AI Blog Google tries to deliver

Google tries to deliver

AI Blog

Issue #25

20/09/2023

Google tries to deliver

Google announced Bard Extensions, which allow you to leverage information from its app ecosystem (Gmail, Docs, Drive, Google Maps, YouTube, and Google Flights & hotels).

Why is this important?

We’ve long said that the huge value for productivity will come when AI can easily and directly leverage our own data. However, it seems that to a large extent Bard can’t seem to deliver exactly what it has promised. Once he instantiates the right file he does a very decent job at summarizing, but it’s a matter of being able to find the file.

However, we find the double-checking with Google Search feature excellent, as Bard highlights the parts of the answer that could be wrong!

We hope that Gemini (the new AI model that Google is partnering with) can solve a lot of these problems. The official announcement/demo:

Resources

AI Insider 📰

The new trend in AI generated images is to create illusions with spiral and checkerboard shapes using Stable Diffusion. This is an offshoot of the QR codes we have seen created with Control Net. To make them you need the Monster Labs QR Control Net, a prompt and a distinct pattern – combining these elements with your own pattern, it takes advantage of a QR code-based control net to produce exciting visual intrusions or effects. You can experiment here and here you will find the complete workflow.
Microsoft made a huge faux pas – its AI research team leaked 38 terabytes of sensitive personal data to its GitHub repo. Well, that went well.
OpenAI is racing to catch up to launch multimodal AI model (i.e. chatbot that can handle text, images, audio and video) before Google. OpenAI had promised that GPT-4 would be like this, but only Be My Eyes was given the full capabilities of the model.
- By the way, guess what happened now that the new academic year has begun? The use of ChatGPT has increased. In June it had dropped significantly, leading to discussions about whether it was just a trend.
40% of the largest early-stage VC investments for Healthcare IT were unexpectedly in AI.

Learning Bytes 🧐

Dictionary.com has added several new terms related to AI. For example:
- generative AI – noun. Computers. artificial intelligence that is designed to process prompts from users and respond with text, images, audio, or other output that is modeled on a training data set.
- chatbot – noun. Α computer program designed to respond with conversational or informational replies to verbal or written messages from users.
- GPT – abbreviation. Computers, Digital Technology. generative pre-trained transformer: a type of machine learning algorithm that uses deep learning and a large database of training text in order to generate new text in response to a user’s prompt.
- hallucinate – verb. Computers, Digital Technology. (of a machine learning program) to produce false information contrary to the intent of the user and present it as if true and factual.
- LLM – abbreviation. Computers. large language model: a type of machine learning algorithm trained on extremely large data sets of existing language and designed to generate new, naturalistic responses to prompts.
Papers we found interesting:
- Cognitive Architectures for Language Agents and its GitHub repo
- Harvard AI study with a field experiment for the influence AI has on knowledge workers
Research from Emory University School of Medicine found that ChatGPT is doing extremely well in diagnosing eye-related diseases.

Cool Finds 🤯

A church in Texas used ChatGPT for its Sunday service. As they stated, this effort is aimed at “understanding and exploring the possibilities of AI without fear, but with an inquiring mind.”