Almost exactly six months ago, Tesla TSLA CEO Elon Musk signed a letter calling for a six-month pause in the development of more powerful artificial intelligence models. In the time since, though, he has been busy developing his own AI, in July announcing the formation of xAI, a new company whose mission is to build "truth-seeking" AI.
On Saturday, Musk revealed the company's first system, a Large Language Model called Grok. The 30 billion parameter model is trained on data from X and designed to have a "rebellious streak."
The company's ultimate goal with Grok is to "assist in the pursuit of understanding."
Related: The ethics of artificial intelligence: A path toward responsible AI
Grok, Musk said, will be provided as a part of X Premium, which costs users $16 per month.
The company released the AI assistant, which remains in early beta testing, to a select group on Saturday. Musk said that the model's real-time access to Twitter data gives it a major advantage over other models.
"In some important respects, it is the best that currently exists," Musk said.
AI experts react
The new model, according to Dr. Noah Giansiracusa, seems to be little more than another example of AI hype.
"I don't see what the big deal is," he told TheStreet via direct messages on X, adding that the only real difference between Grok and other models is its "personality," something that can be achieved through the prompting and fine-tuning of any model. "Seems like a lot of hype and marketing over something with little to no real innovation."
Given credit where credit is due. https://t.co/tYOMCCTXsW
— Gary Marcus (@GaryMarcus) November 5, 2023
And while he's not overly concerned about the impact of potential hallucinations, saying that its X training data likely makes up only a small portion of Grok's total training set, he said that the rhetoric around it could be dangerous.
"By marketing it as having more up-to-date information than other chatbots I suspect it could lead users to believe the info it has on current events is more reliable than it actually is," Giansiracusa said.
Related: Elon Musk reveals his greatest fear about artificial intelligence, and Joe Rogan agrees
Further, the efficacy of AI systems is dependent upon data quality, according to Dr. Srinivas Mukkamala, Ivanti’s CPO and AI authority leader.
Grok's Twitter training "will likely perpetuate the biases and misinformation already present on the platform – not to mention privacy issues this creates by using this data for training," he told TheStreet.
Although it is difficult for any AI company to obtain objectively accurate data, Mukkamala said, specifically designing Grok to be both rebellious and humorous is antithetical to achieving objective data.
"While xAI acknowledges that their model can still generate false or contradictory information, which should set off alarm bells, far more alarming is that we can’t anticipate all the ways in which the live data can be manipulated, biased or be used to spread misinformation," he said.
For Dr. John Licato, a professor of computer science and engineering, Grok's release raises a series of questions that remain unanswered by xAI.
"In this era of hype and false transparency, we need to be especially careful," he told TheStreet. "How do they plan on using our interaction data? Will they update their models with the information we include when chatting with it? And will they tell us exactly which datasets were used to train their model?"
Competition in the AI space, he said, is not a bad thing. But Grok doesn't seem all too different from the competition, and its sarcastic personality represents yet another red flag.
"The idea that it will be 'sarcastic' looks like a gimmick," Licato said.
"Will they achieve this sarcasm by removing some of the guardrails that current LLMs have to reduce the possibility of generating harmful content?"
Musk's AI safety plan
Musk has previously explained that his approach to AI safety revolves around building a curious, truth-seeking AI, a classification that experts have taken issue with as AI models, which are not sentient, cannot be curious, nor can they possess any other human traits or attributes.
"I would love to see more transparency and testing when it comes to throwing around terms like 'witty' and 'rebellious,'" Dr. Sasha Luccioni, an expert who studies the intersection of AI and ethics, told TheStreet.
"Does this come with trade-offs in terms of abusive language and toxicity? We also need more transparency about the data used and how it was filtered (or not) — we all know there's a lot of toxic and harmful stuff on Twitter."
Musk has also often spouted extinction-level fears around out-of-control AI, which experts have said is not supported by the science.
Grok is not yet multimodal, though the company said in a statement that it will eventually be able to process visual and auditory information.
In the wake of the launch, Musk — who helped start OpenAI in 2015 before leaving the board — again criticized the AI firm behind ChatGPT for its quick evolution from a nonprofit to a for-profit company with a $90 billion valuation.
This marks Musk's first push into the "generative AI" space, where Big Tech giants like Google GOOGL and Microsoft MSFT currently reign supreme.
"It's the rhetoric and the dressing around (AI), that carries these tools into places where they don't yet have a justification for being used, but are still being used," Suresh Venkatasubramanian, an AI researcher who in 2021 served as a White House tech advisor, told TheStreet in September.
"And that's where we see the problems."
Related: Experts Explain the Issues With Elon Musk's AI Safety Plan
Get investment guidance from trusted portfolio managers without the management fees. Sign up for Action Alerts PLUS now.
