AI news of the week – 19th May: A quick summary

This past week has been about OpenAI vs Google and who is the likely leader in the world of AI. The OpenAI announcements have also been accompanied by a lot of drama with key people in the team leaving along with some news about how OpenAI treats exists within the company. Not a good look for OpenAI – but we have a very good lesson in PR from Sam Altman as well. That + some interesting updates from Meta and Based Hardware – our favourite open source hardware dev – Nik Shevchenko and a new Humaniod Robot! Let’s dive in:

OpenAI’s new GPT-4o Update

OpenAI announced their GPT-4o model, a new multimodal AI model based on the GPT4 series. The long and short of it is that this version of chatGPT can now see and hear things natively. The live demos were very interesting and this opens up the use cases for building directly atop foundational models as compared to having wrappers and pipes to fit real-world use cases. We’ve done a detailed breakdown here.

Google I/O – and a host of updates from Google on AI

Google I/O had so many angles to it that it’s hard to describe.

• The AI meme lives on
• Google is now the real life ‘Hooli’ from ‘Silicon Valley’
• Almost everything Google showcased was an announcement of a demo
• Google copying SpeedRun? (See tweet below)

https://twitter.com/tkexpress11/status/1790455457879863736?ref_src=twsrc%5Etfw

AI was clearly the central theme of I/O this year. Techcrunch did a good job of summarizing everything.

https://twitter.com/TechCrunch/status/1790504691945898300?ref_src=twsrc%5Etfw

Some key highlights:

1. Gemini gets a big upgrade. Gemini 1.5 pro now has a 2million token context window. The largest context window of any commercially available chatbot in the world. Wow, what a flex 💪.
2. Gemini 1.5 Flash – their lighter and fastest model was announced.
3. Imagen3 and Veo – Google’s Text-to-Image and Text-to-Video tools were announced. (Unofficially we’ve heard that Imagen3 is actually very good. But we haven’t been able to test it yet).
4. Google search will get step-by-step reasoning so that Google Search can do the Google Searching for you.
5. Project Astra is the future of Multimodal AI. This was Google’s response to GPT-4o. The demo was amazing, but Google has a bad history with demos.
6. Gemini will soon be integrated into all Google products. Which means Gemini will be able to do returns for you and a host of other personal tasks since it’ll have access to your e-mails and data.
7. Oh and a lot of mumbo jumbo about AI safety and privacy. 🤷 – We’re way past believing any of that. As long as the service is great, we’ve made our peace with our data being traded for convenience.

You can read a list of 100 things that were announced at Google I/O over at the Google blog.

Drama at OpenAI

Some key senior people at OpenAI resigned immediately after their spring announcement. The list includes OpenAI co-founder and Chief Data Scientist – Ilya Sutskever, Jan Leike – leader of the superalignment group, There’s a lot of speculation around why. Ranging from wild allegations that OpenAI has been sitting on AGI all the way to some reasonably believable arguments that OpenAI was selling out and the purists didn’t like that. Our bet is with the latter. Yann LeCun is in agreement.

https://twitter.com/Dan_Jeffries1/status/1791752281982996590?ref_src=twsrc%5Etfw

Of course, there was news that OpenAI treats its NDA very seriously and that some Dr. Evil level agreements were forcing shut the mouths of those who left recently from speaking against OpenAI. Of course, Sam, being the master he is, taught us a good lesson in PR. See his response below. We’re impressed but we also know that Sam is an extremely smart hustler. So who knows! 🤷

https://twitter.com/sama/status/1791936857594581428?ref_src=twsrc%5Etfw

Opensource hardware is progressing in leaps and bounds

https://twitter.com/kodjima33/status/1789745126992220568?ref_src=twsrc%5Etfw

We are already big fans of Nik Shevchenko – Founder of Based Hardware. They just won a Meta AI hackathon for ‘Open Glass AI’ – open source smart glasses. Open Glass is very similar to Meta’s Ray Ban Smart Glasses. It’s not as trendy, yet, but has great promise. For $20, it’s really value for money! Based hardware released ‘Friend’ – an open source AI wearable that outclasses the likes of Humane’s AI pin a couple of months ago. And for $35 – one can’t argue with the value proposition.

Meta’s new Camerabuds – camera integrated, AI enabled earbuds

If you think that’s a little bit strange, you wouldn’t be the only one. Much like Rabit and Human and some of the others in the space, Big Tech is now trying to find new ways and means to integrate AI (and themselves) in to our lives. There isn’t much news about it other than that they are working on it. We’re looking forward to seeing how these turn out, especially since the Meta Ray Ban Smartglasses are really cool.

ElevenLabs announced Native-Audio – have your web pages automatically be read out by AI

https://twitter.com/elevenlabsio/status/1791502626921607560?ref_src=twsrc%5Etfw

Natively enabling your web pages to be auto-read by AI is really cool. Wonderful for accessibility. Especially convenient for those who like to consume their content while doing other things. What we like most are the setup instructions. Login, whitelist your domain, choose a voice, customize your player and just copy the embed code. No fidgeting with code and trying to get it to work. This is how AI is supposed to work. You can find out more here.

There’s a new humanoid robot on the block: The G1

Unitree revealed a new humanoid robot, the G1, that looks kickass. The robot can do a host of things, include being a ninja. At USD 16k, this robot is a big step forward in accessible robotics and should usher in many new use cases. Check out the video below.

Clearly the G1 is packed with some cutting edge tech – and in a 35kg package, the use case as a personal robot assistant at home becomes more and more tangible.

That’s it for the week. Please do share this article with others who might need a catch-me-up on the AI News for the week. See you in the next one.