SpaceXAI and Tesla head Elon Musk believes that AI is not good enough for the most advanced computer programming tasks. Musk shared his thoughts in a tweet in response to his earlier remarks about his firm’s Grok AI relying on reinforcement learning after finishing its pretraining. The latest Grok iteration, Grok 4.8, relies on 2.5 trillion parameters, which marks a significant bump over the previous Grok models.
Elon Musk Says SpaceXAI’s C/C++ Training Stack Is Written By Humans Since AI Is Not Good Enough
September has been a good month for SpaceXAI after the firm released its Grok 4.6 model and claimed that it offered similar performance to Anthropic’s Fable 5 products. The models came into scrutiny earlier this year after the US government decided to restrict their access to non-nationals. The restrictions went as far as to cover those present inside the US and were lifted in late June.
As per the firm, Grok 4.6 achieved a score of 61 on the Artificial Analyst Intelligence Index (AAI), which ranked it higher than OpenAI’s GPT 5.6 Sol and just a point short of Fable 5. The model also came with a significantly lower price tag as it charged $2 per one million input tokens and $6 per million output tokens. In comparison, Anthropic charges $10 per million input tokens and $50 per million output tokens for the Fable 5.
Not the one to hold back, when a user on X wondered what was next for SpaceXAI in September, Musk chimed in and remarked that his firm’s Grok 4.8 model, which relies on 2.5 trillion parameters, would finish training and enter reinforcement learning this week. He added that the model was trained on SpaceXAI’s C++ software stack.
When a Google AI worker remarked that SpaceXAI’s software stack was likely trained on AI generated by other models such as GPT and Fable, Musk remarked that the C and C++ programming language stack was written by humans.
“Our C/C++ training stack was written by humans. AI is not yet good enough for extremely high-performance software (but it will be).”
However, he didn’t respond when asked whether the training stack relied on using coding agents used by humans.
In another tweet, the SpaceXAI head outlined that the Grok 4.7 model should be on par with Anthropic’s Opus 5.0 model and not the 5.1. The 4.8 model will be a “noticeable improvement,” according to him, with Grok 4.9 “probably” matching the Astra and Fable class of AI models.
About the author: Ramish is a seasoned technology writer and editor with more than a decade of experience. He specializes in semiconductor fabrication and market analysis.
With a background in finance and supply chain management -management from MIT – Ramish combines financial rigor with deep industry insight to deliver accurate and authoritative coverage
Follow Wccftech on Google to get more of our news coverage in your feeds.
