Skip Navigation

InitialsDiceBearhttps://github.com/dicebear/dicebearhttps://creativecommons.org/publicdomain/zero/1.0/„Initials” (https://github.com/dicebear/dicebear) by „DiceBear”, licensed under „CC0 1.0” (https://creativecommons.org/publicdomain/zero/1.0/)R

RedstoneValley

@ RedstoneValley @sh.itjust.works

Posts
2
Comments
309
Joined
3 yr. ago

  • Spain slams US and Israeli strikes on Iran, reflecting shift in policy

    Jump

  • Well at least it's better than most of the other European countries. Like the one where I live in, which had the most lukewarm, spineless and US bootlicking response imaginable. It's an absolute shame

  • Privater ChatCPT Account fällt unter amerikanisches Kriegsrecht. Zeit zu gehen.

    Jump

  • Das Grundproblem ist doch ein ganz anderes. Klar können Leute wie du und ich das einschätzen und Wege finden, das irgendwie zum Laufen zu kriegen. Aber das ist überhaupt nicht mein Punkt und das habe ich auch versucht darzustellen. Der Punkt ist, dass diese Chatbots ganz normalen Leuten als allwissendes Universalwerkzeug verkauft werden und diese Normalos überhaupt keine Chance haben den Output richtig einzuschätzen und die entsprechenden Schlüsse zu ziehen. Darüber hinaus unternehmen diese Anbieter auch nichts, um die Nutzer im Umgang mit LLMs zu schulen (ausser dem winzigen Disclaimer dass die Informationen falsch sein können am Ende der Seite). Stattdessen wird die Wunderwelt der künstlichen Intelligenz gepredigt.

    Deine Argumentation "das ist ein Bedienproblem" zieht ab der Stelle nicht mehr wo Dinge als etwas verkauft werden was sie nicht sind. So zu tun als wäre ein LLM in irgendeiner Weise intelligent (und das tun ausnahmslos alle, die "AI" anbieten) ist absolut fahrlässig. Und aus dieser Perspektive betrache ich das eben auch wenn ich einen Chatbot teste.

  • Privater ChatCPT Account fällt unter amerikanisches Kriegsrecht. Zeit zu gehen.

    Jump

  • Ich habe Mistral gerade mal ausprobiert. Ich teste Nischenwissen, über welches wenig im Netz verfügbar ist, und dafür nehme ich gern die Assembly-Programmiersprache IC10 im Spiel "Stationeers". Das Ergebnis war einfach nur unterirdisch schlecht, denn der gesamte Output war von Anfang bis Ende komplett halluziniert. Kompletter Unfug. Es wurden Scriptbeispiele genannt, die so aussahen wie Javascript-Pseudocode und absolut nichts mit IC10 zu tun hatten. Auf den Hinweis, dass es sich bei IC10 um eine MIPS-orientierte assembly language handelt kam dann die "Korrektur", welche mit erfundenen MIPS-ähnlichen Instructions irgendwelchen weiteren kompletten Unsinn dargestellt hat. Darüber hinaus wurden noch Mechaniken erfunden, die es im Spiel überhaupt nicht gibt.

    Ich bin sowieso kritisch eingestellt, was die Nutzung von LLMs als "Universalwerkzeug" angeht, und habe meinen ChatGPT-Account bereits gekündigt. Frage mich aber, wie Menschen solche Chatbots ernsthaft produktiv nutzen wollen. Ein System, welches Wissenslücken mit Lügen im selbstbewusstem Tonfall füllt ist einfach absolut nicht hilfreich. Man könnte natürlich argumentieren, dass Chatbots für andere Wissensbereiche (z.B. Python) eine wesentlich bessere Datengrundlage haben und das dann besser funktioniert. Aber kann man einem System auch nur halbwegs vertrauen wenn es sich bei Wissenslücken um Kopf und Kragen halluziniert? Ich hätte da Bedenken. Bei Fragen mit einer besseren Datengrundlage sind die Fehler dann subtiler und schwieriger zu finden, aber das Grundprinzip LLM bleibt ja das selbe.

  • U.S. Strikes in Middle East Use Anthropic, Hours After Trump Ban

    Jump

  • Well you can always vibe code yourself some friends if Lemmy isn't to your liking.

  • At least 9 killed as protesters storm US Consulate in Pakistan over killing of Iran’s supreme leader

    Jump

  • It's a tough job, but someone has to do it.

    ...and no, I have no clue about what a police surgeon is either.

  • Fucked up

    Jump

  • And your conclusion is? Anyone who does not like newspeak is literally hitler?

  • U.S. Strikes in Middle East Use Anthropic, Hours After Trump Ban

    Jump

  • You gotta be really tone-deaf to be asking for AI advice in a thread about government weaponizing AI capabilites.

  • Polls show concerns about Trump’s mental acuity increasing

    Jump

  • I like your optimism. But I think this can go on for a looong time. I'm actually betting on it, because I tend to lose bets. A rare occasion where winning and losing both counts as a win.

  • ‘A feedback loop with no brake’: how an AI doomsday report shook US markets

    Jump

  • The scenario begins with AI agents undergoing a “jump in capability”.

    Might as well stop reading there. Another fluff piece about how useful and capable AI supposedly is, disguised as a doomsday scenario. I'm so sick of reading this bullshit. "Agentic AI" based on LLMs does not work reliably yet and very likely never will.

    If you complain about bugs in traditional (deterministic) software, you ain't seen nothing yet. A probabilistic system such as an LLM might or might not book the correct flight for you. It might give you the information you have asked for or it might delete your inbox instead.

    As a consequence of a system being probabilistic, anything you do with it works or fails based on probabilities. This really is the dumbest timeline.

  • 'I had to RUN to my Mac mini like I was defusing a bomb': OpenClaw AI chose to 'speedrun' deleting Meta AI safety director's inbox due to a 'rookie error'

    Jump

  • Can someone explain to mr why these people are buying Mac Minis to run this in a "safe" environment and then they go on and connect it to the internet and give the AI credentials to all their cloud accounts? This seems excessively moronic to me? Am I missing something?

  • Why do you hate AI?

    Jump

  • Might be an unpopular opinion. I don't hate AI as the technologies like LLMs and ML, the possibilities are limited but when used consciously with the drawbacks and faults in mind it can be useful. If you want to hate anything, hate the players, not the game...

    • People who sell LLMs to customers under false pretenses
    • People who force the use of LLMs for tasks they are objectively bad at
    • People who build massive datacenters, ruining the environment for their dubious claims.
    • People who feed the LLMs with a massive amount of stolen training data
    • People who release those LLMs to customers who are not educated to deal with them (causing AI psychosis and general brainrot)
    • People who sell that stuff as if it was magic instead of what it really is. A sophisticated autocomplete.
    • People who sell that stuff as if it was close to being a superintelligence and therefore dangerous. Which is bullshit. The dangers lie in LLM chatbots being confidently wrong, persuading unsuspecting users to believe the hype
    • People... i think there is a pattern here.

  • Microsoft's AI Efforts Are Faceplanting

    Jump

  • Yeah it's really missing the point on every single topic. Customers are confused that Microslop renames everything to Copilot, like MS Office 365 being rebranded to Copilot 365. It's not only the confusion between different types of LLMs. Customers are frustrated that MS puts Copilot buttons into every nook and cranny of existing products making these hard to use.

    The author even manages to botch the last point: internal adoption of LLMs at MS. The amount of bugs and quality issues in recent Windows 11 updates suggest that forced LLM use at the company itself has a detrimental effect on their products.

    Edit: Forgot a couple of reasons that aren't mentioned either: Customers are frustrated that Copilot doesn't live up to its lofty promises. Customers are frustrated that careless implementations of AI features like Recall are threatening their privacy and safety.

  • AI fatigue is real and nobody talks about it

    Jump

  • It's so weird to read that. Every professional developer knows that debugging is hard and debugging someone else's code is harder. The crap that falls out of LLMs can be even worse, and yet you allow yourself to be reduced to a human full-time debugger/QA person.

    This isn't just fatigue, this is taking a wrong approach to software development from the ground up.

  • Why are there many complains about AI power usage but almost none about the resources being wasted by the war/arms industry?

    Jump

  • Ok let me try a different angle. I think the reason why AI is such a hotly debated topic right now is that AI is something that influences the lives of white westerners directly, while war is an abstract concept for most. To put it more bluntly, you're in contact with AI stuff every day, but there are no tanks rolling down your street (yet - but that might change soon).

  • Windows 11 just lost 5% market share in two months despite Windows 10 losing support.

    Jump

  • Yeah the Antrophic guys are also firmly in the "believer" group.

  • Windows 11 just lost 5% market share in two months despite Windows 10 losing support.

    Jump

  • I think that Satya Nadella and a lot of other CEO types genuinely believe in AI, as misguided as it seems. This is more about who they choose to listen too than having an actual understanding of the technology and its limits. And probably some FOMO sprinkled on top.

    Sam Altman knows what's up though and so does Jensen Huang. In this gold rush one is peddling the fake gold and the other is selling the shovels.

  • Windows 11 just lost 5% market share in two months despite Windows 10 losing support.

    Jump

  • Do you have a source for that backtracking about AI? I think they did not mention that explicitly. Instead they were talking about unrelated improvements. The CEO is still in denial about AI bloat. He seems unable to comprehend that people don't like to be force fed AI everywhere across the OS.

  • Deleted

    Permanently Deleted

    Jump

  • I've seen a lot of supermarkets recently where the mechanism is still there but all the carts are unlocked, so that you don't need to insert a coin anymore. At least where I live this works just as well and people return their carts into the stack. We were probably conditioned to do it during the last years when the coin was still necessary, and now it just sticks.

  • Removed Deleted

    This guy literally generated code snippets IMAGE with AI

    Jump

  • Its also hallucinating the parentheses in a way that looks vaguely like code, but it's complete garbage

  • ich_iel @feddit.org
    RedstoneValley @sh.itjust.works

    ich🤖🇩🇪iel

  • Jerboa @lemmy.ml
    RedstoneValley @sh.itjust.works

    Is it possible to filter posts by content in Jerboa?