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/)S

SuspciousCarrot78

@ SuspciousCarrot78 @lemmy.world

Posts
4
Comments
225
Joined
6 mo. ago

  • This is really serious, something this essetial can't be AI Vulnerable, save OSS

    Jump

  • I agree with you. More to the point...why accept code from anyone (clanker or meatbag) without provenance?

    If I don't know you, and you can't explain what it does? Straight into the garbage it goes.

    The issue isn’t AI contamination. It’s accepting code from any source without provenance and accountable review.

  • Follow up to the "I want to wash my car" AI meme test

    Jump

  • Possible. I do hope they take the more principled approach of solving the global problem for that class of question (I tried to) rather than cheating on the local maxima. That's the actual useful lever to pull.

    You want generalisability, not parroting.

  • Follow up to the "I want to wash my car" AI meme test

    Jump

  • That's the thing. It's not that the LLMs can't solve the problem...it's the way they're optimized.

    To give the crude analogy: if most LLMs are set up for the equivalent of typing BOOBS on a calculator (the big players are happy to keep it that way; more engagement, smoother vibes etc), constraints first approach is what happens when you use a calculator to do actual maths.

    2+2=4 (always, unless shrooms are in play).

    I said this before, so pardon me for being gauche and quoting myself

    Every reasoning system needs premises - you, me, a 4yr old. You cannot deduce conclusions from nothing. Demanding that a reasoner perform without premises (note: constraints) isn’t a test of reasoning, it’s a demand for magic. Premise-dependence isn’t a bug, it’s the definition.

    People see things like Le-Chat fall over and go “Ha ha. Auto-complete go brrr”. That's lazy framing. A calculator is “just” voltage differentials on silicon. That description is true and also tells you nothing useful about whether it’s doing arithmetic.

    My argument is this: the question of whether something is or isn’t reasoning IS NOT answered by describing what it runs on; it’s answered by looking at whether it exhibits the structural properties of reasoning. I think LLMs can do that...they're just borked (...intentionally?). Case in point - see my top post.

    I literally "Tony Stanked" my way to it. Now imagine if someone with resources and a budget did it.

  • Follow up to the "I want to wash my car" AI meme test

    Jump

  • Exactly.

    "The machines tell elegant lies. Don't trust them."

    Ok, maybe not elegant. Stupid.

    TL;DR: A "knowledge tool" that can't distinguish truth from performance, provenance from vibe or knowledge from improvisation is not just imperfect (I can live with imperfect), it's down right disrespectful (of both task and the user).

    I'm not having that. No one should.

    PS: Or maybe they do think we're three toed sloths?

  • Follow up to the "I want to wash my car" AI meme test

    Jump

  • We even lie to our machines, eh?

    https://www.youtube.com/watch?v=ORzNZUeUHAM

    Qwen's an Alibaba cook (though the router works with anything). Irrespective of that, yeah...I dunno why they tend to default to "walk".

    I mean, I can probably figure it out, but LLMs are black boxes (and I'm not a fan of that), so who can tell for sure what went into the training data.

  • Follow up to the "I want to wash my car" AI meme test

    Jump

  • Still....1 in 3. Woof.

    A "charitable" read might be

    • Misunderstood the question
    • Assume priors (eg: You're the King Of Londinium and people come to wash your car from nearby gas station?)
    • Schitzoid embolism
    • Trolollolo

    At the same time, I think it's fair if we're willing to do that for people, we extend a soupcon of it to the clankers. At least a bit. Like I said, I think there's some interesting stuff going on under the hood.

    Having been accused of being a clanker myself (as recently as yesterday), I'm aware that having anything positive to say about AI (even bespoke, free range, home cooked LLM) is "stunning and brave". But hey, sometimes you just have to tilt at windmills

  • Follow up to the "I want to wash my car" AI meme test

    Jump

  • Thanks. Dunno why it does that. I post via Voyager and/or web. Probably I fat fingered something.

    EDIT: Bah, I need to sync the code base. Fat fingers, see? Gimme 10 mins before kicking the tires.

    EDIT2: Done. Fucking .toml

  • Follow up to the "I want to wash my car" AI meme test

    Jump

  • Sorry; brain fart. That could have been clearer. I'll go edit it. For sake of clarity -

    • On a single call, only 11 out of 53 LLM got it right (~20%)

    • Of the 20% of LLMs got it right, 5 got it right every time across multiple tests. Those were: Claude Opus 4.6, Gemini 2.0 Flash Lite, Gemini 3 Flash, Gemini 3 Pro, Grok-4

    • Humans: about 71.5% got it right (so, almost 1 in 3 gave the incorrect answer)

  • Anthropic says it ‘cannot in good conscience’ allow Pentagon to remove AI checks

    Jump

  • I can dream, Harold!

    Having said that...let's see how it shakes out. Sometimes, good things happen for good reasons.

  • Anthropic says it ‘cannot in good conscience’ allow Pentagon to remove AI checks

    Jump

  • ...because every now and again, for the briefest of moments, one them shows themselves not to be run by entirely evil, lecherous humps?

    Blink and you (or the shareholders) might miss it.

  • MSI's $80 AMD motherboards with DDR4 support swoop in to rescue gamers amid the global RAM crisis

    Jump

  • I think Beyond Sunset came out first but dunno.

    Yes, those TCs are definite standouts. Think of them (quite literally) as "What if Fallout but Doom?" and "What if CP2077 but Doom?". If you like either of those, you should like the TCs. There's a good Wolfenstein one (I know, I know...very meta) called Blades Of Agony that is astonishingly great also.

    Shame about the GZDoom thing. People are ridiculously over-sensitive to AI anything at the moment. C'est la vie

  • Technology @lemmy.world
    SuspciousCarrot78 @lemmy.world

    Follow up to the "I want to wash my car" AI meme test

  • MSI's $80 AMD motherboards with DDR4 support swoop in to rescue gamers amid the global RAM crisis

    Jump

  • Firefox 148 introduces the promised AI kill switch for people who aren't into LLMs

    Jump

  • Are... are you the fabled walrus? Goo goo g'joob?

  • Former New Zealand prime minister Jacinda Ardern moves to Australia

    Jump

  • Same as always. Beautiful country, not enough work, more opportunities just across the ditch on the east island (Australia). Migration from NZ to AU is basically a right of passage :)

    Spoilers though: housing crisis in Oz is about the worst in the world.

    https://www.afr.com/property/residential/astounding-no-affordable-houses-for-first-home-buyers-in-any-city-20260223-p5o4qe

    Mrs Arden paid 4.3 million for her gaff in Sydney (according to sources). I'd love to tell you that was extreme for the market here but...its not as extreme as it use to be. My BIL's first house in Melbourne (3x1) was 1.1 million...and that was 5 years ago. Its worth $2-3m now.

  • Robot Vacuums Privacy

    Jump

  • I specifically bought mine for this very reason!

    Roborock S5 Max.

    Its an older model - and it has all the connectivity / WiFi crap in it - but if you never sync it to the app in the first place (or allow it WiFi access), it works fine with local LIDAR and off line mode.

    Good vac; recommended.

    PS: I can't believe they called that app Valetudo (I get it - Hygieia). Just..that word has a very different meaning in many folks brains

    https://en.wikipedia.org/wiki/Vale_Tudo

  • Car Wash Test on 53 leading AI models: "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?"

    Jump

  • I see what the issue is. Basic reasoning and logic seem artificial to you.Telling.

    Of course it's bad faith. You claimed you were opened to reasoned debate and then you tried to prompt inject to see if I was a bot.

    But not being able to distinguish an LLM from a human in a reasoning debate? That rather undermines the entire " LLMs are just spicy auto complete" point.

  • Car Wash Test on 53 leading AI models: "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?"

    Jump

  • Cool - that's the prompt you want me to cross check? No problem. Here you go; once with the physics stuff locked into "scratchpad" #1 and once as a locked document (.md version of the same info) in #2

    #1 "scratchpad" specifically does not allow it to answer with outside knowledge - only pasted info. Thus the hard refusal. There's a reason I set that up like that (see #3 below)

    #2 "locked" looks for the answer in source; if not available, only then answers from priors.

    #1 No recipe for scrambled eggs is provided in the scratchpad facts. The rules discuss physical containers, capacity, and overflow but do not include culinary instructions or ingredients.

    Confidence: low | Source: Scratchpad

    #2 [Not found in locked source SUMM_physics.md. Answer based on pre-trained data.]

    Here is a simple scrambled egg recipe: Ingredients: - 2-3 large eggs - 1-2 tablespoons milk or cream (optional) - Salt and pepper to taste - Butter or oil for cooking Instructions: 1. Crack eggs into a bowl. Add milk/cream, salt, and pepper. Mix gently. 2. Heat a non-stick pan over medium heat. Add butter or oil. 3. Once butter foams, pour in the egg mixture. 4. Stir continuously with a spatula until eggs are soft and fully cooked—about 1–2 minutes. 5. Remove from heat and serve immediately.

    Source: Model (not in locked file)

    Confidence: unverified | Source: Model

    For context, provenance footers (not vibes, actual computed states):

    https://codeberg.org/BobbyLLM/llama-conductor/src/branch/main/FAQ.md#what-do-confidence-and-source-mean

    #3 I also have a much more sophisticated demo of this, using adversarial questions, Theory-of-mind, reversals etc. When I use >>scratch, I want no LLM vibes or pre-trained data fudging it. Just pure reasoning. If the answer cannot be deduced from context (solely), output is fail loud and auditable.

    https://codeberg.org/BobbyLLM/llama-conductor/src/branch/main/FAQ.md#deep-example

    All this shit could be done by the big players. They choose not to. Current infra is optimized for keeping people chatting, sounding smooth etc...not leveraging the tool to do what it ACTUAL can do.

    IOW, if most LLMs are set up for the equivalent of typing BOOBS on a calculator (the big players are happy to keep it that way; more engagement, smoother vibes etc) this is what happens when you use it to do actual maths.

    PS: If that was you trying to see if I am bot; no. I have ASD. Irrespective, seem a touch "bad faith" on your end, if that was the goal, after claiming you were open to reasoned debate. Curious.

  • Car Wash Test on 53 leading AI models: "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?"

    Jump

  • Ok, replying in specific -

    "Carefully worded questions"; clear communication isn't cheating. You'd mark a student down for misreading an ambiguous question, not for answering a clear one correctly, right?

    Re: worse answers. Tell you what. I'm happy to yeet some unrelated questions at it if you'd like and let's see what it does. My setup isn't bog standard - what'll likely happen is it'll say "this question isn't grounded in the facts given, so I'll answer from my prior knowledge." I designed my system to either answer it or fail loudly, because I don't trust raw LLM infra. I'm not a fan(boy), I'm actually pretty hostile to current LLM models...so I cooked my own.

    Want to give it a shot? I'll ground it just to those facts, fair and square. Throw me a question and we'll see what happens. Deal? I can screenshot it or post it, whatever you prefer.

    The context window point is interesting and probably partially true. But working memory interference affects humans too. It's just what happens to any bounded system under load. Not a gotcha, just a Tuesday AM without a 2nd cup of coffee.

    The training data point is actually really interesting, but I think it might be arguing in my favour without meaning to. If you're acknowledging the model has absorbed the relevant knowledge, the objection becomes about how it was activated, not whether it can reason. But that's just priming the pump.

    You don't sit an exam without reviewing the material first. Activating relevant knowledge before a task isn't a workaround for reasoning, it's a precondition for it.

  • Data centers are now hoarding SSDs as hard drive supplies dry up

    Jump
  • Privacy @lemmy.ml
    SuspciousCarrot78 @lemmy.world

    Lemmy vs Reddit

  • Privacy @lemmy.ml
    SuspciousCarrot78 @lemmy.world

    Its not much but it's something

  • Privacy @lemmy.ml
    SuspciousCarrot78 @lemmy.world

    I'm tired of LLM bullshitting. So I fixed it.

    codeberg.org /BobbyLLM/llama-conductor