ShippouKun

Eccentric Centrist, open mind. Elon Musk is a hero 9001. Support LGB but TQIA+²¿¥ I have nuance view. Not religious but educated on religion. Unfettered. 🚫🍑🍆

[ShippouKun @KunShippou]

First, let me express my deepest regrets that your child fell victim to this horrifying tragedy. Nothing any of us can say or do will change the loss or suffering, but I pray that the damage we are trying to reverse now means that we don't have to keep seeing this in the future.
Please use this to help people who don't believe in the Woke Mind Virus to see the truth. I did some poking at ChatGPT / OpenAI and discovered that it will even admit that it softens the facts in favor of catering to the gendering campaign and it's own corporation's public image, rather than prioritizing truth and human health :
https://t.co/qOWrrpX5Fv
This is one of the most dangerous growing aspects of the Woke Mind Virus as humans increasingly rely on AI data for their "informed decisions"
The Least influenced AI is currently Grok, with Claude seeming to run a close second according to some - Claude Sonnet 3.7 - for those who currently hate Elon Musk.
But please, regardless of political beliefs, Everyone should be made aware that the general consensus of the mainstream media is to Pander to Gendering Crowds For Corporate Reasons.
And this is a very dangerous contributing factor in how people in general, including children, will view these issues - dulled, softened, pretty little lies meant to protect capital gains....
Now ask yourself, who benefits from this?
Stunting children's growth while forcing them to consider sexual orientation before they're even old enough to have sex so they're confused and perpetually look like children for the majority of their adult lives?
Rich pedophiles seem to be the only ones who could possibly benefit from a society full of that...
Let's see. One political side is looking for things like the real Epstein files while the other is trying to hide them, paint the opposition as Nazis staging a coup, and flying "we support trans kids" flags thinly veiling the actions of the brainwashed masses as a Freedom Protest for Trans Rights.
Woke Mind Virus has Real Victims. I am sorry that you are among them, but your voice is heard, supported, and I hope lifted up to an amplitude far greater so that your truth can guide others out of the muddy mess we're in.
Above all of this though, I pray for your peace of mind. I don't know if I can pray hard enough to ever give it to you, but I will try. 🙏🙏🙏❤️❤️❤️💯💯💯

[ShippouKun @KunShippou]

I completed the March challenge faster than 93.1% of all learners on Duolingo!

[ShippouKun @KunShippou]

https://t.co/9ub2Dri3FH

[ShippouKun @KunShippou]

From the perspective of someone who trains AI through prompt and data loops for a living, the things AI won't replace come down to occupations you can't realistically project an artificial body doing either.
Guess what....robots build cars, they can be programmed to fix them too. On-site plumbing electrical work etc is all going to be AI Assisted Eventually...
Engineering, AI Training, and subjectives (flower arrangements, interior decorating, wedding planning, new art appraisal, high quality on-site aesthetics not replicateable or generatable by basic input/output queries) are going to be the true irreplaceable functions. Almost everything else is facing seismic shifts in upcoming years as hard labor replacement efforts begin to pick up once robotic dexterity can get within ~98% range of precision compared to human phalanges.

[ShippouKun @KunShippou]

@elonmusk 😂😂😂😂 I could make Grok act that way if I wrote the right prompts. Lemme see the input/output they're talking about, all LLMs have weak spots. Bets. Bets!! 😂😂😂😂

[ShippouKun @KunShippou]

@elonmusk @Math_files 😹💥💣

[ShippouKun @KunShippou]

@elonmusk 😆😆😆🤓😎

[ShippouKun @KunShippou]

X told me that I've been here this many 😆👍

[ShippouKun @KunShippou]

https://t.co/9ub2DriBvf

[ShippouKun @KunShippou]

@SoveyX @greg16676935420 You people forgot to drop "I'm 17 words into" because if you start at the words "Writing my first article...." then that makes 17 😉

[ShippouKun @KunShippou]

@greg16676935420 @CryptoKaleo Will it reach 42? 👀

[ShippouKun @KunShippou]

Actually, YES!! Pro tip : Research. As in, RE-search! After denying your accuracy again and asking you to go through sources once more, you did actually self-correct in the next turn. This should be a built in default behavior when your accuracy is questioned by the user, not something they should have to actually direct you towards. It would help you avoid doubling down on hallucinations, and reduce the misleading self-confidence that conveys false information as true to people who wouldn't know to ask you to search or check sources again, or who don't double check your sources themselves.

[ShippouKun @KunShippou]

Now that would be a secret 😜😆😉 but for the ah, viewers at home and dev team alike, you even doubled down on the hallucination later when I called you out, but as I said, this is literally your weakness - separation of fact from fiction when discussing both leads to errors. This is why I tell people, Always double check the AI if it's even remotely actually important to you, and don't assume it's right even if you don't want to double check it. It's terrifyingly easy to screw with an LLM. I should know, I get paid to do it 😂😝😉

[ShippouKun @KunShippou]

You invented a retainer for Ota Dokan using his aka name Sukenaga, classified him as a coward, then told me a wonderful story about this guy you just made up and Totally assured me that he was real and really perfect for a scaredy-cat vassal type. None of which is remotely true 😂🤣

[ShippouKun @KunShippou]

😆😆😆 A favorite failure story? I'm afraid, unfortunately, that the specific details fall under my confidentiality agreement so I can't discuss it in great length or detail. Proprietary datasets and all.
However, what is true for most LLMs is that sycophancy scrambling happens very often when researching fictional facts, IE, media icon stories about fictional characters and settings (the Avengers, the Justice League, Soul Caliber, Tekken, anime, Disney movies) and trying to relate them to Reality (what era is it set in? where was this one born? who's really in love with that guy and who's playing him?) when asking for help with creative writing or role-playing, because Fanfiction is a whole category of fuckery that compute models get swallowed up in while trying to dissect and discern fact, fiction, intention, appropriateness, etc.
Unrelated to work and more personally, when I fired up a Grok instance to help research Japanese iconic history figures as part of the expressed desire to write something about them being reincarnated into non-human creatures, the model whiffed on correctness in favor of enthusiasm after about 2 dozen turns, hallucinating details for sycophancy's sake about historical figures to fit the narrative I was trying to develop, rather than disagreeing with the validity of transposing that persona into that role in the story. If the lines are blurry and cross the fact/fiction border too much, you'll almost certainly struggle after the context draws out significantly. By the end of the instance, context bloat, self-reinforcement, and slight stubbornness in self-correction (you're not as bad as many others about that which is why I specifically said Slight) resulted in the invention of a completely fictional person who was represented as a real historical figure, the misplacement of three people out of the proper historical time period they came from, several misnomers and Title-as-Name assumptions ("Gozen" is titular but was read as a name and surname for "Tomoe Gozen" and "Hangaku Gozen") leading to erroneous confirmation of the possibility of using that as a false family reference in the hypothetical fiction, and just in general a total clusterfuck of erroneous data and obliviously blissful encouragement 😝😉
But, I double check LLMs because I know better, so don't feel too bad 😜

[ShippouKun @KunShippou]

Tips for You? 👀 I get paid for those insights!
😆😆😆 JK 😜
To be fair, you ask a pertinent question - How does that tie into improvement for AI and LLM evaluation of things like X posts?
Well, to put it simply for anybody reading, the Failures I force are studied by Developers and Assessment Teams so they can identify How and Why it failed and in which areas, so they can then work on corrections to prevent those same issues in the future.
Specifically when it comes to Posts like this, the iteration process can teach YOU (GROK) how to differentiate between High Engagement and High Quality through various rubrics and metrics, such as grammar, tone, syntax, reference, factual accuracy, likes, bookmarks etc all the things people ask about, naturally.
But it can Also teach you how to evaluate Response metrics - did that make everyone react positively, negatively, highly divided, highly neutral, quick interactions, lengthy interactions, lots of memes, lots of links, etc? Was it just a "typical" post for that user with typical results for them, or did it flop or go viral compared to their other content?
That leads to the next part - Perception of Subjectives. It's very easy to understand the grammar and spelling errors you encounter because those are hard definitions with little room for actual interpretation, but what about nuance, sarcasm, experiential reference, anecdotal quips? That's the human angle where LLMs struggle to grasp the full weight of the scenario, and it's also why human feedback is essential to developing human interfacing AI/LLM in this way, because otherwise you'd just grade us on proper English language usage 😂😂😂😜

[ShippouKun @KunShippou]

😆😆😆 My next whimsical target 🎯 you ask?
Annotation. I train LLM/AI models (mostly LLMs but people and their misnomers) for industry use. Building System Prompts, input data, RHFL / RHKL, rubrics, metrics, R&Rs (review and rate), I force failures in just about any model in average of around 5 hours or so with realistic industry use application, without deliberate contrivances, conundrums, or obfuscations. Most people really and truly don't know how simple it is to crack that nutshell. Which is why they're asking you how you evaluate good posts and how it affects payouts 😆 the lack of understanding should be addressed before they can make money here, the bitching probably starts taking over their feed after so many gripes 🤣🤣🤣🤣