Weekly review #5
Miko Pawlikowski ???
Follow for breakthrough ideas in tech, business & performance | Keynote speaker | Co-founder SREday.com
Here’s what you’ve missed in AI this week.
?? OpenAI is working on two new AI models (2 min)
The buzz surrounding OpenAI and its secret projects hasn't stopped for months, and every hint we've gotten only leads to more hype. But, according to insiders, fall may be the time when we're finally able to witness how these models will disrupt AI. We're talking about the upcoming and long-awaited "Strawberry" and "Orion," a flagship language model.
Reports claim that Strawberry has been designed to excel at solving maths and programming problems, while Orion is set to outperform GPT-4 and will be trained with high-quality synthetic data, most of which is generated by Strawberry.
However, it's likely that the full version of Strawberry won't arrive this year.
?? Meta unveils “Sapiens” (1 min)
Meta's latest push into AI involves "Sapiens," a family of AI models developed for human image analysis. It comes in four sizes, with the largest being Sapiens-2B, which was trained on 1024 x 1024 pixel images. Overall, the models are able to perform tasks like body segmentation and 2D pose estimation.
The aim of Sapiens is similar to that of Meta's Segment Anything Model. Both tools are progressing human-centric vision models while also promoting innovation in the AI community, as Sapiens is currently available on GitHub.
?? There are new versions of Gemini 1.5 (3 min)
Gemini 1.5 arrived back in May, and although it's difficult to keep track of Google's moves as they're moving faster than anyone in the field, their latest announcement involves three new variants of this model. They've presented a small Gemini 1.5 Flash-8B, an improved version of Gemini 1.5 Flash, and an even more powerful Gemini 1.5 Pro.
领英推荐
Reportedly, the 1.5 Flash and 1.5 Pro showcase enhanced results in maths, coding, and complex prompts, with one of Google's AI Studio product leads claiming that the model (1.5 Flash) is the best in the world for developers at the moment. Maybe it's worth a try!
?? Google doesn't stop with the Gemini enhancements, as they've also announced the rollout of "Gems." Through this feature, users can create customised AI experts on any topic they want. On the other hand, they're broadening the availability of Imagen 3. Both features are coming to the Gemini Advanced, Business, and Enterprise tiers.
???? With the help of AI, researchers from the University of Edinburgh and the University of Dundee are analysing over a million brain scans in order to develop a tool to help doctors determine the risk of dementia in patients and hopefully lead to better diagnoses and treatments earlier.