🎨 IMAGE AI BREAKTHROUGH • DIGITAL NARRATORS • ARTIFICIAL DOCTORS • FAKE DETECTORS | v18
After a little break, The Input is back for 2023! This year will be an incredibly exciting one for AI, as it continues to touch everything around us in ever more powerful ways. From generative visual models to large language models and beyond – I can’t wait to share the latest inspiring progress with you!
🎨 ‘AI THAT MAKES IMAGES’ IS ONE OF THE 10 BREAKTHROUGH TECHNOLOGIES 2023
(→ Technology Review)
As every year MIT Technology Review crowned its 10 Breakthrough Technologies this month. Among them are Diffusion Models, the technology that allows us to generate anything visual from simple text input.
The progress over the last months has been simply astonishing! CLIP, the technique that allowed text-based image generation in the first place, was released just 2 years ago – now we can render high-res images in real-time. So, it is hard to imagine where these models will be in a couple of months. Watch this space!
I’d highly recommend this conversation between Peter Diamandis and Emad Mostaque, the founder of Stability.ai, the company behind Stable Diffusion, if you want to get inspired by how AI will change our lives in the next couple of years.
Also, check out this experiment we did with generative models last year.
Trying to generate Lost Masterpieces with AI.
📚 APPLE INTRODUCES AI GENERATED VOICE-OVERS FOR BOOKS
(→ Apple)
Apple launched what it called digital narration as part of its Books App.
It’s going to be interesting to monitor the reactions to this move, that obviously has the potential to replace human voiceover artists, which is likely to cause a big backlash. Apple on the other hand argues that it is empowering indie writers, who wouldn’t have the means to hire a professional narrator otherwise.
I can also see tech like this become a tool for narrators to improve and scale their work.
🥼 A LANGUAGE MODEL FOR MEDICINE
(→ Interesting Engineering)
Google and Deepmind continue their push into healthcare, arguably one of the sectors with the biggest potential for improvements through AI.
MedPaLM is a Large Language Model, like OpenAI’s GPT, specialized in medical writing, especially question-answer pairs.
Even though the model is still yet to fully outperform human clinicians (it is currently on par with them), an interesting result is that the mistakes the model makes are deemed less dangerous that the ones made by humans.
🔎 GPTZERO: TRYING TO DETECT AI-GENERATED TEXT
(→ BBC)
Is this text written by AI? If you have had the chance to play with or read some of the things ChatGPT generated over the last month you’ll definitely have been thinking about this reading any other text online.
This is a crucial problem to solve as language models become better and better.
Edward Tian, a 22-year-old student at Princeton University is working on an approach to detect AI-generated text. Put simply his system GPTZero is using a measure of how much another AI is surprised by a given text.
This could lead to quite an interesting dynamic of humans trying to write less human-sounding text, as models like ChatGPT are becoming borderline perfect at generating human-sounding text and humans are judged by hoy much they surprise the model.
In any case, an adventurous period for human creativity lies ahead.
The Input is a newsletter made with 🖤 by Nice Outside on planet earth. If you have feedback, are interested in geeking out about any of the things mentioned above or just want to jam on an idea, feel free to reach out to max@no.studio.