Here is today’s top trending news from the world of technology and AI that every tech enthusiast should keep a tab on.
1)
SynthID, a new tool from DeepMind and Hugging Face, watermarks LLM-generated text.
Last week, DeepMind and Hugging Face launched SynthID Text, a groundbreaking open-source tool designed for watermarking text generated by large language models (LLMs). SynthID Text is available for free. The tool has been integrated into the Hugging Face Transformers library, making it accessible for developers to implement watermarking in their applications easily. Notably, It won’t alter the functioning of the LLM or downgrade the quality of the content output. SynthID is likely to improve the transparency and accountability of the AI generated content. Moreover, it can help in addressing the concerns related to AI generated content, particularly in combating misinformation and fake content.
2)
Midjourney’s new web editor lets you edit PC-uploaded images.
Popular AI image generated platform Midjourney has launched a new web editor that allows detailed editing of images uploaded from personal computers. This web editor offers several editing features, like adjusting the composition of images, modifying specific areas with new visuals based on text prompts or changing selected sections of an image. Users who have created at least 10 images on Midjourney’s Discord can access the web editor by visiting Midjourney.com/images. This feature comes at a time when competition in AI image generation space is intensifying.
3)
Sam Altman calls reports of GPT-5 Orion’s upcoming launch as fake
OpenAI’s CEO, Sam Altman, has responded to recent reports regarding the anticipated release of GPT-5 Orion later this year, labeling them as “fake news.” This statement comes amid speculation that the next iteration of OpenAI’s language model could be launched later this year. According to media rumours, GPT-5 Orion would bring significant advancement over its predecessor GPT-4. However, Altman’s dismissal means that OpenAI won’t be launching GPT-5 Orion anytime sooner. OpenAI was recently in news for launching new frontier models o1 preview and o1 mini.
4)
OpenAI’s Whisper has hallucination problems, researchers find.
Slew of researchers have raised concerns about hallucination issues with OpenAI’s Whisper transcription tool. Users have reported that the tool sometimes fails to transcribe audio correctly, especially when the audio begins with non-verbal sounds such as “aah” or “ooh,” or when it starts mid-sentence. Users fear that these issues can lead to significant portions of the audio being deleted from the transcription. OpenAI has so far not commented on this issue.
5)
Meta inks major AI Deal for news
Facebook’s parent company Meta has signed a multi-year agreement with news agency Reuters for providing real-time news through its AI chatbot. According to Axios, users of Meta AI chatbot in the United States will be able to access real-time news and information when prompted about news and current events. Answers and responses will cite Reuters articles and link directly to its coverage. However, both companies have not disclosed the financial details of the agreement. This is Meta’s first big AI deal for publishing current affair news. Separately, Meta also announced last week that it has signed multi-year deal with Blumhouse Productions for testing its new generative AI video model Movie Gen.