In four different tests Grok outperforms GPT-3.5, but not GPT-4
Posted: Thu Dec 26, 2024 9:59 am
Grok aims to become a “powerful research assistant for anyone” by building on the Grok-1 language model. Developed over the past four months after pre-training its predecessor Gork-0 with 33 billion parameters, Grok-1 has been put through a series of evaluations using different standard machine learning benchmarks.
According to xAI, Grok has successfully list of real mobile phone numbers database passed four tests that other AIs have previously faced. The results show that Grok-1 outperforms GPT-3.5 in all tests, although it is easily outperformed by GPT-4, OpenAI's most advanced AI model.
Grok “is only outperformed by models that were trained with significantly more data and computing resources such as GPT-4 ,” xAI notes. “This shows the rapid progress we are making in training the language model with exceptional efficiency,” the company adds.
Although Grok is an efficient AI, it is by no means immune to errors and “can still generate false and contradictory information ,” xIA warns. And such errors are even harder to tackle because Grok is fed from real-time sources. In any case, xIA will take the trouble to monitor all the responses provided by Grok to the user “by searching through different sources, checking intermediate steps and using human feedback” when necessary.
“We believe AI has immense potential to bring significant scientific and economic value to society, and we will work to develop reliable safeguards against catastrophic forms of malicious use,” xAI said. “We believe in doing everything we can to ensure that artificial intelligence remains a force for good ,” Elon Musk’s company stressed.
Grok is currently only available to a “limited number” of users in the United States , whose “feedback” will be taken into consideration by xAi to “help improve its capabilities before a more widespread launch.” To use Grok, you must reside in the United States and first sign up for a waiting list .
According to xAI, Grok has successfully list of real mobile phone numbers database passed four tests that other AIs have previously faced. The results show that Grok-1 outperforms GPT-3.5 in all tests, although it is easily outperformed by GPT-4, OpenAI's most advanced AI model.
Grok “is only outperformed by models that were trained with significantly more data and computing resources such as GPT-4 ,” xAI notes. “This shows the rapid progress we are making in training the language model with exceptional efficiency,” the company adds.
Although Grok is an efficient AI, it is by no means immune to errors and “can still generate false and contradictory information ,” xIA warns. And such errors are even harder to tackle because Grok is fed from real-time sources. In any case, xIA will take the trouble to monitor all the responses provided by Grok to the user “by searching through different sources, checking intermediate steps and using human feedback” when necessary.
“We believe AI has immense potential to bring significant scientific and economic value to society, and we will work to develop reliable safeguards against catastrophic forms of malicious use,” xAI said. “We believe in doing everything we can to ensure that artificial intelligence remains a force for good ,” Elon Musk’s company stressed.
Grok is currently only available to a “limited number” of users in the United States , whose “feedback” will be taken into consideration by xAi to “help improve its capabilities before a more widespread launch.” To use Grok, you must reside in the United States and first sign up for a waiting list .