Elon Musk’s xAI chatbot Grok is finally out, and one of the interesting things is that it took the team to build its prototype in less than four months, while Google took approximately two years to build the Bard, while OpenAI took several years to build ChatGPT.
Interestingly, this was executed by a small team of 16 members. These include experienced professionals, who have previously worked at Google DeepMind, OpenAI, Google Research, Microsoft Research, Tesla, and the University of Toronto.
Musk, who was one of the early investors in OpenAI, must be hoping that Grok stands up well against ChatGPT. Notably, he introduced Grok right before OpenAI’s DevDay. It seems more than just a coincidence! Musk claimed that xAI’s compute is doubling every 2 to 3 months.
Grok also has a super advantage over ChatGPT and Bard-alike because it has exclusive access to X data in real-time, besides scraping from the web. In April, Elon Musk decided to discontinue the free API access to X (formerly Twitter). This decision was made after he recognised that developers were scraping data from the platform, which they could then use to train their own LLMs.
Looks like the battle has now shifted towards the exclusivity of content and data, where Google has Youtube (only recently it changed the policy to train its AI models), and Poe has Quora, while OpenAI has not moved beyond January 2022 information from the web.
According to xAI, Grok’s prototype LLM (Grok-0) with 33 billion parameters approaches the capabilities of Llama 2 (70B) on standard LM benchmarks, using only half of its training resources. The company claims that the latest beta version, Grok-1, is even more powerful, achieving impressive scores of 63.2% on the HumanEval coding task and 73% on MMLU.
Source:xAI
In mathematical tasks evaluated under the MATH benchmark, Grok scored 23.9%, indicating its ability to handle mathematical queries. However, GPT-4 outperformed Grok in this area with a score of 42.5%.
Lastly, in the GSM8K benchmark, used to evaluate the performance of models on multi-step mathematical reasoning, Grok achieved a score of 62.9. Comparatively, GPT-4 achieved a score of 92%. Analysing the benchmark results, it’s evident that Grok trails behind GPT-4. Nevertheless, it has outperformed Llama 2 and GPT-3.5, showcasing its strengths in certain areas.
Grok Vs ChatGPT vs Bard vs Claude
Internet Access: Musk said that Grok has real time info via the X platform, which is a massive advantage over other models.
This could greatly benefit Grok since the X platform is often the primary source of information about global events. In recent months, Musk has also encouraged journalists to post news directly on the X platform.
Example of Grok vs typical GPT, where Grok has current information, but other doesn’t pic.twitter.com/hBRXmQ8KFi
— Elon Musk (@elonmusk) November 5, 2023
In comparison, ChatGPT last month enabled users to browse the net leveraging the ‘Browse with Bing’ feature in the selector under GPT-4. Bard holds an advantage with its direct internet access. Meanwhile, Claude was trained on data up until December 2022, and although it may be aware of events from early 2023, it does not have internet access.
Image Generation: As compared to ChatGPT, Grok still has a long way to go. OpenAI recently made ChatGPT the first true multimodal model with integration of Dall-E 3. On the other hand, Grok lacks image generation capabilities entirely and does not possess other senses, such as vision and audio. Similarly Bard and Claude also lack image generation capability.
Multiformat Capability: With the latest update, ChatGPT not only reads PDFs but also supports a variety of document types within the same conversation, including PDFs, images, CSVs, and more. Claude similarly enables users to upload PDFs and ask questions about the file’s content. However, as of now, Grok and Bard do not offer this capability.
Pricing: Elon Musk announced that Grok AI assistant will be included in 𝕏 Premium+ for $16/month via web. OpenAI’s ChatGPT is free to use, but ChatGPT Plus requires a subscription at $20 per month. Anthropic’s Claude offers a free version, while Claude Pro is available for $20 (US) or £18 (UK) per month. Google’s Bard is accessible to everyone for free.
X has about 528.3 million monthly active users. Meanwhile, according to the latest available data, ChatGPT currently has around 180.5 million users. With the subscription model, it is very easy for Grok to generate revenue. Additionally, xAI does not lack computational resources. Earlier this year, Musk acquired tens of thousands of GPUs. It’s only a matter of time before Grok also becomes multimodal.
The post Elon Musk’s xAI Took Just Four Months to Build Grok appeared first on Analytics India Magazine.
