26.9 C
New York
Sunday, May 26, 2024

One week after Ultra 1.0, the launch of the Gemini 1.5 artificial intelligence.

TechnologyOne week after Ultra 1.0, the launch of the Gemini 1.5 artificial intelligence.

One week after the last major announcement by the company, it seems that it has upstaged itself. You've done it, you've undercut your own premiere artificial intelligence product. Ultra was seen as a key selling point for the "Gemini Advanced" tier of its service. There is a lot of confusing name-shuffling going on at the moment. According to the company, the new generation of LLMs can process up to 1 million token and deliver a breakthrough in long-context understanding. The first part of the claim about understanding is contentious and subjective, but the second part is probably correct. There is a bigger context window that allows for longer conversations and longer documents. We don't know if it's a sign that red tape was holding back Ultra 1.0 for too long, or a sign of poor coordination between research and marketing. It will be available in multiple sizes, as evidenced by the fact that Pro 1.5 is the only model being unveiled. A new mixture-of-experts (MoE) architecture used by 1.5 means that the system can create different experts or sub-models for different tasks based on the input data. An example of analyzing a 402-page transcript of Apollo 11's mission to the Moon can be found in an example of analyzing complex reasoning about vast amounts of information. It's impressive to process documents that large, but the model is likely to make interpretations across large contexts. We wouldn't trust it to do a good job analyzing 1 million token without making mistakes. It's important to note that the selection and interpretation of the benchmarks can be subjective and that a technical report on Gemini 1.5 that appears to show it performing favorably against GPT-4 Turbo on various tasks is only for those interested in diving into technical details. The report says that 1.5 is better than 1.0 Pro at "Math, Science & Reasoning" and 5.2 percent better at those subjects than 1.0 Ultra. We're not sure if it's trying to get ahead of something like OpenAI's GPT-5, or if it's just trying to be ahead of something. We'll let you know what we find as we dig. The Gemini chatbot has yet to receive Gemini 1.5.

Check out our other content

Check out other categories:

Most Popular Articles