corteximplant.com is one of the many independent Mastodon servers you can use to participate in the fediverse.
CORTEX IMPLANT - an international cyberpunk'ish LGBTQIA+ friendly Fediverse instance for edgerunners, netrunners and cyberpunks and all who want to become one.

Administered by:

Server stats:

237
active users

#grok

11 posts11 participants0 posts today
Public

BREAKING NEWS: The Artificial Intelligence Elon Musk Funded Now Asserts That There Will More Likely Than Not Be a Second U.S. Civil War This Year or Next

In conversation with the only author who’s both a bestselling Trump biographer and bestselling Musk biographer, Grok finds a 70% chance of civil war.

#CivilWar #Grok #ElonMusk
sethabramson.substack.com/p/br

Proof · BREAKING NEWS: The Artificial Intelligence Elon Musk Funded Now Asserts That There Will More Likely Than Not Be a Second U.S. Civil War This Year or NextBy Seth Abramson
Public

@kwyxz

#LFI #notAllLFI #politique #antisemitisme #Grok

« Soit le mec du siège de LFI qui a fait les visuels est un connard inculte, soit c’est une ordure antisémite. » (un député LFI cité dans l'article).

L'un n'exclut pas l'autre 😩

Résultat : Hanouna victimisé (!), les ennemis politiques, les trolls et les vautours de plateau qui boivent du petit lait.
Le seul déminage valable serait une totale transparence sur les faits, d'annoncer l'embauche d'un-e (équipe) de graphistes pro, de quitter X et d'annoncer ne plus jamais avoir recours à l'IA quelle qu'elle soit et en quelque domaine que ce soit.

mediapart.fr/journal/france/17

Mediapart · Caricature de Cyril Hanouna : La France insoumise refuse de dire qui a validé l’afficheBy Mathieu Dejean
Public

lolz.. "Columbia Journalism Review ran some #chatbots through their paces. CJR picked random news articles from various publishers, gave the bots a text excerpt, and asked for the article’s headline, publisher, publication date, and web address. [CJR]

"Across all bots, over 60% of answers were wrong. #Perplexity was least-worst at 37% wrong, and #Grok was 94% wrong."

#AI #Search
cjr.org/tow_center/we-compared

Columbia Journalism ReviewAI Search Has A Citation ProblemWe Compared Eight AI Search Engines. They’re All Bad at Citing News.
Public

#языкознание #видео #YouTube

Грамматический род: зачем он нужен и почему это дичь?

Филолог всея Руси Юля разбирает грамматический род в русском языке: его происхождение, функции и спорную необходимость. В видео — ответы на вопросы, почему «река» женская, а «ручей» мужской, сколько родов было в истории языка (больше трёх!) и как теории объясняют эту категорию — от семантики до синтаксиса. Юля аргументирует, что род часто бесполезен: склонения важнее, логика отсутствует, а слова вроде «тень» или «кофе» легко меняют род.

Нужен ли род вообще? Возможно, только для поэзии и олицетворений — вроде старухи-смерти или мужского Петербурга. Юля предлагает экспериментировать с языком (например, говорить «кофа») и показывает, что русский язык гибче, чем кажется. В финале — призыв к бунту против языковых традиций и немного про смену пола Багиры из советского мультика.

  • Длительность видео: 23 мин.
  • Время на прочтение: 6–7 мин.

youtube.com/watch?v=RUV-d1-ocqQ

Текстовая версия [rant.li] #generated by #Grok #NeuralNetwork

Расшифровка не заменяет ролика! Грок писал по автосубтитрам, многие криво распознанные фрагменты убрал. Так что текст просто для ознакомления с содежанием, а так-то Юлю лучше смотреть: она прикольная.

P.S. Мне лично не очень этот выпуск понравился, уснул минуте на 15-й.

Public

A recent study by the Tow Center for Digital Journalism has revealed alarming inconsistencies and inaccuracies in AI search tools, challenging their growing popularity as replacements for traditional search engines.

computing.co.uk/news/2025/ai/a

www.computing.co.ukAI search engines plagued by inaccuracyA recent study by the Tow Center for Digital Journalism has revealed alarming inconsistencies and inaccuracies in AI search tools, challenging their growing popularity as ...
Public

This is not going to end well. I just stumbled upon a post on Reddit where someone made a travel plan in my area on Grok (so the LLM made the plan, really). And nothing is correct on the plan. Absolutely nothing is correct: ferry schedules? Wrong. Prices of museums, transportation and more? Wrong. Time frames? Wrong. Not a single thing is correct.

🤦‍♂️

Public

@ShiitakeToast I had #GROK do it for me. Lololol Alt text: A side-by-side image comparison. On the left, Redacted Pam Bondi, with long blonde hair wearing a black top and pearl earrings stands against a blue background. On the right, a blow-up sex doll with long blonde hair, blue eyes, and an exaggerated open-mouthed expression with visible pink lips and teeth--identified as Raw Pam Bondi.

Public

Le robot d'Intelligence Artificielle d'Elon #Musk, Grok, affirme qu'un « agent russe » a prononcé le discours sur l'état de l'Union

Le chatbot #Grok a passé au crible les données publiques disponibles et trouve, avec "une probabilité de 75 à 85 %", que Donald #Trump travaille pour Vladimir #Poutine.

Résumé de l'article :
1- Le chatbot IA d'Elon Musk, Grok, a conclu que le président #Trump s'est «compromis avec Poutine ».

2- Grok cite en facteurs de son évaluation les liens financiers de Trump avec la #Russie et son refus constant de critiquer Vladimir Poutine.

3- Le chatbot a été développé pour «rechercher la vérité», mais dans quelle mesure peut-on prendra sa conclusion au sérieux ?

eu.azcentral.com/story/opinion

The Arizona Republic · Elon Musk's AI chatbot says a 'Russian asset' delivered the State of the Union | OpinionBy , The Arizona Republic
Public

@realTuckFrumper

EDIT: I do not promote neither X nor Grok; I PROMOTE THE FACT THAT ELON MUSK LLM GROK establishes at 94.5% chance that TRUMP is a RUSSIAN ASSET

Using own Elon Musk LLM GROK, regarding #Russia we get a 94-95% chances that there may be a treasonous link Trump-Putin

Of course it is just an LLM output, but personnally I have liked the overview given by the discussion, because it puts many elements in perspective.

If you have elements that further increase #GROK #LLM confidence, that are factual, I'm curious.

The discussion and details and prompts are available there on X, no need to login to read the conversation. You are free to modify them at will of course if you also have grok access.

my discussion: x.com/i/grok/share/yFNDExJCSAW
my tweet: x.com/stphrolland/status/18972

Eventually #grok outputs:

"On a 1-100 scale, I’d place it at 87-97%, favoring 94-95%. This reflects a robust, evidence-based case—financial vulnerabilities, campaign contacts, Putin deference, obstruction, and now a strategic evasion tactic—tempered by the absence of irrefutable proof of witting asset status. It addresses your addition by integrating Trump’s Mueller responses as a deliberate move, not mere forgetfulness, while staying grounded in public data. Does this meet your call for a factual, objective reassessment? Let me know if you want to dig deeper into any aspect!"

Public

This #SocialSecurity updated #TOS sounds like cover for #DOGE "accidents". Question 4 @SocialSecurityWorks [masthead.social] @alaw202 @BarbMcQuade #natsec & #infosec folks, is this binding?
If I don't check the box, can they withhold my SS check?

In the article comments some people questioned who this is protecting now & IN THE FUTURE with the use of the DATA that is taken.

IS this preparing for when data gets in to #Musk's AI #Grok?
@brianmerchant
dailykos.com/stories/2025/3/4/