Dawid Ostrowski’s Post

Strongly recommended for anyone interested in biotech, drug discovery and rare diseases!

Usually when the technology moves on, we are able to offer more! :)

As I continue to analyze the Apple Health report (https://lnkd.in/dJr6KgCE - worth reading!) I've created this chart, describing various "Kits" (APIs/SDKs) together with some examples of apps that use them. I couldn't find much information about utilization of those components, with one exception - at WWDC 2020 it's been mentioned that HealthKit is being used by 18k+ apps. Good result!

I rarely write about Apple, but I did a little analysis of the health and fitness features on Apple Watch, iPhone, and iPad, and I have to say it's a very impressive set! I took data from the Apple's Health Report - https://lnkd.in/dJr6KgCE I created this visualization by myself :) (With the help of d3.js, observablehq.com, and Gemini 1.5 Pro)

If you are based in Europe, Google AI Studio wasn't available to you at the launch of Gemini - well, be informed that it has been recently published! It's a very convenient environment to test various models with your prompts. There are bunch of useful features: - Regular (conversational) and structured prompts (table based interface) - Creating new tuned models from an existing structured prompt or by importing from Google Sheets or a CSV - Integrates with Google Drive (to share prompts, easier access to tuning or context information) - Choose between Gemini 1.0 Pro (30k), Gemini 1.5 Flash (1M), Gemini 1.5 Pro (2M) or Gemma 2 (8k) - Interesting prompt gallery (Math tutor, blog post creator, object identifier, research assistant, regexed, and many more)

Last Thursday, at Google I/O Connect Berlin, I received an unexpected assignment to staff the Google Cloud's booth. I quickly made myself familiar with the topic of multimodal search, and it's really an interesting solution, so I'm sharing more details here: - You can already deploy Multimodal semantic search with LLM intelligence thanks to Vertex AI Multimodal Embeddings - The search results are real business data from your dataset, and no artificial texts, summarization, or images will be added - you can deploy this solution to production today without having to worry about the unexpected behavior of LLMs - the search results are returned in tens of milliseconds, without having to wait for sluggish text generation and incurring much higher costs - It's all in General Availability Learn more at: https://lnkd.in/dwuSEgRR

Follow me now to be my 5000th follower! (offer valid only for new followers 😂)

Wrapping up GDE Forum 2024 I'm flying back energized and inspired by meeting so many of our fantastic Google Developer experts!

Yesterday we released the Gemma 2 open model. - Available in 9 billion (9B) and 27 billion (27B) parameter sizes - higher-performing and more efficient at inference than the first generation - significant safety advancements built in - Runs on a single NVIDIA H100 Tensor Core GPU or TPU host, significantly reducing deployment costs 27B version offers competitive alternatives to models more than twice its size, delivering the kind of performance that was only possible with proprietary models as recently as December.

Google I/O Connect Berlin! Tons of great content, and a great opportunity to reconnect with the community!