AI Weekly: GPT-4o, AlphaFold 3 revolutionizing drug discovery, Apple’s M4 chip, and more – 05/15/2024
👇 ChatGPT can now converse with you through video, detect sarcasm, read emotions in real time… and you don’t need a Plus subscription to try it!
AI advances sure feel exponential. This week’s AI Weekly edition features a long list of exciting releases from OpenAI, Google, Apple, Microsoft, and Neuralink, most of which are bound to have an enormous impact on your life in the coming months and years.
From this week’s selection: GPT-4o, AlphaFold 3 from Google DeepMind & Isomorphic Labs, Apple’s M4 chip, Microsoft's MAI-1, and Neuralink’s human brain implant.
📢 OpenAI launches GPT-4o: accessible for all subscription tiers
OpenAI has announced GPT-4o, their new flagship model capable of real-time reasoning across audio, vision, and text. GPT-4o ("o" for "omni") enables natural human-computer interaction by processing any combination of text, audio, and image inputs and outputs. With response times averaging 320 milliseconds, it matches human conversation speed and improves on non-English languages, vision, and audio understanding compared to previous models. It is also 50% cheaper and faster in the API.
The model showcases capabilities such as real-time translation, interview preparation, sarcasm detection, and customer service. Unlike previous models, GPT-4o is trained end-to-end across text, vision, and audio, enhancing its ability to interpret and express emotional nuances.
GPT-4o's advanced language tokenization reduces tokens across various languages, significantly improving efficiency. It also incorporates solid safety measures, including new safety systems for voice outputs. Initially, text and image capabilities are available, with audio and video capabilities to follow for trusted partners.
It’s now accessible in ChatGPT's free tier and to Plus users with higher message limits. Developers can also access GPT-4o via the API, offering twice the speed, half the cost, and higher rate limits compared to GPT-4 Turbo.
🧬 Google DeepMind and Isomorphic Labs have released AlphaFold 3
"In a paper published in Nature, we introduce AlphaFold 3, a revolutionary model that can predict the structure and interactions of all life’s molecules with unprecedented accuracy. For the interactions of proteins with other molecule types we see at least a 50% improvement compared with existing prediction methods, and for some important categories of interaction we have doubled prediction accuracy." – Google DeepMind AlphaFold team & Isomorphic Labs
Google DeepMind, in collaboration with Isomorphic Labs, has announced the release of AlphaFold 3, the latest iteration of its revolutionary protein structure prediction model. This new version significantly improves upon its predecessors by not only predicting protein structures but also their interactions with other biomolecules like DNA, RNA, and small molecules, crucial for drug discovery.
While AlphaFold 3 will not be open-sourced, Google has introduced the AlphaFold Server to provide researchers worldwide free access to the model for non-commercial use, thus supporting scientific advancement while managing intellectual property concerns. The introduction of AlphaFold 3 promises to enhance our understanding of biological processes and could lead to significant breakthroughs in medicine and biology.
Recommended by LinkedIn
🍏 Apple introduces M4 chip
Apple has officially unveiled its new M4 chip, a significant advancement in its silicon technology, designed specifically for the latest OLED iPad Pro models. This new chip, utilizing second-generation 3nm technology, is capable of handling a staggering 38 trillion operations per second, marking a notable leap in performance from its predecessors. While it outpaces previous Apple chips, it still lags slightly behind Qualcomm's Snapdragon X Elite.
The M4 features a 10-core CPU that balances performance and efficiency cores, and it supports advanced graphical technologies such as mesh shading and ray tracing. With this release, Apple enters into competition with Microsoft's upcoming Arm-powered hardware. The race for dominance in CPU and AI capabilities in mobile and portable devices is on!
🤠 Microsoft is developing a new AI model: MAI-1 in the works
Microsoft is developing an AI model called MAI-1, intended to rival OpenAI's GPT-4 and Google's Gemini. This new model, led by Mustafa Suleyman (CEO of the newly founded Microsoft AI), who previously worked at Google AI, reportedly works with around 500 billion parameters. Developed using technology and data from the recently acquired startup Inflection, MAI-1 would mark a significant push towards Microsoft’s independence in AI innovation, reducing its reliance on collaborations with OpenAI.
Read Microsoft CTO Kevin Scott’s post mentioning MAI-1 on LinkedIn.
😮 Neuralink’s first human brain implant experiment faces setback
Neuralink has encountered a setback with its first human brain implant, the N1 chip, where some micro-thin threads retracted after implantation, possibly due to air trapped in the skull. Despite this, the company managed to enhance the device's data transmission speed by optimizing their recording algorithm and improving neural signal translation. The initial participant, Noland Arbaugh, has since been able to control digital devices more effectively.
Neuralink is planning further implants and continues to refine their technology, working closely with the FDA to ensure safety and efficacy. This development is a part of their broader effort to improve digital interaction for people with quadriplegia.