Comments by "miraculixx" (@miraculixxs) on "IBM Technology" channel.

  1. Well.explained. Loved the benefits, very portable :)
    8
  2. Most importantly these are engineering challenges & tasks, not legal issues as all the laywers would want you to believe.
    4
  3. Lot's of hyperbole, little substance 😢
    3
  4. ​ @vintastic_  human expert review for every response
    2
  5. In a nutshell, LLMs are not fit for purpose as fully automated systems. Scary stuff.
    2
  6. XSS: if some guy can insert links on your web page you have a lot of problems worse than XSS
    2
  7. ... yet
    2
  8. This
    1
  9. Yeah people calling experts to clarify what the AI says. Not gonna happen. People believe everything AIs produce.
    1
  10. Wont you just get back to the input prob distributions?
    1
  11. "Repository" is a very recent word. Nobody used that prior to ~2015, certainly not in connection with data analytics.
    1
  12. Trends #10 - #20 - back to the roots. AI is so much more than the generative variant. Models like regression, logreg, svm, random forests/boosted trees, and also time series etc have a proven value proposition, we know how they work and they are easy to understand. In fact the EU AI Act implicitely mandates these simpler models for many applications due to requirements of transparency and the right to a human judgement.
    1
  13. The distinction between software, text and other IP licenses is not juristiction. Juristiction is about legal authority, that is some geographical or organisational scope. The distinction between licenses are about the terms used in the licence text and wording. Apache makes explicit references to software, whereas CC BY does not.
    1
  14. The difference between mainframes/general computing and Gen AI/LLMs is that the former actually worked as in generating marginal value. LLMs, not so much. Moving fast, as in new versions, is not a hallmark of useful. Sure AI is here to stay but all the grifters are gonna be gone. That's a good thing. Stop the hype already ;) so we can get back to building useful stuff.
    1
  15. "You could almost use it to generate these RFPs" - the almost is doing a lot of lifting there 😂 also wait until clients generate responses, and vendors use LLMs to summarize from that. fun insues 😂😂😂
    1
  16. Source for those 600bn$?
    1
  17. "how to spend energy to answer questions you already know the answer to" 😅
    1
  18. Which technology are you referring to?
    1
  19. SAM is not strictly speaking GenAI. It's an image segmentation mode. Really wished people would put in more effort to distinguish terms.
    1
  20. Of course technology is a driver. Think Uber. Without technology this would never have happened.
    1
  21. The key problem about gen ai et al is not depth but scale. These models are nice tools for individual productivity, but they are terrible for automating processes at scale. Very much like Office tools like Excel - great tool for IC, worst tool ever to for scaling automation. That's why majority of Gen AI projects will be of no value, because the models don't work without human supervision, thus the business case doesn't work out. The solution of course is to use gen ai for prototyping and then move to more classical AI for scale.
    1
  22. I don't get the purpose of this system. Sun screen has absolutely no connection to my vacation day allotment. Also to stay safe just use factor 50 sunscreen irrespective of the time you want to spend in the sun. The number of sun hours is not a determining factor, just uv radiation is, and that's typically high even in overcast weather, so there's really no need to look at the weather report. This must be the worst explanation of agents ever.
    1
  23. It's not right to assume that RAG solves hallucinations, it doesn't. Generative AI is inherently prone to hallucinations, in fact all it does is hallucinate - it just produces "most likely" text vv the prompt, and we can't know what that is beforehand. Thus we have to verify every output in order to be sure it is accurate. Studies show 50% of outputs are factually wrong, even when curated input is used.
    1
  24. Weid taxonomy. They are really one and the same thing, except perhaps multi agent.
    1
  25. They are not. Just appear to be. Dangerzone
    1