Jump to content

CM0

Members
 View Profile  See their activity

  • Posts

    365

  • Joined

  • Last visited

 Content Type 

Everything posted by CM0

  1. CM0
    I don't know . But seriously, larger LLMs and "reasoning" llms do perform better at some tasks. But all of the machines are essentially probability machines. That doesn't mean they are useless, but it does place limits on their capabilities. They require immense training data for any capability you want to produce. The AI Labs are selling us mostly hype when they talk about "reasoning" and AGI. They have to keep the money flowing for these expensive machines that still are not profitable. They can become really good at some tasks, but the tendency for them to make strange and unexplainable errors, hallucinations, is not going away. Hallucinations are simply the result of a machine that doesn't have true understanding and is only constructing its output from probabilities of text patterns.
  2. CM0
    ChatGPT doesn't know what it is talking about. It is just making stuff up. Nonetheless, it is partially true. In order to be very good, it will need many human examples. However, considering the base language is javascript and there are plenty of examples of coding in javascript. If you feed the model the API documentation it will probably do somewhat ok. But you can't ask the model what it can do. Models do not know. FYI, I write extensively on this topic. For anyone interested, BTW all illustrations done with Affinity - Explanation of the lack of reasoning in LLMs
  3. CM0
    Affinity v2.6.3, Windows 10 When using paint mixer or smudge, if you turn on size jitter you will get square artifacts in your brush stroke. The following is with jitter about 50% and mixer strength 60% using Basic round brush. It seems to occur with all brushes, but the basic round makes it more noticeable. I also discovered it often occurs if using sub brushes, you get these odd square artifacts as well. Brush settings:
  4. CM0
    It is in the picture. It is rendering outside of the ArtBoard. It should be clipped to the ArtBoard layer.
  5. CM0
    When you have an effect like motion blur combined with curves adjustment, contents outside the bounding area of the artboard are rendered on screen. Affinity 2.6.3, Windows 10. draw outside artboard.afpub
  6. CM0
    Any updates on this issue?
  7. CM0
    I have addressed this substantially in the past. I have logged a substantial number. Each one often becomes a battle with the mods or others on the forum to even understand the issue, which I'm certain would be immediately understand by a dev. It is time consuming and exhaustive. Furthermore, Affinity has no priority system. They have no idea what is the most important items to fix. You cannot properly extract this information from a public forum. You need a proper customer facing tracking system. FWIW, development processes was part of my day job, so yes I'm opinionated here. If I have a hundred bugs, Affinity has no idea which ones are the most important to me. Since the likelihood of any ever being fixed is already nearly zero, if I submit them all then then chances of the most important ones being fixed first is even less likely. Therefore, I started only just mentioning the absolutely most painful. Start fixing the ones I've already submitted is the best way to fix this. Then you have ensured it is worth my time to submit the rest.
  8. CM0
    I have a log of so many bugs I've never even reported because I lost faith in them ever being fixed.
  9. CM0
    Any updates on this?
  10. CM0
    So, implying that Illustrator seemed to handle it ok? BTW Inkscape uses Python for scripting. Not sure if you have tried that.
  11. CM0
    I want a high performant scripting language that uses JIT. Affinity has always prioritized performance. I've never considered the Affinity app as intended to be scientific level of accuracy or any other application in its class. What is the end user use case? Does Illustrator support the use case(it uses JS)?
  12. CM0
    It is the Python Decimal module. However, it is still done in software. So it's performance is slow in comparison. It is not useful in general for graphic applications. This kind of accuracy is for mostly scientific purposes.
  13. CM0
    Spline runs in the browser - https://spline.design/#features Graphite runs in the browser - https://graphite.rs/ Modyfi runs in the browser - https://www.modyfi.com/ Don't worry, this is not an issue.
  14. CM0
    I just checked and this still is not fixed in RC1. It is very disappointing that af-399 was marked fixed, but in reality it was only a partial fix. All that needs to happen is the same fix applied for erase mode. I fear it will never be fixed if not addressed while we had attention on this issue.
  15. CM0
    I couldn't agree more. There are so many bugs that don't even have work arounds. Fixing bugs is like getting new features. You can do things you couldn't do before.
  16. CM0
    I have made this same comment so many times. I also emphasized this on every user survey. Fixing all the bugs is like getting new features because some things are so broken they can't even be used. It is so frustrating. One of the long standing bugs that was important to me was finally marked fixed this release. I was was so excited, but then I tested it and it wasn't fixed. Reported it as not fixed, so now there is a new bug logged that probably won't get fixed. I would really love a major dedicated bugfix only release.
  17. CM0
    Any updates on this one? Really hoping it can get included as the fix af-399 was targeted for this, but missed this use case. Hopefully the same fix is straightforward to apply.
  18. CM0
    Sure, attached is the same example switched to using erase mode as a mask. passthrough mode blending using erase.afpub
  19. CM0
    This was marked fixed; however, it is not fixed for erase mode. When using erase mode as a mask, the passthrough problem still occurs.
  20. CM0
    Please note this is not fixed for erase mode that I documented above. If you use erase mode as a mask, this problem still occurs in passthrough mode.
  21. CM0
    Yes, this would probably yield the best result. As seen from the competitor products, ML isn't pure magic. Nobody has a solution that works perfectly. Having a combination of effective tools is still required.
  22. CM0
    The most data gathered in another sense is vision. Models trained on video data have also demonstrated they are unable to make sense of that data as well. We already have machines that can experience their environment with LLM models at the core, but they still are unable to reason. True humanlike reasoning is going to require new innovations. Scaling LLMs does not seem to be the path to get there.
  23. CM0
    I just linked to a test that any human would pass, but all AI's fail. Additionally, humans outperform LLMs in reasoning with a fraction of the data that LLMs require. Current LLMs have already nearly consumed all text data that exists.
  24. CM0
    Current AI systems are mostly representations of the data on which they are trained. Once you move outside the distribution curve it gets everything terribly wrong. Complex relationships of data and its ability to create permutations out of that creates a convincing intelligence mimic. But it actually understands none of it. Very simple tests can expose the failures, such as LLM's don't know what they know and don't know.
  25. CM0
    Awesome, glad to see this is finally addressed. Will lower the temperature on my laptop. Note, as pointed out in this thread, the ruler also causes this problem. Hopefully that is addressed as well?
×
×
  • Create New...

Important Information

Terms of Use | Privacy Policy | Guidelines | We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.