Adobe have released Firefly V2 some time ago. It’s very good, random sample from the net below. All my previous criticisms are void now. Two things I have noticed, though: the NLU (natural language understanding) capabilities of the input text prompt is not as strong as with OpenAI DALL-E 3. I may(!) have seen slight blurriness introduced ... Read more 16 Dec 2023 - 1 minute read
Mistral.AI 🇫🇷 have launched their Platform, which gives API access to their “best models” on a pay-per-use basis. Currently available on an invite-only, early-access basis. Background: Mistral have been in the news for making publicly available the weights etc. of rather small Text Generation models that show surprising performance in standard ... Read more 14 Dec 2023 - 1 minute read
Google Gemini got launched (“our most capable AI model”). Upfront summary: „Overall, you should probably keep using GPT-4/Bing for any complex tasks - it is still the best AI available“ (Ethan Mollick) Literature: Google Blog, Bard Blog, DeepMind Landing Page, Gemini Technical Report AlphaCode 2 Tech Report Available through Bard ... Read more 06 Dec 2023 - 2 minute read
Overview of Model deployment options on AWS, which also takes into account cost implications: LinkedIn. It’s important to note that the comparison may not be entirely straightforward, as the slides alone do not provide a full picture. However, they do reveal some significant cost variances that may need to be considered. In particular, AWS Lamb... Read more 29 Nov 2023 - less than 1 minute read
Amazon have launched a preview of Titan Image Generator - three screencaps from the AI/ML keynote attached below. This is very similar to the DALL-E that is available through labs.openai.com (DALL-E 2?): only very limited resolutions available, both for Generation and Editing (screenshot #4). result is always PNG Unlike DALL-E: I hav... Read more 29 Nov 2023 - 1 minute read