
I’m not sure about chamfer (I haven’t used it in my sketches) but fillet is definitely included in the Sketcher workbench.
- Comment on Fusion360 free Startup option is going away! 1 year ago:
Entirely fair! I think FreeCAD is still fine for hobbyists like myself though. It does take quite a bit of getting used to (I came from Fusion360 and Inventor first) since it operates somewhat differently, but it’s good that we have at least one option.
Hopefully it’ll see more development and become substantially more viable in the future.
- Comment on Fusion360 free Startup option is going away! 1 year ago:
This isn’t true? Fillets and chamfers are available in the PartDesign workbench.
- Comment on Framework 13 With AMD Ryzen 7040 Series Makes For A Great Linux Laptop 1 year ago:
The big thing you get with frameworks is super simple repairability. This means service manuals, parts availability, easy access to components like the battery, RAM, ssd, etc. Customizable ports are also a nice feature. You can even upgrade the motherboard later down the line instead of buying a whole new laptop.
- Comment on ‘ChatGPT detector’ catches AI-generated papers with unprecedented accuracy 1 year ago:
I haven’t read the article myself, but it’s worth noting that in CS as a whole and especially ML/CV/NLP, selective conferences are generally seen as the gold standard for publications compared to journals. The top conferences include NeurIPS, ICLR, ICML, CVPR for CV and EMNLP for NLP.
It looks like the journal in question is a physical sciences journal as well, though I haven’t looked much into it.
- Comment on ‘ChatGPT detector’ catches AI-generated papers with unprecedented accuracy 1 year ago:
I’m curious what field you’re in. I’m in computer vision and ML and most conferences have clauses saying not to use ChatGPT or other LLM tools. However, most of the folks I work with see no issue with using LLMs to assist in sentence structure, wording, etc, but they generally don’t approve of using LLMs to write accuracy critical sections (such as background, or results) outside of things like rewording.
I suspect part of the reason conferences are hesitant to allow LLM usage has to do with copyright, since that’s still somewhat of a gray area in the US AFAIK.
- Comment on ‘ChatGPT detector’ catches AI-generated papers with unprecedented accuracy 1 year ago:
Also one very important aspect of this is that it must be possible to backpropagate the discriminator. If you just have access to inference on a detector of some kind but not the model weights and architecture itself, you won’t be able to perform backpropagation and therefore can’t generate gradients to update your generator’s weights.
That said, yes, GANs have somewhat fallen out of favor due to their relatively poor sample diversity compared to diffusion models.
- Comment on The Future of Machine Learning: A New Breakthrough Technique (MLC) 1 year ago:
For reference, ICML is one of the most prestigious machine learning conferences alongside ICLR and NeurIPS.