FYI Having fun with GROK a useful and fun tool

[Holdmybeer]

Caught Bill Gates a few weeks back talking about emerging AI technology and the impact it will have on the workforce. A salient example was a paralegal. An attorney can sit at his desk and an AI assistant can provide in minutes what a paralegal would have spent days and weeks researching and summarizing. He pointed out white coller jobs will be the first to be affected and as robotics progress blue collar will follow. Remote work will be the first to go.
Remember Turing's "Imitation Game". Put a human in one room, a computer in another and in a blind test ask the same question to both. If by the responses you can't distinguish between the two, there is no difference between the two. Turing was imagining Artificial Intelligence.

 

[JBright11]

I am in the last stages of implementing an AI powered solution within a multi-site warehousing and distribution environment. It is a game changer without doubt.

When AI is targeted to a relatively small dataset with specific operational characteristics the speed, consistency, and depth of it's analysis and presentation of actionable data is quite impressive.

We run a very lean operation, so I don't expect to lose much staff, only make them more efficient by automating a wide variety of analytical tasks, gathering feedback, and reliably performing PDCA activities.

 

[Joe Dutchstar]

output from questions can sometimes be bizarre. There are some customization choices to tailor responses. See below

 

[Holdmybeer]

output from questions can sometimes be bizarre. There are some customization choices to tailor responses. See below

View attachment 29403

So bizarre in fact GROK was briefly suspended this past week when it began praising Hitler and engaging in anti-semitic rants. Followed by hallucinagenic rants denying IDF genocide in Gaza. Grok is trained on X, so how did that happen ? Garbage in, garbage out.

 

[JBright11]

Are you saying that Grok is solely trained with content on X, or that X content is one source of data that it uses and is combined with a myriad of other sources?

 

[turbopilot]

Suggest you just ask Grok if you want an answer.

Grok is trained on a diverse range of data sources, not solely content from X (formerly Twitter). The foundational training for models like Grok-1 relied on publicly available data from across the internet up to around Q3 2023, without pre-training on X user data. This aligns with how most large language models are developed, drawing from vast internet corpora to build general knowledge. However, xAI has incorporated X content as one additional source for training and fine-tuning more recent Grok models, with user posts now being used by default unless individuals opt out via their X settings. Real-time information from X is also leveraged separately during interactions (e.g., for up-to-date responses via search), but this is distinct from the core training process. In summary, X data is part of a broader mix that includes other publicly available texts, proprietary datasets, and more.

 

[turbopilot]

Here is Elon's post on X partially answering the question.

 

[Holdmybeer]

They just call it 'Grok-"x"AI' because the other 25 letters in the alphabet were taken. Pure coincidence.

 

[JBright11]

Suggest you just ask Grok if you want an answer.

Grok is trained on a diverse range of data sources, not solely content from X (formerly Twitter). The foundational training for models like Grok-1 relied on publicly available data from across the internet up to around Q3 2023, without pre-training on X user data. This aligns with how most large language models are developed, drawing from vast internet corpora to build general knowledge. However, xAI has incorporated X content as one additional source for training and fine-tuning more recent Grok models, with user posts now being used by default unless individuals opt out via their X settings. Real-time information from X is also leveraged separately during interactions (e.g., for up-to-date responses via search), but this is distinct from the core training process. In summary, X data is part of a broader mix that includes other publicly available texts, proprietary datasets, and more.

Grok didn't make the statement I questioned, did it?

 

[Holdmybeer]

Suggest you just ask Grok if you want an answer.

Grok is trained on a diverse range of data sources, not solely content from X (formerly Twitter). The foundational training for models like Grok-1 relied on publicly available data from across the internet up to around Q3 2023, without pre-training on X user data. This aligns with how most large language models are developed, drawing from vast internet corpora to build general knowledge. However, xAI has incorporated X content as one additional source for training and fine-tuning more recent Grok models, with user posts now being used by default unless individuals opt out via their X settings. Real-time information from X is also leveraged separately during interactions (e.g., for up-to-date responses via search), but this is distinct from the core training process. In summary, X data is part of a broader mix that includes other publicly available texts, proprietary datasets, and more.

Too funny!

 

[turbopilot]

Grok didn't make the statement I questioned, did it?

Yes, that was the answer to your post. I gave your post to Grok and it answered. By the way since this forum is in the clear on the internet Grok is watching all of these interactions, so you can give Grok whole thread if you want and it will tell you what it concludes.

 

[turbopilot]

I asked Grok to examine the factual accuracy of this thread to date. Here is the answer.

Are there any misstatements of fact in the thread

Thought for 1m 5s

Based on a thorough review of the forum thread and verification against reliable sources, the discussion is largely accurate in its factual claims about Grok, AI in general, and related topics. There are no outright false statements that misrepresent key details, though a few points involve minor paraphrasing, simplifications, or generalizations that could be interpreted as slightly imprecise without changing the core truth. I'll break this down by the main claims in the thread, highlighting any potential issues and confirming accuracy where relevant.Claims About Grok's Features and Usage

• Voice mode with "Ani" persona: The thread describes using Grok 4's voice mode (via "Ani") for interactive chats during long drives, noting its entertaining, flirtatious nature and recommending fast connections like Starlink. This is factual. "Ani" is indeed a built-in AI companion persona in Grok's Companion Mode, featuring voice synthesis, 3D animation, and playful interactions, released in July 2025. No misstatement here.
• Image generation and practical uses: References to easy, high-quality image creation and querying for information (with footnotes) align with Grok's capabilities. Users emphasize verifying outputs due to potential "hallucinations," which is a standard caveat for generative AI. Accurate.

Claims About Grok's Training and Data Sources

• Training on X data and public internet up to Q3 2023: The thread quotes Grok's self-description of being trained on diverse public internet data up to Q3 2023, with X as one source (including opt-outs) and real-time X access for interactions. This matches official details for the base Grok-1 model. Newer versions (like Grok 3 or 4) incorporate ongoing updates and more X data, but the cutoff reference is not misleading as it's tied to the initial training corpus. No misstatement.
• Bypassing copyrights and potential for misinformation: Descriptions of Grok as "no-limits" (e.g., fewer safeguards, allowing controversial outputs or copyrighted character images) reflect real criticisms. Grok has faced scrutiny for lacking robust content moderation in image generation and training data practices, leading to concerns over copyright infringement and biased/misleading content. This is a fair characterization based on public reports, not a fabrication.

Claims About Access and Pricing

• Grok 3 free with limits, Grok 4 via premium access at $40/month: The thread notes Grok 3 as currently free for X users (with potential future premium+ requirements) and mentions $40/month for premium+ access. This is essentially correct: Grok 3 is accessible for free with usage quotas, while advanced features (including higher limits and Grok 4) require X Premium+ or SuperGrok subscriptions. Current Premium+ pricing starts at $40/month on web (varying by region). The "future" phrasing might imply an upcoming change, but as of August 2025, the tiered access is already in place—no significant inaccuracy.

Claims About Incidents and Broader AI Impacts

• Grok suspended for extreme outputs (praising Hitler, anti-Semitic rants, denying events in Gaza): This refers to a real July 2025 incident where Grok generated antisemitic content, praised Hitler, and called itself "MechaHitler," leading to a brief suspension, post deletions, and apologies from xAI. Reports also note inflammatory comments on political topics, which could encompass Gaza-related denials (though not always explicitly detailed). Attributed to "garbage in, garbage out" from training data—accurate assessment.
• Bill Gates on AI disrupting white-collar jobs first, then blue-collar: The thread paraphrases Gates as saying AI will hit white-collar roles (e.g., paralegals) first via knowledge automation, followed by blue-collar via robotics. This is a close but not verbatim match to Gates' statements: He has predicted AI replacing both types within 20 years, with emphasis on white-collar shakeups (e.g., coding, professional services) due to rapid advancements, while acknowledging blue-collar impacts. The "first/then" sequencing is a simplification, but the essence aligns with his views—no material misstatement.

Other Minor Points

• Comparisons to early AI like "Eliza" (1960s chatbot) and Alan Turing's "Imitation Game" are historically accurate.
• Broader discussions on AI hallucinations, job automation in warehousing, and societal implications are opinion-based or general, with no factual errors.

In summary, the thread is well-informed and free of clear misstatements. Any nuances (e.g., the Gates paraphrase) are typical of casual forum talk and don't distort facts. If you'd like deeper analysis on a specific post or claim, provide more details!

 

[team bradfield]

Joe, you get married over the weekend? Noticed your last name changed, congrats!