- author: Matt Wolfe
Recent Advancements in AI: Open-source Models and Generative AI Tools
Despite the impression that there has been a slowdown in AI news, there have been numerous advancements in the field over the past week. While there have not been any groundbreaking announcements like the release of GPT-5 or mid-journey version 6, there have been significant developments that are adding up to a lot in the AI space.
Open-source AI Models
Last week Meta released an AI music generation model, which is open-source and free to use on Hugging Face. Compared to other models such as Refusion or Music LM from Google, the model from Meta is superior. In fact, many people are already experimenting with this model, producing impressive results including 90's rock songs and early 2000's pop-punk anthems. It is important to note that this model can be installed locally using a graphics card with at least 16 GB of memory, or used on Google Collab.
Moreover, Meta is not the only tech giant investing in open-source AI models. OpenAI, DeepMind, and Anthropic have also decided to open up their models to the UK government for research and safety purposes. This move comes as various governments are requesting access to these models for long-term implications of AI.
Generative AI Tools
Recently, Adobe made some updates to Adobe Express (similar to Adobe's version of Canva) and announced that Adobe Firefly generative AI comes to Adobe Express. You can now prompt images and add text effects using AI directly inside of Adobe Express.
Moreover, Adobe Illustrator has also added generative AI tools, including generative recolor that allows the uploading of a black and white image or design to be regenerated with other color schemes. With the help of AI, Adobe Illustrator finds color palettes that look good together and re-colors images with interesting new variations.
Additionally, Google has also integrated AI features into some of its products such as Gmail and Google Docs. Users can now use the "help me write" button to create prompts and quickly generate persuasive emails and articles. Although these tasks could be accomplished simply by using larger language models like Chat GPT or BART, Google's AI integration into these tools is saving users extra time.
More Human-like AI
Meta has also created a human-like AI model. Based on Jan Lacun's vision for more human-like AI, this model can take a small piece of an image for context and find the rest of the image by applying what it already knows about the subject matter. Therefore, when given a piece of a dog's head, for example, the model has been trained to recreate the rest of the dog's body. This technology is available on Github and can be used for other image-generating applications.
AMD's Partnership with Hugging Face
Finally, AMD has partnered with Hugging Face, which is a platform where people can upload their machine learning models and their code. During their recent live event, AMD made two significant announcements. The first is their collaboration with Hugging Face, providing the compute power behind this platform. Second, they are building hardware specifically tailored for AI that competes with Nvidia's chips. AMD is aiming to create chips that are more powerful than Nvidia's by designing Mi 300X. Through this collaboration, AI democratization will be more accessible for everyone.
Overall, these recent advancements highlight the ongoing growth and evolution of AI. It will be interesting to see what the future holds for AI and how these new models and tools will affect various industries.
AI Hardware: AMD vs Nvidia
At a recent event, AMD announced its plans to build hardware specifically tailored for AI, called the Mi 300X, which is designed for generative AI and optimizes for the larger memory requirements of large language models. The chip combines cna3 with an industry-leading 192 gigabytes of HBM3 memory, which is 2.4 times more memory and 1.6 times more memory bandwidth than Nvidia's competing H100 chip. While these chips are not likely to be available for consumer PCs in the near future, this competition between AMD and Nvidia highlights the increasing focus on creating more powerful GPUs for training large language models.
OpenAI Updates: GPT-3.5 and Chat Bot Integration
OpenAI also made several updates, including the release of a new 16,000 context version of GPT-3.5, making it four times larger than the previous version, which had 4,000 context. The API costs for GPT-3.5 usage will also decrease, although it remains to be seen if companies using the API will pass the cost savings on to their consumers. Additionally, tensions have reportedly arisen between OpenAI and Microsoft, who have a unique partnership as both companies support and compete with each other in AI features. OpenAI had warned Microsoft to proceed cautiously with integrating GPT-4 into Bing search to avoid inaccurate and unpredictable responses, but the company went ahead despite the warnings.
AI and Music: New Beatles Song
Speaking of AI in music, Paul McCartney recently revealed that AI technology was used to create a final Beatles song. The technology was used to extract John Lennon's voice from old demos, allowing McCartney to complete the song, which is set to release this year.
Virtual Try-Ons and AI Modeling
Google also announced a new generative AI model for virtual try-ons that allows users to generate try-on images with AI. Using diffusion models similar to those used in Stable Diffusion, this AI modeling can be used when selling clothes online, where users can upload pictures of themselves to see how different clothes fit. This new model was created using Google's shopping graph, featuring images of people wearing various clothing styles from brands such as Anthropologie, LOFT, H&M, and Everlane. It will become more accurate over time and expand to other Brands.
Shopify: Democratizing Technology for Entrepreneurs
Shopify is enabling entrepreneurs to become successful by providing a platform where anyone can set up their own store to sell physical products. Setting up an eCommerce store doesn't have to be complicated, and Shopify makes it simple for anyone to build their own business online. By using Shopify, Matt Wolf was able to set up the Future Tools merchandise store, and anyone can do the same. Thanks to Shopify for sponsoring this video and enabling people to start their own online stores with ease.
Using AI to Create Unique Images
Finally, the latest trend on Twitter is using control net and stable diffusion with QR codes to create unique images. This trend has been gaining interest with tweets from Rowan Chung, Linus Eckenstam, and copy Sutra, who have all shared tutorials on how to set up the technique. Michael GAO also created a tutorial, showing that anyone can try this out easily. AK on Twitter also shared that there's a model that you can download directly. As AI continues to evolve, we can expect more exciting discoveries and trends to emerge.
The Latest Tech Trends: From QR Code Art to Skin Condition Detection
This week has been jam-packed with exciting tech news and trends. Here's a quick rundown of the most interesting updates:
QR Code Art: The Latest Trend
QR code art has been taking over Twitter lately, with artists creating unique and intricate designs that can be scanned with a smartphone's camera. While this trend is incredibly hot right now, it's not quite perfect yet. Generating QR codes that both look good and scan effectively poses a challenge, but several tutorials have been published online to help enthusiasts achieve the best results.
- Linus Eckenstam's buddy, Copy Sutra, created a three-step tutorial on how to create QR code art.
- Michael GAO and AK on Twitter also shared their tutorials on the subject.
- Rowan Chung's tutorial on QR code art recently went viral with 1.3 million views.
While the technology still has some kinks to work out, QR code art has the potential to be used in the future for businesses and marketing purposes.
Google Lens Helps Identify Skin Conditions
In a recent blog post, Google announced that Google Lens can now help users identify and search for skin conditions. Users can upload a photo of any unusual bumps, discolorations, or rashes, and Google Lens will detect the skin condition and provide relevant information. While this technology is not meant to replace medical professionals, it does provide an extra resource for those who are curious or concerned.
Re-Rendered Video Improves Video Quality
Researchers have recently announced a new technology called re-rendered video, which significantly improves the quality of videos. Unlike other video models like stable diffusion and gen 1, re-rendered video generates videos without flickering or incoherencies. While it's not available to the public yet, researchers hope to make it accessible to everyone soon.
Mid-Journey Releases Version 5.2
Mid-Journey, a popular text-generating AI, is expected to release its new version 5.2 any day now. The update will include new tools and features, including limited Discord-compatible outpainting and a prompt analyzer that reduces word barf prompts. Mid-Journey is also working on improving moderation to prevent certain prompts from being denied or rejected.
11 Labs' AI Speech Classifier Fights Against Deep Fakes
11 Labs, known for its ultra-realistic text-to-speech voices, recently introduced an AI speech classifier, which can identify whether a particular audio file was generated by 11 Labs. The company hopes to use this tool to prevent deep fakes and malicious use of AI technology. While 11 Labs has been criticized in the past for making it easy to deep fake anyone's voice, this tool is a proactive step in the right direction.
Overall, this week's tech trends showcase the incredible advancements being made in AI and computer vision technologies. While these updates have their challenges and limitations, they also provide a glimpse into what the future of technology may hold.
Conclusion and Call to Action
In conclusion, AI generated audio is becoming more advanced and accessible, with new developments and applications being announced every week. As exciting as these advancements are, it is important to acknowledge the potential for misuse and take proactive measures to prevent malicious use. It's encouraging to see AI companies taking responsibility and working towards a more ethical future.
If you're interested in staying up to date with the latest AI news and developments, be sure to subscribe to this channel and check out futuretools.io for real-time updates. Additionally, there are a variety of exciting tutorials and videos being produced, covering topics such as GPT prompts, AI music generation and warp fusion.
Thank you to our sponsor, Shopify, for making these videos possible. And thank you to all of our viewers for tuning in and supporting this channel. Don't forget to like, subscribe, and hit the notification bell to stay up to date with all of our latest content.