The big happenings in the world of AI this week were dominated by Google’s I/O conference. It’s a sign of the changing times that a Google conference would announce a lot of AI services. I think the launch of ChatGPT has certainly shaken up the landscape and forced companies like Google to accelerate their plans. Here are some that caught my eye:

Google announces its next-generation Large Language Model, PaLM 2. Sadly I’m old enough to remember the Palm devices, so it does seem a little retro to me. It os said to be a new general-purpose large language model that can generate a wide variety of human-like text in response to human input. The model is skilled at math, software development, language translation reasoning and natural language generation.

Google is bringing Generative A.I. directly to Search.

Not surprisingly and widely predicted, Google is bringing Generative AI features to search called “Search Generative Experience,” or SGE. It’ll be able to synthesize search results from complex queries.

It’ll be available to Labs users over the “coming weeks,” and users can access it by tapping on the Labs icon in the Google app or on the desktop version of Google Chrome. However, it’s not clear when it will be available to everyone.

Google removes waiting list for Bard chatbot.

Google is upgrading its Bard artificial intelligence chatbot, making it more widely available and giving it a variety of new capabilities.

In March, Google started letting users in the U.S. and the U.K. test Bard, a rival to ChatGPT, the viral chatbot that Microsoft-backed startup OpenAI released in November. Now Google is opening access to the broader public, releasing it in over 180 countries and territories.

Wendy’s will open a new AI-powered drive-thru in Columbus, Ohio, in June. You knew it would happen and I’ve also heard of some automated McDonalds restaurants. Wendy’s FreshAI will be powered by Google’s AI chatbot service. The chatbot will use voice recognition to take orders from customers. A human employee will monitor the service to ensure that it is working correctly. Customers will have the option to speak with a human if needed. Wendy’s is not planning to replace existing workers with the technology.

Google launches a GitHub Copilot competitor. Google has launched an integrated chat tool for coders. Codey was designed specifically to handle queries related to coding and also to Google Cloud general queries. It will be available for Google’s cloud-hosted Workstations service, Visual Studio Code, the Google Shell Editor, and JetBrains’ IDEs. Codey is currently only available to a few trusted testers.

Developers increasingly get advice from AI chatbots and GitHub CoPilot rather than Stack Overflow message boards. It seems these chat tools for coders are pretty effective as they are driving down traffic to the traditional resource for answers to all questions on coding.Interesting Projects

Interesting Projects

Here’s a couple I came across this week.

TinyML Made Easy: Image Classification. I love this project from Marcelo of MJRobot and documented on Hackster where he takes the incredibly low-priced Seeed Xiao ESP32S3 Sense, which for $13.99 includes a 240MHz processor, camera & microphone, among other things and walks you through an image classifier for bananas, apples & potatoes using TinyML.

TinyML: An Always-On Audio Classifier using Synthetic Data. This is another Audio project using Synthetic data to train an Audio monitor based on the Arduino Nicla Voice. It’s very similar to the one I shared recently for Baby Cry detection and uses a similar workflow and tools. I’m intrigued by the whole concept of using synthetic data to train the AI.

‘Godfather of AI’ quits Google with regrets and fears about his life’s work. Geoffrey Hinton, known as one of the ‘Godfathers of AI’, has resigned from his position at Google to express his concerns about the potential risks of AI technology. Hinton’s work has directly contributed to developing technologies such as ChatGPT and Google Bard. He warns that the competition between new AI technologies will be impossible to stop, leading to a world filled with so much misinformation that distinguishing truth from falsehood will become impossible. Moreover, the technology could potentially eliminate jobs and even pose a threat to humanity itself as it becomes capable of writing and running its own code.

Hinton’s departure from Google is significant, given his stature in the AI community. He is often referred to as the “Godfather” of AI. In his resignation statement, he expressed his desire to speak openly about the dangers of AI without worrying about the impact on his employer. He even admitted to regretting his life’s work, believing bad actors would inevitably exploit the technology. With other concerned critics like Yudkowsky and Gary Marcus, Hinton’s resignation is a significant development in the ongoing debate about the potential risks and benefits of AI.

Scientists develop A.I. system focused on turning peoples’ thoughts into text. Semantic decoders are AI systems that can translate brain activity into text and potentially benefit patients who have lost their ability to communicate physically. Researchers at the University of Texas have developed a non-invasive semantic decoder system that uses a transformer model. It can produce text that closely or precisely matches the intended meaning of words around half the time without requiring any surgical implants. However, it still cannot be used outside of a lab setting because it requires an fMRI scanner. Nevertheless, the researchers believe they could eventually create a more portable system.

Amazon is developing an improved LLM to power Alexa. Amazon’s Alexa might finally be getting the upgrade many have been waiting for since the rise of language models last year. CEO Andy Jassy leads the charge in building the world’s best personal assistant, with Alexa being the natural starting point. Although it can sometimes be frustrating to use, incorporating powerful state-of-the-art language models could be the solution. Amazon is currently developing an improved language model (LLM) to power Alexa, which will be capable of understanding more complex and nuanced language, making it easier for users to interact with the voice assistant. The new model will also support more languages, making Alexa more accessible globally. Additionally, Amazon is working on enhancing the privacy and security of Alexa to instil greater confidence among users when using the voice assistant in their homes.

Interesting Projects

Nvidia Jetson Iron Man HUD. Engineer and avid cosplayer Kris Kersey, has created a superhero helmet with the help of NVIDIA technology. The helmet, inspired by the superhero Iron Man, is equipped with two cameras — one by each eye slot — that see what the helmet’s wearer is seeing. The HUD then presents information, including the current temperature, humidity, altitude and GPS location. It can also classify what’s in the user’s view based on deep neural networks for object detection.

Nvidia Jetson Iron Man HUD. Engineer and avid cosplayer Kris Kersey, has created a superhero helmet with the help of NVIDIA technology. The helmet, inspired by the superhero Iron Man, is equipped with two cameras — one by each eye slot — that see what the helmet’s wearer is seeing. The HUD then presents information, including the current temperature, humidity, altitude and GPS location. It can also classify what’s in the user’s view based on deep neural networks for object detection.

Hardware

M5Stack Launches the Compact, Camera-Packing CoreS3 with the IoT and TinyML in mind. M5Stack has launched a new compact camera, the CoresS3, designed with IoT and TinyML in mind. At its heart is the Espressif ESP32-S3, featuring a dual-core Tensilica Xtensa LX7 processor running at up to 240MHz with 512kB of static RAM (SRAM), 384kB of on-chip flash, Wi-Fi, and Bluetooth 5.0 Low Energy (BLE) connectivity and vector acceleration specifically designed for tinyML workloads at the edge.

UK to invest in AI by creating a taskforce. The UK is trying to get on the AI bandwagon and not be left behind. It has pledged £100 million to establish a Foundation Model Taskforce. The team will develop AI that ideally makes the country “globally competitive”. Call me a cynic (soapbox alert), but the UK needs to change a few things to become competitive again – one being to repeal IR35, which hamstrings hiring the best talent.

Apple is to introduce an AI-based Health Coaching Service in 2024. Codenamed Quartz, the AI-based service, will help encourage users to exercise, improve their eating habits, and take steps to improve their sleep.

Hugging Face releases it’sown version of ChatGPT. Hugging Face, the AI startup backed by tens of millions in venture capital, has released an open source alternative to OpenAI’s viral AI-powered chatbot.

Interesting Projects

Zap Weeds with AI-Power Lasers. Yes, a fricking laser! I love the premise of this project, in order to avoid spraying nasty chemicals on your crops to destroy weeds, tow along an AI-powered weed spotter and terminator. Of course, that just leaves trying to eliminate the use of pesticides – I wonder how that will be solved (hopefully more lasers, mwah-hah-ha with a little finger in the side of the mouth).

Chatbot in a Speaker. Hoani Bryson decided to make a standalone voice-operated ChatGPT client in a Speaker complete with a ring of cool mood LED’s. 

Australia aiming to use drones to detect Koalas. Australia is expanding its project to survey koalas. The idea is to use citizen scientists to fit thermal cameras to drones and collect data which is then fed to the project’s AI to process and count the koalas.

Baby cry detection using ChatGPT.ChatGPT, TinyML and Text-to-Audio technologies were used to create artificial data for detecting baby crying. The project uses ChatGPT to generate a load of text to describe scenarios of a baby crying, ie. “Tearful baby during bath”. These are fed to another AI that takes test descriptions and tries to generate audio of that. These audios are used as the dataset to train an AI model using Edge Impulse and then deployed on an Arduino Nicla Voice.

Cool Stuff

MatrixSpace announces AI-enabled Radar Modules. MatrixSpace announces the first in a series of small Radar modules that combines sensing, real-time AI edge processing, and RF communication in a package the size of a cellphone to equip large-scale use cases, such as semi-autonomous mobile platforms, general aviation, and portable or fixed systems. Digitizing the outdoors is the ability to measure objects’ size, location, and movement in time, shifting reliance from human sensing to a far more detailed recognition of objects and their movements.

“Liquid” Neural networks adapt on the go. Recent research has shown that small AI networks called Liquid Neural Networks have enabled drones to navigate new environments, outperforming more advanced systems. These are a class of networks that learn on the job, continually adapting to new information just like our brains.

Facebook’s Meta AI has announced its Segment Anything project. Segmentation is a fundamental part of computer vision and is the ability to identify which image pixels belong to an object. Using AI for this task requires training with large datasets of suitable images. What Meta AI has done with their Segment Anything Model (SAM) is pre-train it over billions of images (probably all our holiday snaps ), so it can reliably segment just about anything you ask it to in an image. To experiment with it and run it, you’ll need Python (>=3.8), PyTorch (>=1.7) torchvision >= 0.8, and to poke around their Github.

Interesting Projects

I came across an interesting project at Hackster : “Face Detection with MTCNN and TensorFlow Lite for ESP32-S3” enabling Face Detection and alignment on an ESP32 dev kit. Detection of a face is around 1 second.

This Week in AI

Elon Musk, shortly after calling for a pause in AI, which cynics felt he just wanted to catch up, has announced his rival AI company which is called X.AI.

All the excitement over ChatGPT and the fact that Microsoft has a version built into Bing has forced Google to accelerate its plans to incorporate AI into its search. This is allegedly happening in two stages, the first being project “Magi”, which will add Generative AI features to search. This will allow it to answer questions and follow-up questions and even write code. Further down the line, following Magi is an all-new search engine powered by AI. It has also merged its Brain team and Deep Mind to form Google Deepmind. Google Deepmind aims to significantly accelerate the company’s progress in AI.

OpenAI’s CEO says the Age of Giant AI models is already over.  OpenAI CEO Sam Altman says that the research strategy that birthed ChatGPT is played out and that further progress will not come from making models bigger.

Microsoft reportedly working on it’s own AI chips that could rival Nvidia’s. Microsoft is reportedly working on AI chips for training large language models. The chips have been in development in secret since 2019. They are currently being tested with large language models like GPT-4.

This is potentially a massive announcement. Sony, the company that already manufactures the Raspberry Pi has announced that they have made a “strategic” investment in Raspberry Pi Ltd to bring edge AI to the Raspberry Pi!

“Our goal is to provide new value to a variety of industries and support them in solving issues using our innovative edge AI sensing technology built around image sensors,” said Terushi Shimizu, President and CEO of Sony Semiconductor Solutions Corporation. “We are very pleased to be partnering with Raspberry Pi Ltd. to bring our AITRIOS platform — which supports the development of unique and diverse solutions utilising our edge AI devices — to the Raspberry Pi user and developer community and provide a unique development experience.”

So AITRIOS is Sony’s platform for developing Edge AI systems rather than a chip or family of chips. Will this be a native software capability integrated into Raspberry Pi OS or some hardware added to a new Raspberry Pi device (Raspberry PI 5?). It’s unclear, and Eben Upton, CEO of Raspberry Pi says:

“This collaboration further expands our relationship and introduces products using Sony Semiconductor Solutions’ AI technology into the Raspberry Pi ecosystem. This will allow users to use machine learning at the edge to create new attractions. We will help you to build a realistic application.”

Eben Upton, CEO Raspberry Pi

But nonetheless, it shows both companies’ direction in easier and more powerful focus and support for AI. How exciting is that?

Hardware

OpenMV, which makes camera/processor modules announced a new upgraded camera in the Open MV Cam RT. This is an upgrade to their Cam H7 Plus, which they’ve struggled to manufacture due to the ongoing chip shortages. So instead, they decided to design a better camera designed around parts they could get hold of. The Cam RT is based around a Cortex-M7 @ 600 MHz w/ Double Precision FPU, is around 1.25 faster than the H7, and has 4x the program space, High-speed USB, 10/100 Ethernet, Wifi and BLE onboard. They’re slating a 2k production run in June.

ArduCam is saying that their Arducam Mega – SPI Camera will start shipping out in the next few weeks. This is an autofocus camera that uses SPI to link to any microcontroller. What is unique is that both commands AND image data are sent across a ubiquitous SPI bus. I was sceptical that you could get sufficient bandwidth to do this, but the demos seem to suggest a decent throughput can be obtained. I’m a backer, so I will be excited to try this new camera out.

Whilst on the subject of Arducam, their Pico4ML-Pro TinyML Dev Kit is on presale for shipment in April and they’ve just announced that you’ll also get an enclosure for this bad boy. This is a Raspberry Pi 2040-based TinyML development kit no onboard camera) that can do Person detection, Wake Word detection and Magic Wand. It comes with a 2.4’ LCD Display and can connect to 4 Arducam Mega Cameras, all for the very respectable $59.99.

Crowdfunding Now

Duovox Ultra: Military-Grade Night Vision Monocular is a “Full colour at low light & 7-level IR illuminator at pitch darkness,1080FHD, up to 500m range, up to 10 hrs battery”. Pretty neat, with pledge levels starting at $129 and 22 days to go.

This Week in AI

Chinese technology company Alibaba announced plans to roll out it’s ChatGPT rival called Tongyi Qianwen. Tongyi Qianwen roughly translates as “seeking an answer by asking a thousand questions.”

Some slightly older ChatGPT news: ChatGPT banned in Italy over privacy concerns. They said there was no legal basis to justify “the mass collection and storage of personal data for the purpose of ‘training’ the algorithms underlying the operation of the platform”. They are going to try and get OpenAI under the General Data Protection Regulation (GDPR) regulations. OpenAI has “20 days to say how it would address the watchdog’s concerns, under penalty of a fine of €20 million ($21.7m) or up to 4% of annual revenues.”

And who can ignore the worry that AI is going to end the World as leading figures call for a suspension of AI Training. They want “training of AIs above a certain capacity to be halted for at least six months.” Leading luminaries, including Elon Musk and Steve Wozniak, Apple’s co-founder, have all signed an open letter calling for the halt. You could argue that a lot of those asking for the halt are all engaged in their own AI tools and feel that they need to catch up. In fact, what a surprise, Elon now announces his own rival AI company. Funny that.

The sad reality, I feel, is that whatever the perceived dangers of AI represent, the genie is already out of the bottle. You can’t stop everyone now advancing it, and particularly you can’t sit idly by whilst certain not-so-democratic regimes steal a march on the rest of the World. The AI arms race has already begun, and I don’t think we can stop it.

“AI humans plot and gossip like us”. AI researchers at Stanford created human-like entities with ChatGPT to “create 25 characters who “live” in a fictional world called Smallville” (where’s Superman???). “They slept, cooked meals and went to work. They formed opinions, made new friendships, nurtured grudges and gossiped”. !!