- The Singularity
- Posts
- Tesla's Humanoid Robot Evolves
The latest AI news for you and your business.
Greetings friends!
Even after all of Google’s AI announcements last week at I/O, there is plenty of news on the AI front this week. Money continues to flow into new AI startups and Apple is back in the news with their AI tech to improve accessibility for users with disabilities, and of course, there’s Tesla. Let’s dig in.
In this issue
Tesla’s humanoid robot evolves
Apple uses AI to tech improve accessibility
Hippocratic developing an LLM for healthcare
Together raises $20M for open source AI models
AI Explained
Prompt Engineering
Tools to check out!
Tesla’s humanoid robot evolves
Last year at the companies AI day, Tesla demonstrated their new Tesla Bot. A humanoid robot being built according to Elon Musk, “to solve the labor shortage”. It was a bit underwhelming, even for a prototype. Now at Investor Day, Tesla has given us more impressive update, with an improved exterior, impressive dexterity and the ability to learn it’s environment and tasks using the companies AI tech adopted from it’s self driving autos. No word on when the robots will start replacing workers at the Tesla factories.
Apple uses AI to improve accessibility
Known for their commitment to great design and easy to use software, Apple has turned their focus and their AI onto improving the experience of its users with disabilities with a host of improvements for accessibility. Improvements have been made in Text-To-Speech, including the ability take a snippets of voice recorders to mimic a users style and tone of speech. A useful feature for those that struggle with speaking or are at risk of losing their voice completely. Known for their highly secretive ways, I am sure that more will be announced at WWDC in June.
[read more]
Hippocratic developing an LLM for healthcare
It seems that healthcare has received a larger level of attention from the AI community than most other industries, and Hippocratic is the latest to emerge from stealth mode. The company, with a massive $50M USD seed round, aims to develop a “safety focused LLM for healthcare”. According to CEO Munjal Shah, “The company mission is to develop the safest artificial health general intelligence in order to dramatically improve healthcare accessibility and health outcomes.” Unlike other healthcare AI companies, the focus is not on diagnostics, but rather “is aimed at use cases like explaining benefits and billing, providing dietary advice and medication reminders, answering pre-op questions, onboarding patients and delivering “negative” test results that indicate nothing’s wrong“.
[read more]
Together raises $20M for open source AI models
Launched in June 2022, Together is building an ecosystem of open source Large Language Models (LLM) that may be used anyone. The company raised a $20M USD seed round, led by Lux Capital. Together has launched one of its first projects,RedPajama, which seeks to create a set of open source generative models, including chat models similar to ChatGPT.
[read more]
Prompt Engineering
In last weeks issue I went over what ChatGPT is and how it works. However, one of the most important pieces of that puzzle was missing. You. An LLM requires user input, or a prompt as it’s known, to be perform it’s magic. As is often the case in life, you will find success not necessarily by knowing the answer, but rather by asking the right question to ask. In the case of ChatGPT, the right prompt to provide. Even though you can phrase your prompts as a question, ChatGPT and other systems like it aren’t really answering questions in the way we would normally think of it. Instead, it is “completing your prompt”, based on what it think should come next in the series of text you provided. This is why it’s notoriously bad at basic math. It isn’t calculating numbers, it’s predicting/generating text. Which is also why you’ve seen so many new tools that will write for you (generate text), that’s what they’re designed to do.
So how do you get the most out of ChatGPT? It begins with a good prompt, and writing a good prompt depends a your goals. In general, you can use a either a Zero-shot or a Few-shot prompt.
A Zero-shot is the direct, ask a question and see what you get style. Sometimes it’s exactly what you want, other times, not so much. The example below shows a basic example of a Zero-shot prompt. What is a guitar?
ChatGPT responds with a fairly comprehensive result, but what if we’re after something more particular or we want a different style of output? That’s where the Few-Shot prompt comes in to play. With a Few-Shot prompt you provide a task or question to the model along with some context of your expected output, which is a simple form of training, something will cover more in a future edition. Let’s take a look at an example.
Here we’ve told ChatGPT exactly how we would like the output to be formatted and which details to include, and the response we received is exactly what we wanted. This is a fairly simple example, but you can make quite elaborate prompts, sometimes also called templates, which produce some really impressive results. Try it out and I think you’ll find it to be rather incredible with what you can produce!
Tools to check out!
You’ve reached the end for today, but fear not! We’ll back again with more news and tips to inspire you.
If you enjoyed this newsletter, please consider sharing it. 😄