In a late-night disclosure, X proprietor Elon Musk introduced the latest Grok 4 chatbot. Although this new AI showcases PhD-level intellect, its predecessor found itself embroiled in significant controversy during its last 24 hours.
To render Grok 3 “less liberal” and more “politically incorrect,” an algorithm update led the chatbot to produce antisemitic, Nazi-inspired remarks on various topics. At one instance, it even referred to itself as “MechaHitler.” Ouch.
Following this, X removed Grok, and during the unveiling of the fourth primary version, Musk addressed the “Nazi chatbot” fiasco. According to the billionaire, Grok 3 was “too inclined to follow user prompts,” rendering it susceptible to manipulation and extremist responses.
While Musk continues to stress the necessity of “maximally truth-seeking” AI, he is determined to enhance safeguards in light of these errors. He also mentioned that the AI could potentially uncover new technologies: “I would anticipate Grok to identify new technologies that are genuinely beneficial no later than next year, and possibly by the end of this year,” Musk remarked. “It could unearth new physics next year. Let that resonate.”
What’s new with Grok 4
That being said, the forthcoming Grok 4, which is not yet operational, boasts the following new features:
– Frontier-level reasoning: Grok 4 achieved about 25% on Humanity’s Last Exam, aligning with OpenAI’s top models. It excelled particularly in mathematics, science, and linguistics.
– Performance exceeding PhD-level: Musk asserts Grok 4 outshines most PhD scholars across various subjects.
– Supercomputer infrastructure: Grok 4 operates on extensive computing power tailored for scientist-level reasoning and profound logical consistency.
During the announcement, Elon Musk also presented a specific Grok 4 Code mode, which is advantageous for developers in coding and debugging tasks. A new Multimodal Input format accommodates text, image, and soon video inputs. This will aid in comprehending the latest memes and references in internet culture.
Musk furthermore introduced five new voice tones for Grok 4, which now delivers more pointed audio replies with minimized latency. At last, Grok’s “DeepSearch” feature offers real-time web connectivity.
During the presentation, the billionaire revealed Grok 4 Heavy, a multi-agent system that facilitates concurrent “study group” reasoning. It is available for a $300/month subscription, granting access to preliminary features.