The models of generative artificial intelligence they promise to profoundly transform the way we work but also and above all the way we communicate and interact with technology. Like it or not, a revolution is already underway. Here we think that the best approach is to try to know and understand the tools that AI makes available today, always maintaining wit and critical thinking without apocalyptic excesses or uncritical enthusiasm, if only to be prepared for the world to come.
Many of the services found online today, advertised by a new crop of techno-hucksters on social media and YouTubeare often nothing more than attempts to earn easy money by repurposing GPT-3.5 and GPT-4 in a different guise and perhaps applying OpenAI models to a specific domain.
However, there are already many legitimate and useful applications of generative AI. Some services and platforms (such as Microsoft) enhance GPT-3.5 and GPT-4 with custom data and context. Others rely on completely different models to transcribe audio, edit video, or generate illustrations and images. We have selected 10 that we think are already worth using, or that it is good to know to stay informed on the developments of generative AI.
Artificial intelligence
Google’s generative AI integrated into Gmail and Docs: “It will help write texts”
by Emanuele Capone
Copilot per Office
Microsoft’s investments in OpenAI have already started to pay off. Not only with the integration of the GPT model within the Bing search engine, but also thanks to Copilot, the new function for Office 365 (for now accessible only in beta) which allows you to use artificial intelligence within the programs of the Microsoft suite. The strong point of Copilot compared to other solutions is the integration with the context accessed by the programs of the Office suite, which allows it (for example) to create powerpoint presentations starting from text documents or commands in natural language, but also to take notes, transcribe meetings, compose emails and much more.
GitHub Copilot
Speaking of the generative AI revolution, we often focus on writing texts, and therefore articles, press releases, messages, emails and so on. However, there is another writing domain in which ChatGPT excels more than any other: programming. GitHub, the software repository acquired by Microsoft, has already integrated GPT into code editors with the Copilot function, and the results are amazing: programmers can complete handwritten code in seconds, or have Copilot write functions and other pieces of code with input in natural language. There is still room for improvement, but in the hands of an experienced programmer, i.e. someone who can already conceptualize the programmatic structures he wants to achieve, Copilot is a very powerful tool, already today. It is available as an extension for the Visual Studio Code, Visual Studio or Neovim code editors and costs $10/month or $100/year. It’s free for students and programmers working on open source projects.
Speakers
As students and journalists well know, transcribe a long tape or an interview it is a long and tedious operation. Fortunately, AI has already solved it for some time, with services and solutions whose quality and reliability have increased exponentially over the past two years. There are many solutions and services, but here we particularly appreciate SpeakAI, primarily for the results we have managed to obtain, and then because the service not only carries out the transcription (also in Italian) but also allows you to analyze the text for identify themes and keywords. Transcripts are paid based on the length of the recording.
Descript
Descript combines a very faithful speech-to-text transcription with another very useful function for video and audio editing based on the transcribed text. It works like this: you enter a video or a recording into the app, wait for the AI to transcribe the speech and then you can proceed to edit the text as in a word processing app. When you remove a portion of text in the transcript, the app proceeds to delete the corresponding clip in the audio or video. It generally works very well and is a tool that can revolutionize your workflow of those who have to edit many texts and audio. In the Pro version it is also possible to generate a clone of the speaker’s voice in order to correct audio portions with the artificial voice.
Jasper AI
Jasper is an online service for writing marketing copy. Compared to ChatGPT has the advantage to be based on a model specifically trained to optimize the commercial yield of the generated text and make it more attractive to search engines. In addition to writing articles and slogans, Jasper can also translate and summarize long texts. Like all AI services, you pay over a certain threshold, but the first 10,000 words generated are free.
Murf.Ai
You need a speaker for video or audio, but you don’t want to use your voice e you don’t have the budget for a voice actor quality? Again, the AI intervenes. Again: there are many services and they are all similar to each other, but Murf.ai is among the most interesting. It allows you to choose from many different voices, suitable for different types of video and audio.
Resemble.ai
It offers functionality similar to that of Murf.Ai, but with one more possibility and that is the creating a voice clone based on a human voice. We tried it with English and the result is decidedly surprising, so much so that we fooled some friends who we played a recording made with the cloned voice. If you are wondering if it is possible to use Resemble.ai to generate deepfake audio, the answer is no: synthetic voices are created by recording at least 25 clips with randomly generated phrasesprecisely to avoid the possibility of cloning third-party voices from recordings.
Synthesia
Synthesia is a platform for video generation with AI. The service provides a series of virtual faces capable of reciting a text provided as a script. The result is very good, but it is very easy to understand that the speaker is an artificial avatar. After all, the purpose of Synthesia is not to create deepfakes, but to provide a quick solution for the making educational videos or for internal communication in companies. In short, before seeing an AI head and shoulders on the news, it will take a little longer.
MidJourney
If OpenAI is the most important player in the field of text-generative AI, MidJourney is establishing itself as a reference model in the context of image generation. Version 5 of the software has seen an extraordinary leap in generation quality. And it has also caused discussion, thanks to two collections of images (the Pope with the white duvet and the fake arrest of Donald Trump) that have gone viral. To try MidJourney you need to familiarize yourself with the Discord platform (necessary to use the service) and with English and it is mandatory to pay: due to recent controversies, the free tier has been discontinued and the paid ones remain, which start at 8 dollars a month.
IA generative
Glaze protects artists from artificial intelligence: how it works and how to use it
by Francesco Marino
Adobe Firefly
The company that develops Photoshop and the Creative Suite has been integrating AI-based features into its programs for several years now, thanks to a proprietary model called Sensei. It has recently though open the public beta of Firefly, an image generation platform that aims to compete with MidJourney and which will then be integrated into future versions of Photoshop (or at least, that’s the intention). The first results are exceptional, even if characterized by a well-defined style that easily identifies Firefly. The platform allows you to select styles, lighting and other characteristics of the image to then generate 4 different but similar results. For be able to use the Firefly beta you must sign in with an Adobe account or create one, and then request to join the waitlist. When you save an image, Adobe adds a watermark to show that it was AI-generated.