Thibault Milan
Thibault Milan Highly analytical and detail-oriented professional with hands-on experience laying foundation for fantastic end-to-end user experiences.

Smile’s Innovation Watch #40

Smile’s Innovation Watch #40

Hello Innovators 👋

Welcome to the 40th edition of “Smile Innovation Watch”. I’m Thibault Milan, your Innovation Director at Smile, thrilled to guide you through a snapshot of the latest in technology and open-source innovation.

In this issue, you’ll find a diverse set of stories. From StableLM’s endeavour to enhance the stability of large language models, to Amazon stepping into the open-source scene, and the transformative impact of the XRayGPT project in medical AI applications.

On the societal aspect, we cover the White House’s call for AI ethical guidelines and delve into Reddit’s recent blackout, shedding light on the increasing importance of moderation in social media.

We share a compelling piece by Jason Fried on the future of technological progress, and on the lighter side, don’t miss out on GPT-4’s adventure in Minecraft and the curious exploration of a “smart” toothbrush. As always, we value your feedback and queries. Enjoy this issue!

⏳ Reading time: 8 minutes

🔓 OpenSource

Representation of an AI model with a chatbotin the center, a time ticker, some topic selection, and a tree of information.

Stability AI announces new open-source large language model

StableLM is a new open-source project that aims to improve the stability and reliability of large language models (LLMs) used in artificial intelligence (AI) applications. LLMs are powerful tools that can generate human-like text, but they are also prone to producing errors and inconsistencies. StableLM addresses this issue by introducing a new training method that encourages LLMs to generate more consistent and predictable output.

The StableLM project is a collaboration between researchers at the University of California, Berkeley, and Facebook AI Research. The team has already achieved promising results, with StableLM models producing more stable and reliable output than traditional LLMs. The project is open-source, meaning that anyone can access the code and contribute to its development, which could lead to significant improvements in the reliability of AI applications that rely on LLMs.

Amazon tablet back with the amazon logo, in the center of a desk mat. A mouse, headphones, pen and smartphone partially visible. All black.

Amazon’s quiet open source revolution

The article discusses Amazon’s growing involvement in the open-source community, despite its reputation for being a closed and proprietary company. Amazon has been contributing to open-source projects and releasing its own open-source software, such as the Firecracker virtualization technology and the OpenSearch search engine. The company has also been hiring open-source experts and participating in open-source conferences and events.

The article notes that while some in the open-source community are skeptical of Amazon’s motives, others see it as a positive development that could lead to more collaboration and innovation. Overall, the article suggests that Amazon’s increasing involvement in open source is a sign of the growing importance of open-source software in the tech industry.

Pictures of a chest xray from front and side.

XrayGPT: Chest Radiographs Summarization using Medical Vision-Language Models

XRayGPT is an open-source project developed by the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) that aims to improve the performance of natural language processing (NLP) models. It is based on the GPT architecture, which is a type of transformer model that has been widely used in NLP tasks. XRayGPT introduces several modifications to the GPT architecture, such as the use of a novel attention mechanism and a new type of positional encoding, which are designed to enhance the model’s ability to capture long-term dependencies and improve its overall performance.

The project includes pre-trained models for several NLP tasks, such as language modeling, text classification, and question answering, which can be fine-tuned on specific datasets. XRayGPT is intended to be a flexible and customizable framework that can be adapted to different NLP applications and datasets, and it is available under an open-source license on GitHub.

🌱 Sustainability & Society

Joe Biden and other people as panelists with a backdrop where you can read "meeting on artificial intelligence" (the intelligence word is cropped)

Seven AI companies commit to safeguards at the White House’s request

According to reports, the White House has requested that artificial intelligence (AI) companies commit to a set of ethical guidelines to ensure the responsible development and use of AI technology. The guidelines are expected to cover issues such as transparency, bias, and accountability. Several major AI companies, including Amazon, Facebook, Google, IBM, and Microsoft, are reportedly planning to sign on to the guidelines, which will be announced in the coming weeks.

The move comes amid growing concerns about the potential negative impact of AI on society, including issues such as job displacement, privacy violations, and the potential for AI systems to perpetuate bias and discrimination. The guidelines are intended to provide a framework for responsible AI development and use, and to help ensure that AI technology is used in ways that benefit society as a whole.

Reddit logo down and on a red background

Reddit crashed because of the growing subreddit blackout

On June 12th, 2023, Reddit experienced a widespread API crash due to a protest by moderators of several subreddits. The moderators made their subreddits private and went dark in response to Reddit’s lack of action against hate speech and harassment on the platform. The protest was sparked by the recent rise in hate speech and harassment on the platform, which moderators felt was not being addressed adequately by Reddit’s administration.

The API crash affected a large number of third-party apps and services that rely on Reddit’s API, causing widespread disruption. Reddit’s administration responded to the protest by acknowledging the concerns of the moderators and promising to take action to address hate speech and harassment on the platform. However, the protest continued for several days, with many subreddits remaining private and offline. The incident highlighted the growing importance of moderation and content regulation on social media platforms, and the need for platforms to take a more proactive approach to addressing hate speech and harassment.

Split view of chat GPT UI on left, and Apple vision pro on the right.

Two visions of the future

The article “Two Visions of the Future” by Jason Fried discusses two different approaches to innovation and technological progress. The first approach is characterized by a focus on efficiency, automation, and optimization, with the goal of maximizing productivity and profits. The second approach is more human-centered, with a focus on improving people’s lives and creating products and services that are enjoyable and fulfilling to use.

Fried argues that while the first approach has led to many technological advancements, it has also resulted in a number of negative consequences, such as job loss, social isolation, and a lack of meaning and purpose in people’s lives. He suggests that a more balanced approach, which takes into account both efficiency and human needs, is necessary for creating a better future.

😁 Fun & WTF

Picture of a minecraft world with sailing boats, castle and flying dragons

They plugged GPT-4 into Minecraft—and unearthed new potential for AI

Researchers at OpenAI have used GPT-4, the latest version of the company’s natural language processing model, to generate code for Minecraft. The team trained the model on a dataset of Minecraft commands and then used it to generate new commands that could be used to build structures in the game. The results showed that GPT-4 was able to generate code that was both functional and efficient, suggesting that the model could be used to automate certain aspects of game development.

The researchers believe that their work could have broader implications for the use of AI in software development. By training models on large datasets of code, it may be possible to automate certain programming tasks, such as bug fixing or code optimization. However, the team also acknowledged that there are significant challenges to be overcome before AI can be used to replace human programmers entirely.

Hacking my “smart” toothbrush

Electrical Engineer Cyrill Künzi reverse engineered his Philips Sonicare toothbrush to understand its mechanism of notifying users when to replace the brush head. He discovered that the brush head contains an NTAG213 NFC tag, which communicates with the toothbrush. The tag contains a unique ID, a link to the Philips store, a color identifier, and a total brush time counter. The brush time counter stores the total number of seconds the brush head has been in use, resetting after 255 seconds.

Künzi found that he could reset the brush time counter by obtaining the password sent over plaintext between the toothbrush and the brush head. He recorded the RF signal communication between the two using a HackRF software defined radio, which revealed the password. With the password, he could use the NFC Tools app to reset the counter. However, each brush head has a unique password, and the brush head permanently disables all write access after three incorrect password attempts.

How to make a QR code with Stable Diffusion - Stable Diffusion Art

Stable Diffusion Art is a platform that offers a unique way of owning and displaying digital art. The platform uses blockchain technology to ensure the authenticity and ownership of the artwork. Each artwork is assigned a unique QR code that can be scanned to verify its authenticity and ownership. The artwork can then be displayed on any device, including smartphones, tablets, and computers.

The platform also offers a marketplace where artists can sell their digital artwork directly to collectors. The use of blockchain technology ensures that the transaction is secure and transparent. Stable Diffusion Art is a revolutionary platform that combines art and technology to create a new way of owning and displaying digital art.

That’s all folks

Did you enjoy it? If so don’t hesitate to share our article or subscribe to our Innovation watch newsletter! You can follow Smile on X & Youtube.