For the latest updates and exclusive contents of the industry’s best AI application, join the daily and weekly newsletter. Learn more
Less than two years after its launch, XAI has shipped what can be the most advanced AI model to date. GROK 3 matches all major benchmarks and the most advanced models in the user evaluation. Chatbot stadiumAnd the training has not been completed yet.
Since the team has not yet published a paper or a technical report, there is still no details about Grok 3. However, based on the content shared by the presentation in the presentation and the various experiments that AI experts executed in the model, we can guess how Grok 3 can affect the AI industry in the next few months.
Faster launch
As the competition between the AI laboratory increases, it is expected that the model release cycle will be shorter. In the GrOK 3 presentation, XAI founder Elon Musk said the user said, “You can notify you almost every day because you continue to improve the model.”
“DEEPSEEK and Grok’s competitive pressure, integrated into the changing political environments of domestic and international AI, will make the existing major laboratory ships faster.” Nathan LambertMechanical Allen Institute for AII. “Due to the increase in competition and the reduction of regulations, users are more likely to provide much more powerful AI for much faster timelines.”
On the other hand, this can be a good thing for the user because it can continue to access the latest models without waiting for a month -long rollout. On the other hand, it can have an unstable effect for developers who expect a consistent movement in the model. According to the user’s previous study and empirical evidence, various versions of models can react differently to the same prompt.
Companies need to develop customized evaluations and run regularly so that new updates do not stop the application.
Scaling
The recent launch of DeepSeek-R1 has weakened large-scale spending to create a large computing cluster. However, the sudden rise of XAI is a demonstration of large -scale investments made by a technology company in AI accelerator. Grok 3 was trained in a record time thanks to Memphis’s Collosus SUPERCLUSTER.
Lambert said, “We have no details, but it’s still safe to take data points for scaling. “XAI’s approach and messages were to get the largest cluster online as soon as possible. The razor description of OCCAM was helpful for scaling until there was more details, but most of Grok’s performance can come from technologies other than innocent scaling. ”
Other analysts XAI pointed out that the ability to expand computer clusters is the key to Grok 3’s success. however Musk implied There is something more than scaling at work. For more information, wait for the paper.
Open source culture
The transition to LLMS (Lange CANGE SOURCING Large Language Model) is increasing. XAI already has an open source grok 1. According to MUSK, the company’s general policy is to open all models except the latest version. Therefore, when Grok 3 is completely released, Grok 2 becomes open source. (SAM Altman also did entertaining Ideas to open some models of Openai.)
XAI will also not show the entire chain token of the Grok 3 reason to prevent the competitors from copying. Instead, it will show you the detailed outline of the model’s reasoning tracking, just as Openai performed O3-Mini. The entire bed is available only to the XAI Open SOURCES GROK 3, probably after the release of Grok 4.
Check your own atmosphere
Despite the impressive benchmark results, the reaction to Grok 3 was mixed. Former Openai and Tesla AI scientists Andrej karpathy Along with O1-PRO, he pointed out that he was inferred to the “state-of-the-art art”, but he pointed out that he was lagging behind other state-of-the-art models in creating configurable vector graphics or some work. Ethical problem exploration.
Other users pointed out Grok 3’s defects of coding ability There are many cases of Grok 3 compared to other models, Impressive coding achievements.
It is a good idea to check and study your mood based on my own experience of the main model. I do not judge a model based on one shot prompt. There is a series of tests that reflect the types of tasks achieved in the organization (see some examples here). The right approach allows you to make the most of these advanced models.
