A week in Generative AI: Translation, Inpainting and testing consciousness
A week in Generative AI: Translation, Inpainting and testing consciousness
News for the week ending 27th August 2023
Another week of big news and some really interesting long reads. I’d say that generative AI has now solved ‘translation’ with the release of Meta’s new language AI and MidJourney’s new inpainting feature brings advanced photoshop capabilities to anyone who can use a mouse and type…
If AI becomes conscious, how will we know?
Nice little summary of a discussion paper that proposes 14 criteria to suggest (but not prove) an AI is conscious.
Meta unveils breakthrough language AI
This is the first all-in-one multilingual multimodal AI translation and transcription model that can perform speech-to-text, speech-to-speech, text-to-speech, and text-to-text translations for up to 100 languages.
OpenAI brings fine-tuning to GPT-3.5 Turbo
This is a great announcement for businesses looking to fine-tune GPT to add domain-specific knowledge and specific behaviours. The fine-tuning will also allow for better guardrails to be built into the base model.
Midjourney released a new feature: Inpainting
This is a really cool feature - this example shows how a Studio Ghibli image can be amended into a Harry Potter theme. The speed at which these new updates and capabilities are coming is mind blowing!
ChatGPT custom instructions are now live in the EU and UK
I haven’t had a chance to play around with this too much yet, but glad to see this feature come to the EU and UK about a month after it was released.
This week’s long reads
The Atlantic - Revealed: The Authors Whose Pirated Books Are Powering Generative AI
Napkin Math - The Internet Is Still Early
Ars Technica - How ChatGPT turned generative AI into an “anything tool”
“The future is already here, it’s just not evenly distributed.“
William Gibson