1. Meta releases two multimodal Large Language Models (LLMs)
On November 16, Meta released Emu Video, a text-based video model, and Emu Edit, an image editing model, on its website. According to the introduction, Emu Video can generate high-quality videos through natural language, and it divides the video generation into two steps, first generating an image conditional on text, and then generating a video conditional on both text and the generated image. Emu Edit is a multitasking image editing model that can edit images through natural language. In training, the team formulated all region editing, free-form editing, and computer vision tasks as generation tasks and provided learning task embeddings.
Paper address:
emu-video.metademolab.com/assets/emu_video.pdf
emu-edit.metademolab.com/assets/emu_edit.pdf
Project homepage address:
2. Cruise Continues Autonomous Vehicle Tests Abroad Amid US Safety Review
General Motors’ self-driving car unit, Cruise, is continuing on-road testing in Dubai and Japan, even as it halts all vehicle operations in the United States for a full safety review, according to information obtained by Reuters.
The company has expanded the scope of its internal investigation to ensure the utmost safety of its autonomous vehicles.
Earlier this month, Cruise made the decision to suspend all fully autonomous rides and recalled approximately 950 vehicles. Despite the ongoing safety review in the U.S., the company’s self-driving cars are still on the road in Dubai and Japan, underscoring its commitment to perfecting autonomous driving technology.
Stay tuned for more updates on this developing story about Cruise’s ongoing self-driving car efforts.
3. Google Releases New Document on ‘An AI Opportunity Agenda’
On November 14, local time, Google released An AI Opportunity Agenda, which provides specific recommendations for governments on investing in AI infrastructure, building human capital and an AI-ready workforce, and maximizing AI accessibility and adoption.
“An AI Opportunity Agenda” document address:
https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/AI_Opportunity_Agenda.pdf
4. NVIDIA Announces Next-Generation H200 AI Chip
NVIDIA unveiled its next-generation AI chip, the H200 Tensor Core GPU, the first GPU with HBM3e, which delivers more memory and speed to accelerate generative AI, large language models (LLMs), and scientific computing for high-performance computing (HPC) workloads.
Systems with the H200 from server manufacturers and cloud service providers worldwide are expected to begin shipping in the second quarter of 2024.
Early cloud service providers using the H200 include CoreWeave, Lambda, Vultr, AWS, Google Cloud, Microsoft Azure, and Oracle.
5. ChatGPT Plus Pauses New Subscriptions
On December 15, OpenAI CEO Sam Altman announced on the social platform X that new ChatGPT Plus subscriptions would be suspended because the surge in usage after the developer conference was more than OpenAI could handle and the company wanted to make sure that every user had a good experience.
Users can still sign up within the app to be notified when Plus subscriptions resume.
6. ChatGPT Web Version Bug: Change The URL To Upgrade GPT-4
According to today’s report by IT Home, a number of social platform X users recently found that the ChatGPT web version has a major loophole, just modify the URL, and you can upgrade GPT-3.5 to GPT-4.
It is reported that users only need to enter the ChatGPT official chat page and add “/?model=gpt-4-gizmo” after the URL to use the GPT-4, or there is a serious “temp” level bug in the authentication system. The bug has now been largely fixed.
