GPTNext is coming, Alibaba’s Qwen2-VL, Cohere updates Command R

Here’s what you’ve missed in AI this week.

?? OpenAI Japan announces what’s “Next” (1 min)

We're always on the lookout for what's coming from OpenAI, and insights into their moves sometimes come from where we least expect them. Japan is the place this time, as the company's CEO in the country, Tadao Nagasaki, announced at the KDDI Summit that a new version of their artificial intelligence models, labelled "GPT Next," is likely to be launched before this year ends.

But this is not the first time we've heard about "Next," as OpenAI's CTO Kevin Scott claimed earlier this year that there would be a more powerful model by the end of 2024. Initially, we were led to believe this was "Strawberry," which we now know is in the works, but if rumours turn out to be true, "GPT Next" could be linked to "Orion," the successor to GPT-4, which will be powered by data generated from Strawberry, with increased reliability and reasoning.

?? Alibaba’s Qwen2-VL is a powerful visual model (1 min)

Alibaba Cloud has recently introduced Qwen2-VL, and this vision-language model showcases enhanced visual understanding, video comprehension, and multilingual text-image processing. Yet, what sets it apart from current offerings is that it can analyse long-duration videos, up to 20 minutes, for example, and provide answers about their content.

Qwen2-VL supports several languages, with English, Chinese, and Vietnamese among them, and its benchmark results have nothing to envy leading models like Llama 3.1 or GPT-4. For Alibaba, this could mark significant progress in the field of visual AI.

???? Cohere has improved its Command R series (2 min)

Cohere keeps improving its Command R series. The RAG-optimised large language models now have updated versions that display enhanced capabilities for multilingual applications, and structured data analysis, as well as major improvements in coding, maths, logic, and latency.

Command R and Command R+ are directed at developers and companies, and they're available through Cohere's hosted API and Amazon SageMaker.

??Although Anthropic's approach has always been directed at business users, they've decided to enhance Claude's capabilities with a new subscription plan aimed at enterprises and make it even more useful. The enterprise version allows the upload of proprietary data to the chatbot, from which it can analyse, answer queries, create graphics and pages, and overall be a customised AI assistant.

??OpenAI's former chief scientist Ilya Sutskever announced his exit from the company in May, and he shared details about its upcoming project, an AI startup called Safe Superintelligence (SSI). Now reports claim that the startup has raised $1 billion in funding, which will further nurture Sutskever's goal of developing "safe" AI systems.

Eagle: Exploring The Design Space for Multimodal LLMs with Mixture of Encoders

Apple, Nvidia in talks to join OpenAI funding round, media reports say

Generative Verifiers: Reward Modeling as Next-Token Prediction

Ask Claude: Amazon turns to Anthropic's AI for Alexa revamp

Midjourney says it’s ‘getting into hardware’

Bill Gates has a good feeling about AI