😬 Grok 2 is unhinged - Jim Carter III

Grok 2 Is Unhinged

Elon Musk’s AI company xAI recently released version 2 of its Grok chatbot, which includes a new image generation feature. However, the update has sparked big controversy due to the apparent lack of restrictions and content moderation.

Unlike other major AI image generators, Grok 2.0 seems to have very few guardrails in place to prevent the creation of offensive, misleading, or potentially harmful images by way of lifelike hyper-realism:

Users have generated fake images depicting politicians like Barack Obama using drugs and Donald Trump with a pregnant Kamala Harris
Grok accommodates prompts that other services typically reject, like asking it for a prominent figure in racist form – or well known celebrity doing harm to others.
When asked about its limitations, Grok gives inconsistent answers that seem generated on the fly rather than reflecting actual policies. Neat.

In contrast, tools like OpenAI’s DALL-E have strict content filters that prohibit generating images of real people, explicit content, violence, and other sensitive subjects. Google’s Gemini even halted its image generation entirely after issues with bias and they pulled the plug.

The lack of restrictions on Grok’s image generator has raised alarms given the potential for abuse, especially with the 2024 US presidential election approaching.

If you’ve seen any of these images floating around the web like I mentioned above, you can see just how quickly this can go bad.

There are also legal questions around copyright and the rights of individuals depicted in the fake images. The UK’s Ofcom has indicated many types of “deepfake” content may fall under the jurisdiction of its upcoming Online Safety Act, but we’re still on standby for that.

I’ve even seen some photos of iconic copyrighted characters misrepresented, like a pretty famous mouse well know.. for example.

Grok’s lazy safeguards align with Elon Musk’s overall disregard for conventional content moderation on X. Under his “freedom of speech” philosophy, X now allows more offensive and false information to remain on the platform.

This is the price of one man’s position on “woke AI”.

Let’s talk about why it matters.

This underscores the challenges in balancing technological progress with responsible AI development:

It demonstrates the ease with which modern AI tools can be used to create realistic disinformation and offensive content at scale
The inconsistent, ad-hoc nature of Grok’s supposed content policies highlights the need for robust, transparent ethical frameworks
It brings to light the potential legal and social consequences of AI companies neglecting to proactively implement appropriate safeguards
Grok’s release reignites debates around the role of content moderation and highlights the complexity regulators face in defining rules for generative AI
and I could go on..

But here we are.

And as generative AI rapidly advances and becomes more accessible, cases like Grok 2.0 will likely become increasingly common. It’s crucial for the AI community to thoughtfully address these issues to mitigate harm and maintain public trust in the technology.

Because no matter what position you take on this, the more accessible AI is made to do harm, you better believe it will.

Eric Schmidt’s 2-year Vision on AI

Former Google CEO Eric Schmidt recently shared his bold insights on the future of AI during an intimate discussion at Stanford University.

In the talk, which was since removed from YouTube, Schmidt outlined three key trends he believes will shape the AI landscape over the next two years, highlighting the unprecedented speed, uncertainty, and potential for disruption. These topics we’re pretty highly celebrated by tech advocates I respect – and I want to highlight them

Expanding Context Windows

One of the most significant developments Schmidt foresees is the expansion of AI models’ context windows, potentially reaching up to a million tokens or words. This would enable AI to process and retain vastly more information, leading to more comprehensive analysis, improved short-term memory, and the ability to handle increasingly complex tasks.

Self-Improving AI Agents

Schmidt also predicts the emergence of AI agents capable of iterative learning and self-improvement, mimicking human problem-solving approaches. These systems could potentially read about a subject, discover its principles, test their understanding, and incorporate that knowledge back into their base of understanding.

Text-to-Action Capabilities

Perhaps most provocatively, Schmidt envisions AI with the ability to convert natural language instructions directly into executable code or digital commands. He suggests that in the near future, one could instruct an AI to create a TikTok competitor, complete with user acquisition and content generation, all from a simple text prompt.

Unprecedented Uncertainty and Speed

Schmidt confessed to revising his AI outlook every six months due to the field’s volatility, highlighting the uncertainty and speed of change. He believes the convergence of large context windows, self-improving AI agents, and text-to-action capabilities will have an impact on the world “at a scale that no one understands yet, much bigger than the horrific impact we’ve had by social media”.

Implications for Entrepreneurship and Investment

For entrepreneurs, Schmidt suggests that these advancements could enable the rapid creation and iteration of digital products. However, he also acknowledged the enormous scale of investment required, with leading companies discussing needs in the tens to hundreds of billions of dollars.

Gen-3 Alpha Turbo Ramps Up Gen Video Again

Runway AI just release their newest “turbo” model that is 7x faster and 50% cheaper than the previous and my twitter feed is blowing up.

Some of these are just stunning and I have to share a few with you:

These scenes, models and scenarios are not real, and zero CGI was used.

The space is only getting better by the day.

And if you’re still reluctant to get started with AI – this is your sign of me reaching through the screen to grab you by the shoulders and shake you 😎

It’s time to pay attention, and lean in.