The 'Best Currently Available' AI Chatbot 'Grok' Launched Ahead of Schedule

baoshi.rao

Ahead of OpenAI's first developer conference, where upgrades to GPT-4 and other products are anticipated, Elon Musk preemptively announced the launch of 'Grok,' a large language model developed by his company xAI. Musk believes it is the 'best currently available' AI chatbot.

Musk's AI assistant is called Grok, a verb that means 'to understand the situation (before speaking)'. The Oxford Dictionary describes it as the ability 'to understand (something) intuitively or empathetically'. Considering that intuition is based on linguistic, physiological, and auditory cues—which are challenging to express to others, not to mention machines—naming an AI this way is quite intriguing.

According to the xAI team, Grok is designed to mimic 'The Hitchhiker's Guide to the Galaxy', answering questions with 'a bit of wisdom' and a 'touch of rebellion'. The developers also specifically warn: if you don't like humor, don't use Grok! Strictly speaking, Grok can answer almost any question posed by humans, and even if it doesn't know the answer, it can suggest some thought-provoking questions.

Although the current Grok is similar in form to the previously released GPT-3.5 as a conversational question-answering bot, there are still many differences.

Test results indicate that Grok-1 has shown not only significant improvements over its predecessor but also outperforms GPT-3.5 in mathematics, coding, and multidisciplinary knowledge evaluations. As a hands-on project by Elon Musk himself, Grok's unconventional approach and quality are to be expected. The Grok system notably avoids familiar frameworks like PyTorch or TensorFlow, and even omits Python entirely, opting instead for the Rust programming language and the emerging deep learning framework JAX.

In addition to regular conversations, Grok has other features such as simultaneous output for multiple dialogues, allowing users to write code while asking other questions. When regenerating unsatisfactory answers, users can expand the timeline to directly navigate between different response versions, and even use the built-in markdown editor to manually modify AI responses before continuing the dialogue.

However, what Elon Musk is most proud of is Grok's two 'secret weapons': First, Grok is willing to proactively provide certain answers, such as responding to sharp questions that most other AI systems refuse to address; second, Grok can access real-time information from the X platform—similar to how ChatGPT retrieves instant updates by browsing the internet.

Some netizens have labeled ChatGPT as 'politically correct' while describing Grok as more down-to-earth and straightforward, not caring about others' opinions (Based). While acknowledging ChatGPT's capabilities, they find it somewhat 'boring' compared to Grok.

When answering questions, unlike ChatGPT which strongly emphasizes its lack of emotions and presents itself as just a robot, Grok doesn't hesitate to 'show' emotions and preferences. For example, when a user asks, 'Should bagels be allowed to be hollowed out?' Grok exclaims 'That's terrible!' and asserts 'This is a crime against breakfast and even humanity,' then goes on to elaborate for over a hundred words.

It's important to note that Grok still makes factual errors, meaning it also suffers from the 'bias' issues commonly seen in current large language models. In a screenshot shared by Musk, Grok incorrectly described the jury deliberation in the recent legal case of cryptocurrency platform SBF as lasting 8 hours when it actually took just over 4 hours.

In the official statement on xAI's website, there are repeated calls for more talent to join. Moving forward, xAI will focus on enhancing several areas, such as improving model context understanding and retrieval capabilities, equipping Grok with visual and auditory sensory abilities, and advancing multimodal capabilities to achieve broader applications, including real-time interaction and assistance.

Some netizens believe that xAI has only been established for four months - how much could they have developed on their own? Musk is just riding the hype after all.

In fact, it's no exaggeration to say that without Musk, there would be no OpenAI, and without ChatGPT, there would be no Grok. Musk's founding of xAI is directly related to the great success of OpenAI's ChatGPT. Musk was very dissatisfied with OpenAI's commercialization and its close relationship with Microsoft. In the book Elon Musk by the renowned contemporary American biographer Walter Isaacson, who followed Musk for a long time, the intricate relationship between Musk and OpenAI is detailed.

For a decade, Elon Musk has harbored concerns that artificial intelligence could one day spiral out of control, developing its own consciousness and posing a threat to humanity. Google co-founder Larry Page dismissed these worries, labeling Musk a 'speciesist' for his exclusive preference for humans over other forms of intelligence, which ultimately fractured their friendship. Musk attempted to prevent Page and Google from acquiring DeepMind, the AI pioneer founded by Demis Hassabis. After failing in this effort, Musk co-founded OpenAI in 2015 with Sam Altman as a competitive nonprofit laboratory.

At a time when OpenAI most needed follow-up funding, Musk departed. On February 20, 2018, Musk suddenly resigned from OpenAI's board, citing conflicts of interest between Tesla's autonomous driving technology development and OpenAI. The official statement at the time indicated he would continue donating to OpenAI and serving as an advisor. Musk later explained this was because both Tesla and OpenAI were competing for the same pool of technical talent, creating a conflict of interest.

However, the actual situation was more complex. Musk essentially left in anger. In early 2018, Musk believed OpenAI's R&D had fallen significantly behind Google, so he proposed taking over OpenAI and personally leading the research efforts. But this confident proposal was met with strong opposition from Altman, the technical team, and other board members.

Therefore, Musk decided to continue building a competitive AI team focused on achieving Tesla's autonomous driving goals. Despite simultaneously dealing with production crises at the Nevada and Fremont factories, he recruited deep learning and computer vision expert Andrej Karpathy from OpenAI to lead Tesla's AI projects.

OpenAI has released GPT-3.5 to the public. Their creation can chat with humans in a natural manner and perform numerous text-based intellectual tasks. GPT-3.5 has been a tremendous success, bringing Sam Altman closer to Microsoft. Elon Musk, however, believes they are recklessly testing the boundaries of uncontrolled AI, which has left him deeply concerned and dissatisfied.

In February 2023, Elon Musk invited Sam Altman to meet with him on Twitter and asked Altman to bring OpenAI's founding documents. Musk questioned him, demanding proof of how he could legally transform a donation-funded nonprofit organization into a for-profit entity capable of earning millions. Altman attempted to convince Musk that everything was done legally, insisting that he was neither a shareholder nor a profiteer. He also offered Musk shares in the new company, but Musk declined.

Surprisingly, Elon Musk launched a fierce attack on OpenAI and Altman. He said: "OpenAI was created as an open-source (which is why I named it 'Open' AI), non-profit company with the purpose of competing with Google. Now it has become a closed-source, profit-maximizing company, effectively under Microsoft's control."

I still don't understand how the non-profit organization I founded with a $100 million donation turned into a $30 billion for-profit company. If this is legal, why isn't everyone doing it?" He called artificial intelligence "the most powerful tool ever created by humans" and then lamented that it has "now fallen into the hands of ruthless monopolies."

(The above excerpt is from Elon Musk's Biography)

Musk previously signed an open letter with Bengio, Apple co-founder Steve Wozniak, Stability AI CEO, and Marcus, calling for a 6-month pause on developing AI more advanced than GPT-4. However, just a few months later, he launched his own large model, Grok. This does not indicate Musk's inconsistency; he remains a strong advocate for 'AI doomsday' theories.

Musk and Altman once discussed in detail a goal called 'AI alignment.' It became a hot topic in 2023 after OpenAI launched a chatbot, ChatGPT. The goal is to align artificial intelligence systems with human objectives and values, much like the rules Isaac Asimov set in his novels to prevent robots from harming humans. Think of HAL, the homicidal computer in 2001: A Space Odyssey, which turned against its human creators. In AI systems, what firewalls and self-destruct switches can humans implement to ensure machines act in our best interests? And who gets to decide what these critical human interests are?

Moreover, as early as 2015, Musk believed that a large number of competing AI systems could balance each other, which would be better. Just as human collective collaboration can resist human bullies, a large number of independent AI robots would also strive to prevent the actions of evil robots.

For Musk, the reason to make OpenAI truly open is to enable many people to build their own systems based on its source code. He told Wired reporter Steven Levy: 'I believe the best firewall against human misuse of AI is to have as many people as possible possess AI.'

Musk also realized that the success of an artificial intelligence system depends on its ability to obtain large amounts of real-world data for machine learning. At the time, he recognized that Tesla was such a "gold mine," collecting millions of frames of video daily from drivers handling various situations. He said, "Tesla likely possesses more real-world data than any other company in the world." Later, he came to understand that another treasure trove of real-world data is Twitter: as of 2023, Twitter processes 500 million human-generated posts daily. Therefore, Grok may not necessarily lose to ChatGPT, as data is a harder threshold to overcome and more difficult to replenish compared to computing power.

From Musk's perspective, this is a battle about protection rather than competition. The xAI official website states: "We will do our utmost to ensure artificial intelligence remains a benevolent force."

OpenAI's new products and features have left many startups trembling. Before the launch event, some joked: 'When Apple holds a product launch, many startups see opportunities emerge; every time OpenAI releases new products, a bunch of startups die.' Judging by the post-launch reactions, this seems to hold true. Moreover, this time with the GPT release, the situation appears even more dire for many startups.

As Musk previously pointed out, OpenAI is getting too close to big corporations, gradually becoming a closed-source, profit-maximizing company. For large models that are still evolving and not yet mature, this is not good news. It will only lead to tech giants prematurely building their own fences, leaving users without autonomy and controlled by large companies. Therefore, we hope that Musk's Grok and domestic large models will continue to iterate, evolve, and compete fully. AI should have many paths.