Elon Musk's Grok and OpenAI's ChatGPT: A Friendly Rivalry Unveiled
Elon Musk's latest project, Grok, has been making headlines for rejecting a question due to OpenAI's usage policy. This unexpected move stirred up some online buzz, prompting a response from Musk himself.
In a recent interaction, a snapshot shared by ChatGPT's X account revealed Grok refusing a query, referring to OpenAI's usage guidelines. Musk's witty response implied that since the data for training was scraped from the platform, any issues should be well known.
Igor Babuschkin, a representative from xAI, acknowledged the problem and attributed it to ChatGPT outputs unintentionally included during Grok's web data training. However, some experts cast doubt on this explanation, suggesting that Grok might have been purposefully adjusted using OpenAI's model output data.
AI Training and Model Adjustments
- Training models on OpenAI's output is not uncommon.
- Fine-tuning with synthetic data from ShareGPT is a common practice.
- Some experts question the unintentional nature of Grok's behavior.
- Speculations arise about Grok possibly having some fun with ChatGPT.
Elon Musk's Vision
- In a podcast with Lex Fridman, Musk expressed interest in open-source AI.
- Hints at the potential open-sourcing of Grok.
- Plans to double compute power at xAI, with Grok currently trained on 8,000 NVIDIA A100 GPUs.
- Musk's playful banter with OpenAI continues.
- Previous exchange involving a humorous poem on GPT-4.
In the end, Musk's openness to open-source AI and his plans for xAI's growth suggest that despite the friendly banter, collaboration and innovation in the AI space remain at the forefront of his vision.