ByteDance is reportedly secretly developing multiple AI products.
Meta is expected to unveil a new model, Llama 3, in July.
Tim Cook states that Apple will disclose its plans for generative AI later this year.
Apple shareholders reject proposal for AI transparency report.
Microsoft researcher publishes the first "reverse engineering" report on the Sora model.
Microsoft proposes a variant of the 1-bit large language model.
OpenAI and Microsoft are sued for piracy.
Adobe releases AIGC music creation tool.
Hugging Face launches the StarCoder 2 series of models.
Lightrick announces the AIGC movie production platform LTX Studio.
AI movie production platform Morph is officially launched.
AI image generation startup Idegram raises $80 million in Series A funding.
Amazon focuses on investing in AIGC startups in areas such as robotics.
AI customer service efficiency significantly surpasses human, causing related company stocks to plummet.
Paper Link: 1-Bit Large Language Model Variants
GitHub Repository: SoraReview
Microsoft Introduces 1-Bit Large Language Model Variant
On February 27th, Microsoft proposed a 1-bit large language model, presenting a variant called BitNet b1.58. The research introduces this variant, demonstrating its comparable performance in perplexity and final task performance to models with the same size and training tokens, while significantly improving cost-effectiveness in terms of latency, memory usage, throughput, and energy consumption. Furthermore, the 1.58-bit large language model defines a new scaling law and a novel approach to training next-generation large models that are both high-performing and cost-effective. Additionally, it enables a new computing paradigm and offers new insights for designing specific hardware tailored for 1-bit large language models.
Yesterday, news outlets Raw Story, AlterNet, and The Intercept filed a lawsuit against OpenAI and Microsoft, alleging copyright infringement, including the removal of authors, titles, and other copyright information when training AI large models.
On February 28th, Adobe unveiled its latest generative AI tool, Project Music GenAI Control, for creating and editing custom audio. With this tool, users can generate music from text prompts and perform fine-grained audio editing. Its creator, Nicholas Bryan, likened it to the "Photoshop for audio."
Hugging Face, in collaboration with cloud computing company ServiceNow, released the open-source code generation model series StarCoder 2 on the 28th. It includes a 3-billion-parameter model trained by ServiceNow, a 7-billion-parameter model trained by Hugging Face, and a 15-billion-parameter model trained by NVIDIA. The first two models can run on most consumer-grade GPUs.
Today, Lightrick announced the first-generation generative AI movie production platform, LTX Studio. Users can leverage AI to generate short films and meticulously edit scripts, characters, scenes, styles, dialogues, plots, and more. The platform will be officially launched on March 27th and is now open for reservations.
Reservation Link: https://ltx.studio/
Today, AI startup Morph Studio, in collaboration with Stability AI, launched the AI movie production platform Morph. Based on a text-to-video model, users can create and edit scenes by inputting textual prompts of different scenarios and combining them into a coherent story.
According to Bloomberg, Canadian AI startup Idegram announced its Series A funding plan on Wednesday, raising $80 million. Founded less than two years ago, the company focuses on developing AI image generation technology.
According to the Financial Times, Franziska Bossart, head of Amazon's corporate venture capital division, stated on Wednesday that generative AI has significant potential applications in robotics and automation. The company will prioritize investments in this area this year. Amazon's $1 billion Industrial Innovation Fund will increase its investment in startups in this field.
Swedish fintech company Klarna announced on the 27th its AI assistant application powered by OpenAI technology. The assistant engaged in 2.3 million conversations within a month, accounting for two-thirds of the total customer service conversations, equivalent to the workload of 700 full-time employees. Customer satisfaction is comparable to humans, with the average problem-solving time reduced from 11 minutes to 2 minutes. Bloomberg attributes the industry's concern about AI replacing human customer service to this statement, leading to a significant drop in the company's stock price on Wednesday.
