This edition of Tooploox AI news covers the significant legal changes in the AI landscape, with the EU adopting the AI Act and the UN issuing their AI resolution. This issue also includes new applications of generative AI in media and healthcare.
Also, March has shown that Large Language Models are surprisingly capable of handling math-related problems.
03.05.2024 Orca-Math outperforms LLMs in math-related fields
Orca-Math is a Small Language Model (SLM) aimed at solving particular, narrow problems – math operations, in this case. The model itself was developed using Mistral AI’s Mistral 7B. With its smaller size and significantly reduced power consumption, the model can be used to aid its larger counterparts when more sophisticated math operations are necessary.
Orca currently beats the vast majority of LLMs in Math-related tasks. More on the matter can be found in Microsoft’s press release.
03.13.2024 Anthropic releases Claude 3 Haiku
Anthropic released a faster and more affordable version of Claude 3, nicknamed Haiku – a direct competitor for ChatGPT4 Turbo. This type of model favors performance to be used in particular and specific use cases. The system completes the set of available models – Claude 3 Opus and Claude 3 Sonnet – providing a variety of models for different business use cases.
More information about the new model is available in the Anthropic press release.
03.13.2024 EU votes AI Act into Law
European lawmakers voted in favor of the AI Act – a law that aims to regulate the usage of AI-based systems. The law also provides regulation on which use cases need to be supervised and names use cases where AI usage is forbidden. The AI Act was proposed in 2018, and the law has been in process since then. Now, the AI Act will regulate the use of AI-based solutions in the EU starting from the end of 2024.
More information can be found in this Reuters report.
03.13.2024 Amazon aims to build product pages from links
Amazon is releasing a new, generative AI-based feature that aims to build a product page from the product link. The page consists of complete written descriptions and images. The goal of this feature is to ease the start of selling new products on Amazon when the shop exists somewhere else (for example, in an owned e-commerce store).
More about the new feature can be found in The Verge story.
03.13.2024 Google prohibits Gemini from talking about politics
The potential influence of AI-based tools on the 2024 elections is considered one of the greatest threats to governments and politics around the world. AI-based tools may produce misleading content en-masse and support the perpetuation of it. To avoid the risk of the model being abused to interfere with elections, Google decided to ban Gemini from talking about politics, limiting users from getting information about candidates, parties, or other elements related to politics.
More can be found in the Guardian’s report.
03.20.2024 The first fairly trained LLM is here
KL3M is the first Large Language Model trained using only fair and legally acquired data. The data is available in the Kelvin Legal DataPack, a commercial dataset that now contains two trillion tokens of legal, financial, and general texts. According to 273 Ventures, the company behind the model, the tool shows low toxicity scores and ranks high when it comes to generating law-related content.
More can be found in the 273 Ventures press release.
03.21.2024 AI tool predicts side effects in breast cancer treatment
The technology trialed in the UK, France, and the Netherlands, aims to predict whether patients treated with radiotherapy or surgery are likely to experience complications or problems related to it, such as chronic arm swelling. The data to train the tool was collected from 6,361 patients. The tool was accurate at 81.6% when it comes to predicting complications and 73.4% when predicting a lack of complications.
Details can be found in the Guardian.
03.21.2024 UN passes first global AI resolution
Shortly after the EU resolution, the General Assembly of the United Nations adopted the first global AI resolution. All member states agreed that the use of artificial intelligence needs to respect human rights and comply with local laws. Also, the resolution highlights the need “to govern this technology rather than let it govern us.”
More info can be found in this official statement from the UN.
03.23.2024 Financial Times introduces Ask FT chatbot
Financial Times, the broadsheet British newspaper started in 1888, embraced generative AI technology by launching the Ask FT chatbot. The system is trained on the company data and provides users with answers using natural language based on a collection of information from all available FT texts.
Apart from an answer, the system provides the user also with information about the source article used.
More can be found in this story from the Verge.