The saying goes, “The fewer the words, the bigger the story.”
Musk took to X (Twitter) to announce that Grok, an artificial intelligence assistant from his AI company xAI, will be open-sourced this week.
But Musk didn’t reveal much more than the announcement of open-sourcing Grok this week.
However, we can take a look at all the news and assume that what xAI will be announcing this week will likely be the Grok V1.5 chatbot.
As early as February 22, Musk revealed on his personal social media that he would release the new version in two weeks at the beginning of March, and this week is just in the “early March” timeframe.
According to the update rhythm of other AI companies, xAI has also reached the time to update the next generation of large models.
Since its release last November, xAI’s first product, Grok AI, has generated a lot of discussion in the industry and then quickly cooled down. Its popularity is not on the same level as ChatGPT, mainly because the so-called “sense of humor” selling point of Grok AI didn’t attract much interest.
Before that, Inflection AI, a dark horse in the VC circle, had already made a mature and out-of-the-ring “emotional chatbot” product, compared to which Grok AI’s performance was more like a “half-finished product”.
In addition, Grok AI relies on 33 billion parameters behind the core engine Grok-1 at the time to see the performance of GPT-3.5, but did not catch up with the GPT-4, and now OpenAI has been updated to the GPT-4 Turbo version, Grok-1 is even more difficult to “big job”.
Therefore, the release of a new generation of large models has been a top priority for xAI.
From what we’ve heard, the new version will bring several features, including the much-anticipated “Grok Analysis” button. Simply put, this button automatically summarizes entire conversations and responses, helping users easily understand complex conversations. In addition, Grok v1.5 will also be able to assist users in creating posts, making content creation more efficient.
However, these features are basically “standard” for large models, and open source is a good choice to attract more vendors, developers, and users in a short time.
It is worth mentioning that a major feature of the Grok model is the main “lightweight”.
In the test, the Grok-1 prototype version of Grok-0 performance can already be comparable to thethisture LLaMA 2 (70 billion parameters), but only uses half of the training resources.
The vast majority of AI applications are now developed in Python, which is gradually becoming the compiled language of choice in AI because of its rich libraries that allow for simplified program code to build neural networks, fill in parameters, import data, and call execution functions for training.
At this cost of this, however, Python has become overly bloated and difficult to speed up, which is a paradoxical “drawback” for large thistles that desperately need speed.
In a study testing the data, Rust was shown to be 50% more efficient than Java and 98% more efficient than Python in terms of energy utilization. As GPUs continue to be ethical, the need for this efficient language is even greater for large models.
However, although Rust is good, but the learning threshold is high, the development is difficult, and the ecology is not as good as C / C + + +, JAVA, and other old languages, so in the good “open source” as a selling point, may be able to attract more people to join the development of Rust.
In addition, Musk chose to announce open source at this time because there is a very obvious intention – to OpenAI an open source “to make a sample”.
Just this past weekend, Altman announced his return to the OpenAI board of directors, ending a court battle that had lasted several months. Off-site, Musk appealed OpenAI’s request for an “open source” lawsuit, which seems to have not yet reached a climax.
In a comment that “so should OpenAI, if they are as ‘Open’ as their name suggests,” Musk couldn’t help but respond: “OpenAI is a lie. “
Musk’s lawsuit against OpenAI centers around three main areas, “whether it should be profitable,” “open source to closed source,” and “doubts about the independence of the board of directors.”
On the question of whether OpenAI should be open source, Musk argues that the GPT-4 is already very close to being human, and that OpenAl is developing a secret algorithm called Q*, which is potentially powerful enough to be a general-purpose AI, and therefore must be used for the benefit of the general public.
There is no accurate information about what exactly Q* is. But what is certain is that OpenAl has put very heavy pressure on Musk’s AI layout.
In addition to OpenAI, xAI has to face the challenge of Google, Meta, and even Anthropic, Mistral AI, such a rising star, its competitive pressure can be imagined, at this time will be a passive situation into an active offense, perhaps a little more opportunity.
Overall, Musk’s pursuit of artificial intelligence road is not so smooth sailing, open source Grok is not a “gimmick”, so wait for this week’s latest news.
