Anthropic researchers show how a series of fabricated dialogues can manipulate AI responses
The researchers published a paper that details a vulnerability in the current generation of large language models that can be used to trick an AI model into providing responses it’s programmed to avoid, such as those that could be harmful or unethical. The vulnerability is a new one, ensuing from the improved “context window” of the latest era of LLMs.
Source
The US and UK have signed an agreement focusing on the safety of AI technologies
Under the deal, the UK’s new AI Safety Institute and the upcoming US organisation will exchange research expertise with the aim of mitigating AI risks. In addition, the two countries agreed to forge similar partnerships with other countries to foster AI safety around the world.
Source
Yahoo acquired Artifact, an AI news app made by Instagram co-founders
Artifact, which was launched last year, offered users to link their contacts and see what others are reading, among various other features. It also used AI to summarize long news stories in a few bullet points.
Source
Microsoft upgraded Copilot for 365 with GPT-4 Turbo
With the new AI model, users will receive faster and more comprehensive responses, whether working within Microsoft 365 applications or using Copilot for web searches. Also, Microsoft removed limitations on conversation length and frequency.
Source
New products
iQOO Neo 9S Pro
iQOO Neo 9S Pro will have an OLED display, Dimensity 9300 processor and 12 GB of RAM. It will come with a 50-megapixel primary camera.
Source
Vivo V30 SE
The phone will have a Snapdragon 4 Gen 2 processor, 8GB of RAM and Adreno 615 GPU.
Source
