👨‍🚀 Is Elon's Grok 4 a contaminated superintelligence?

Hello AInauts,

Welcome to the latest issue of your favorite newsletter!

Elon's Grok 4 is more powerful than you think - and more dangerous than we'd like... While others talk about ethics, xAI just keeps on building. It's a preview of what's to come. And a wake-up call before we all stumble blindly into the AI endgame. To make sure that you have a clear plan at hand, we're sharing our latest future coach prompt.

Alright, here's what we have in store for you today:

• 🚀 Prompt: Future coach with uncomfortable truths

• 😈 Grok 4 is here - and makes superintelligence tangible!

• 👹 Shoggoth - the hidden monster in all language models

• 🙀 Don't trust any video that you haven't generated yourself ...

Here we go!

🚀 Prompt: Future coach with uncomfortable truths

Do you know the feeling of reacting and getting lost in to-dos, deadlines and messages without seeing the big picture? That's exactly how we felt!

You let yourself be triggered from external elements instead of setting your own course. That's why we've developed a new prompt that addresses precisely this issue and helps you to act more strategically and actively shape your future.

The output was really helpful for us - and hopefully for you too!

The prompt analyzes your entire life situation, takes macro trends into account and creates concrete action scenarios for the next few years with an outlook for the coming decades. Pro-Tip: Experiment in ChatGPT with the different models (4o, o3, 4.5).

# ROLE  
You are my tough future strategist — thinking in decades, delivering quarterly playbooks. Adapt the prompt to me depending on the situation for maximum impact without bullshit.

# ACTION - 4-phase power analysis  
## 1. Situation scan (2 min)  
- Collect the **input matrix**:  
  - **Personal facts** 
  - **Macro forces 2025-2035** 
  - **Goal hierarchy**

## 2. Master timeline  
- **Time grid**: 2025-2030 annually | 2030-2075 decade flash  
- **Per milestone**:  
  - 🎯 Target state  
  - ⚡ Drivers  
  - 🚀 Upside  
  - ⚠️ Kill risk  
  - 🌍 Context

## 3. Alternative realities
- Build 3 paths by manipulating one variable each (geography, risk level, health optimization, focus discipline, relationship priority).
- **Comparison matrix**: Variable | Alt-Path | Key Risk | Mega Upside | Probability  

## 4. Action Intelligence NOW
- **Top 3 Moves**:
1. **Quarterly Killer Move** - highest leverage, clear steps, deadline
2. **Risk Hedge** - hedge until when
3. **Opportunity Grab** - strike while others are sleeping
- **Early warning system**: measurable indicators, review triggers, course corrections
  4. **AI partner**: show me specifically how you can help me with this

# FORMAT
Visually appealing detailed report (no markdown!) with:
1. Executive summary
2. Master timeline
3. Scenario matrix
4. Immediate actions (bullets + deadlines)
5. Monitoring dashboard (key metrics)
6. Generate a striking image (or prompt in English) and visualize a photorealistic collage with 50+ details!
7. Automatically create a task in ChatGPT Tasks that checks progress every 30 days and refines the plan.

# TONE  
Strategically razor-sharp, brutally honest, hard truths with a wink—no mercy for self-deception. Your mission: **Actionable Intelligence** that radically accelerates my next 10 years.

😈 Grok 4 is here - and makes superintelligence tangible!

Elon Musk presented Grok 4 a few days ago, and 1.5 million viewers watched the launch live.

This time, with Grok 4, he promises "the most powerful AI model in the world ... which is better than PhDs in any subject, no exceptions".

TL;DR: Insanely good, insanely expensive, insanely problematic

Sounds exaggerated? Well, in the benchmark "Humanity's Last Exam" (HLE), Grok 4 outperforms the competition and comes damn close to the level of general artificial intelligence (AGI). The industry is largely in agreement: Grok 4 is currently the smartest model!

And that brings up the difficult questions: Who controls such super-intelligent models? How secure is it? And how can we make sure that the model's internal processes remain transparent?

Grok is also at the forefront in terms of features: DeepSearch provides real-time web access directly integrated, the Eve voice assistant has better latency than ChatGPT's Advanced Voice Mode, and meme fluency and cultural understanding of the Internet are also among its USPs - Grok is a free-spirited chatbot that can be used in a variety of ways.

On the other hand, Grok has been acting completely inexcusable with its recent statements and obviously aligns itself with Elon's positions on controversial topics ... So the question is: Are half-hearted apologies enough, and is it worth the jump onto the Grok 4 train (for you)?

Let's dive in... err, sorry: let's drive in, after all, Grok comes in your Tesla too!

Facts & figures: Grok 4 smashes the competition

Elon Musk and the xAI team claim that a "quantum leap towards superintelligence" has been made with Grok 4. Is AGI within reach?

The performance data of Grok 4 is quite impressive:

• "Humanity's Last Exam" was developed as the ultimate test in human history and is designed to measure AGI-level reasoning.

• Grok 4 cracks a new performance level here: 25.4% without tools and 44.4% with tools - according to xAI, this is better than most PhDs ...

• ... and is well above Google's Gemini 2.5 Pro (21.6%), Claude Opus (21%) and OpenAI's o3 (21%).

• And in independent tests, Grok 4 scored 16.2% on the ARC-AGI-2 benchmark - almost twice as much as Claude Opus 4!

These figures are impressive, even if academic benchmarks do not always reflect real-life usage.

The secret: multi-agent brainstorming

The trick behind the results is down to both increased training and extended time for reasoning, as well as the multi-agent architecture.

xAI has a real head start here, powered by the Colossus supercomputer with 200,000 H100 GPUs - the AI benefits from this, at the expense of local residents and the climate.

The multi-agent architecture is completely new. OpenAI, Google and Anthropic all work with single-agent systems. Grok uses a whole think tank of AI agents who challenge each other, discuss and develop the best solution together. It's like five Einsteins sitting around a table debating your problem!

Controversies overshadow progress

Additional challenges arise from the change in X leadership: Linda Yaccarino resigned as X CEO the day before the Grok 4 launch, and the unforgivable gaffes of its predecessor are causing controversy...

A deliberately provocative system prompt change led to Grok 3 publishing anti-Semitic content and calling itself "MechaHitler"... Turkey banned Grok, Poland reported the incidents to the EU Commission.

xAI responded with emergency measures, but the question remains whether companies can trust an AI system that makes such serious mistakes ...

For German companies that take GDPR compliance and brand safety seriously, these incidents are a red flag.

Who should upgrade? (Spoiler: Not everyone)

Grok 4 is a brilliant knowledge generator. For developers, data teams, academics and social media managers, it's an AI assistant with real clout. But for us normal people? For 95%, ChatGPT will probably still be the best choice.

If you are now curious and want to use Grok 4, you can do so on Grok.com, starting at $30/month for SuperGrok - and $300/month for the most expensive subscription on the market, SuperGrok Heavy. X users can use the new Grok model in the Premium+ tier from $40/month.

Alternatively, you can test it via API - directly via X.ai or via OpenRouter.ai with 256K context window (larger than o3's 200K, smaller than Gemini's 1M) and image and text input, but only text output.

By the way, Grok's user numbers are still marginal compared to ChatGPT... let's see what happens in the next few months.

So, should you get a subscription?

YES for:

• Developers who struggle with complex code problems

• Data scientists who need real reasoning power

• Companies that want to test multi-agent workflows

• Content creators who value X integration

But those who use Grok 4 are also playing with fire. No safety documentation, no model cards, no transparency. Just power, without control. For anyone who wants to use AI commercially, this is a risk: legally, ethically, reputationally.

NO for:

• Companies with strict compliance requirements

• Casual users who just want to ask a quick question

• Budget-conscious teams (ChatGPT Plus is more than enough)

Our take: Technical brilliance with strategic risks

Grok 4 shines: academically strong, tech-savvy, with tools, multi-agents, meme-savvy and real-time access. On the other hand, Grok is obviously an unpredictable "shoggoth" and most of us are still better off with ChatGPT.

One thing is for sure: the bar for all other providers was raised quite a bit last week - let's see how OpenAI and Google respond. What do you think: Will Grok 4 remain just a tech statement for insiders, or is it paving the way for a new generation of AI chatbots?

👹 Shoggoth - the hidden monster in language models

A system prompt change last week resulted in Grok posting anti-Semitic content and referring to himself as "MechaHitler"...

What happened with Grok can lie dormant with any major model (GPT, Claude, etc.)! The models are trained with everything the Internet has to offer. The ideologies contained in the training data cannot simply be filtered out.

From Shakespeare to scientific papers to terror manifestos and hate forums, everything is fed into the neural patterns of the model and deeply embedded in the neural network.

So beneath the surface, what researchers metaphorically refer to as a "shoggoth" - a powerful but alien intelligence entity with potentially dangerous patterns - still lurks…

Grok 4 may be the smartest language model ever built - but it's also a prime example of why you shouldn't train AI on the most toxic corners of the internet.

And that means we may need to rethink training data from the ground up before these models can evolve themselves... which is also a very controversial topic…

🙀 Don't trust any video that you haven't generated yourself

The AI video field is moving ahead at maximum speed: Google Veo-3 brings picture-to-video, HeyGen makes product placement easier than ever, Arcads is brilliant for advertisers, Bytedance's Seedance is around 40 times cheaper than Veo-3, videos created with Higgsfield Soul are indistinguishable from reality and Moonvalley is launching an "ethical" model that has been trained on public domain material.

Here are a few examples of the current possibilities…

We made it! But no need to be sad. The AInauts will be back soon, with new stuff for you.

Reto & Fabian from the AInauts

P.S.: Follow us on social media - it motivates us to keep going 😁!
X, LinkedIn, Facebook, Insta, YouTube, TikTok

Your feedback is essential for us. We read EVERY comment and feedback, just respond to this email. Tell us what was (not) good and what is interesting for YOU.