Just imagine a world with inhabitants fully controlled by a benevolent AI system, instructed to do their daily jobs, eat, sleep, and talk in the way people truly would. Now it is possible to run this on a personal computer – the scientific project has been released as open-source, open for tinkering by some programming demiurges.
Generative Artificial Intelligence services adoption is not slowing down on its march to get a dominant position as a working tool for world companies. This edition of Tooploox CS and AI news brings new facts to light about the arguably most popular GenAI-based tools in the world.
Generative AI is quickly infiltrating organizations, McKinsey reports
McKinsey is one of the companies that leverages the power of generative AI in their daily work. The company is undoubtedly proof that the technology is gaining wider and wider adoption. And according to its recent report – the technology can already be considered fully adopted.
The vast majority (79%) of respondents asked by the company if they had at least some exposure to generative AI, either in work or outside it, confirmed this. The survey gives more in-depth information about the industries using generative AI technology.
The report can be found on the McKinsey website.
OpenAI adds a ‘huge set’ of ChatGPT updates, including suggested prompts and multiple file uploads.
ChatGPT was rolled out in November last year. If you missed that (probably due to living under a rock), you could read what ChatGPT is on our blog. Since then, the technology has seen a set of major updates. Yet the most recent has come with a set of improvements that enhance its notoriety.
Recently, the feature of ChatGPT plugins has also been introduced.
The system comes with suggested prompts to get a good start. Also, the company delivers suggested follow-up questions to improve responses. The users can also upload files to the chat and leave some custom instructions to help the Chat cooperate in the way they desire every time.
More can be found in the release notes.
LLM-powered life simulation made open-source.
An interesting experiment was initiated in April and took its next steps in August – the researchers have delivered a small city inhabited by LLM-powered agents who exchange information and behave the way the creator desired. The inhabitants of the city run their errands, arrange meetings, and just live life(?).
The research project has been released as an open source to make it accessible to other researchers and tinkerers from all around the world. The AI-powered social sandbox can be an inspiration for the gaming industry, both as a way to improve player experience and to build entirely new games.
The full information can be found in the Arxiv paper. Also, the working simulation can be seen on their demo page.
AI produces counterfeit books sold as written by real authors.
Jane Friedman is an author who focuses on the publishing and writing market, advising new writers on starting their careers and building their income on writing. She has discovered that both Goodreads and Amazon have been selling books on the subjects she writes about, with her listed as an author, yet they are books she has never written. According to the author herself, the content was probably delivered by an AI model, and the producer was preying on people who were seeking her knowledge.
What’s even more interesting is that she asked both Amazon and Goodreads to remove the books from their offers, yet both companies initially declined. After the author reached out to the media with the story, Goodreads removed the counterfeit title, while Amazon asked the author to provide proof of any trademarked goods being violated. When the author responded that she had no copyright for her name, the company refused to remove the fake books.
The story in detail can be found in the author’s blog post.
US Court – AI-generated content is not copyrightable
A judge of Washington court affirmed the decision to reject an application filed by computer scientist Stephen Thaler on behalf of the DABUS (Device for the Autonomous Bootstrapping of Unified Sentience) computer system.
Both the patent office and court rejected the possibility of a computer system holding the intellectual right to its creation and highlighted the fact that only human creation can be copyrighted.
More can be found in Reuter’s news.
OpenAI supports companies in fine-tuning ChatGPT Turbo.
This new feature, introduced by OpenAI, enables companies to tune the model using proprietary data owned by them. By that, the system can be adjusted to the business needs of a particular organization, for example, by gaining a better and deeper context of the job to be done and tasks to be delivered. By that, building some more awesome prompts for ChatGPT is way easier and more efficient.
According to the company, a fine-tuned GPT 3.5 Turbo can outperform the higher-version model of GPT4 in certain tasks.
More information can be found on the OpenAI website.
McKinsey introduces Lilli – an internal generative AI-powered assistant
According to a report issued in June, nearly half of the giant’s employees are using generative AI to complete their day-to-day tasks. Primarily, ChatGPT has been their tool of choice, yet the company has designed a new generative AI product – Lilli.
Lilli basically serves as a knowledge management tool, with the system providing insights, data, and ideas about what experts to hire for a particular project. According to company officials, the chatbot can be compared to asking a question to “a totality of the company knowledge.”
The system was named after Lillian Dombrowski, the first woman McKinsey hired for a professional service role back in 1945.
More information can be found in this press release.
Meta releases SeamlessM4T – an all-in-one, multilingual multimodal AI translation model
Meta, the company behind Facebook and Instagram, has released a multi-purpose model that can perform a variety of tasks, like text-to-speech, speech-to-text, speech-to-speech, and text-to-text tasks.
Before this, these tasks had to be performed by separate, narrow models trained solely to fulfill each particular task – for example, text-to-speech.
Furthermore, the model is capable of delivering translations in multiple languages. So, a user, for example, can deliver text in English and have it spoken in Spanish.
More details can be found on Meta’s blog post.