DeepMind releases AlphaFold 3 - Weekly News Roundup - Issue #466
Plus: OpenAI releases Model Spec; Neuralink publishes progress update; Tesla shares new video of Optimus; growing meat with Gatorade; and more!
Hello and welcome to Weekly News Roundup Issue #466. This week we will take a closer look at the AlphaFold 3, a newly released biology prediction tool from Google DeepMind and its drug discovery spin-off, Isomorphic Labs.
In other news, OpenAI published Model Spec, a glimpse into the set of rules and guidelines defining the behaviour of ChatGPT. Neuralink shared the progress update of its first human trial. Meanwhile, in robotics, Tesla has shared a new video of its humanoid robot Optimus, and the US military is testing robot dogs with guns attached to their backs. We will finish this week’s issue with TSMC’s announcement of a new process node beyond N2.
Enjoy!
In 2021, DeepMind released AlphaFold 2, an AI model designed and trained to solve the protein folding problem - determining the three-dimensional structure of a protein based solely on its amino acid sequence. It is a challenging problem, which normally takes years of hard work and experimentation to produce the 3D structure of just one protein. AlphaFold 2 was able to solve the protein folding problem and predict with good accuracy what the structure of the protein would look like. Not only that, a year later, in 2022, DeepMind uploaded the structures of around 200 million proteins from 1 million species, covering nearly every known protein on the planet, to the AlphaFold Protein Structure Database. These structures were predicted by AlphaFold and are available for free for everyone to use for free and advancing scientific research.
Now DeepMind takes AlphaFold to the next level with the release of AlphaFold 3. The new, improved model can not only predict the shape of proteins but also nearly all of the elements of life - DNA, RNA, ligands (small molecules interacting with proteins which play a big role in drug development), and more. AlphaFold 3 can also predict how those molecules will interact with each other, a key functionality required in drug development or any kind of molecular engineering task.
In a paper published in Nature, researchers from DeepMind and Isomorphic Labs (DeepMind’s spin-off company focusing on applying AI in drug discovery) write that AlphaFold 3 achieves higher performance in all but one category of tasks over systems built specifically to perform that task. For example, AlphaFold 3 is better at predicting the structure of ligands than specialised physics-based systems widely used in the industry. However, we will need independent tests to verify DeepMind’s claims.
The main change between AlphaFold 2 and AlphaFold 3 is the replacement of a graph-based module called Evoformer which “imagines” how the proteins should look like with a new, simpler module called Pairformer. There is also a new diffusion module, similar to those used in text-to-image models. But instead of generating a requested image from noise, this diffusion module generates 3D molecular structures.
Unlike AlphaFold 2, whose code was open-sourced, DeepMind does not plan to do the same with AlphaFold 3, as MIT Technology Review reports, quoting a call with Demis Hassabis. Instead, DeepMind and Isomorphic Labs launched AlphaFold Server, where everyone can play with AlphaFold 3 for free, provided it's for non-commercial purposes and does not violate the terms of service. However, this move limits access to AlphaFold 3. AlphaFold Server seems like a preview of what AlphaFold 3 can do and I can see a future in which full access to the model is available behind a paywall, as a subscription. AlphaFold Server might be the first step towards commercializing the technology under Isomorphic Labs.
The primary application of AlphaFold 3, according to a press release, is to “transform our understanding of the biological world and drug discovery”. With AlphaFold 3, researchers can develop new drugs quickly and cheaper than before. New therapies, which previously were unavailable to the general public, could have a massive impact on health and the quality of life for millions of people. But I also would like to highlight potential applications of AlphaFold 3 outside medicine. The biotech revolution is coming. Right now, we are in the middle of the AI revolution which will be soon followed by the robotics revolution. Next in line will be the biotech revolution. Together with other computational tools, AlphaFold 3 opens new possibilities for what we could do with biology.
If you enjoy this post, please click the ❤️ button or share it.
Do you like my work? Consider becoming a paying subscriber to support it
For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.
Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.
🦾 More than a human
Neuralink: PRIME Study Progress Update — User Experience
Neuralink has shared the results of the PRIME Study whose goal was to demonstrate that the company’s brain-computer interface (BCI) is safe and useful in real life. Thanks to the implant, Noland Arbaugh, a PRIME Study participant, could control his laptop and play games for the first time in eight years. During his first-ever research session, Noland set a new world record for human BCI cursor control. However, in the weeks following the surgery, a number of threads retracted from the brain, resulting in a decrease in the number of effective electrodes.
Baby born deaf can hear after breakthrough gene therapy
Thanks to gene therapy, an 11-month-old baby girl born deaf due to auditory neuropathy can hear for the first time. The therapy involved an infusion containing a harmless virus delivering a working copy of the OTOF gene directly into her cochlea. Within four weeks, she began responding to sounds, and by 24 weeks, she achieved near-normal hearing for soft sounds.
🧠 Artificial Intelligence
OpenAI: Introducing the Model Spec
OpenAI has published Model Spec, a collection of high-level rules that indirectly govern ChatGPT and other models. The document gives a glimpse into how OpenAI has conditioned ChatGPT’s behaviour and its rules to determine when to reject a request. However, that does not mean OpenAI is fully opening its models. Model Spec is intended for developers and researchers to better understand how ChatGPT works under the hood. Model Spec is also a part of OpenAI’s ongoing public conversation about how models should behave, how desired model behaviour is determined, and how best to engage the general public in these discussions.
OpenAI considers allowing users to create AI-generated pornography
OpenAI is considering allowing users to create AI-generated not-safe-for-work (NSFW) content, including erotica and nudity, while maintaining a strict ban on deepfakes. The company aims to explore responsible ways to enable the generation of such content in age-appropriate contexts. Critics, however, argue that this undermines OpenAI's mission to develop "safe and beneficial" AI, with concerns around the potential for misuse in creating explicit images, including those involving children.
Rishi Sunak promised to make AI safe. Big Tech’s not playing ball.
Six months ago, at the first AI Safety Summit in Bletchley Park, UK, the leading companies in the AI space agreed to share their closely guarded AI models with the British government to assess their safety. However, as Politico reports, the AI Safety Institute, which was founded to perform safety checks, has problems getting access to evaluate the new models before they are released and is failing to do its job. Of the leading AI labs, only Google DeepMind has allowed anything approaching pre-deployment access, according to the article.
Ukraine unveils AI-generated foreign ministry spokesperson
Ukraine unveiled an AI-generated spokesperson called Victoria who will make official statements on behalf of its foreign ministry. The ministry said it would “for the first time in history” use a digital spokesperson to read its statements, which will still be written by humans.
Eight newspaper publishers sue Microsoft and OpenAI over copyright infringement
Eight US newspaper publishers sued Microsoft and OpenAI over copyright infringement, adding one more case for lawyers of both companies to deal with. According to the publishers, ChatGPT and Copilot have been “purloining millions of the publishers’ copyrighted articles without permission and without payment.”
Chinese Large Model Benchmark Evaluation April 2024 Report
If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?
🤖 Robotics
Tesla shares video of Optimus autonomously sorting battery cells
Tesla shared a new video of its humanoid robot, Optimus, showing the robot’s ability to recognise battery cells, grab them and put them in a box. According to the video, Optimus is being tested in one of Tesla’s factories.
Rifle-Armed Robot Dogs Now Being Tested By Marine Special Operators
The United States Marine Forces Special Operations Command (MARSOC) is currently testing robot dogs with rifles attached to their backs. These robots, made by Onyx Industries, could be the first of their kind used by the US military if the tests are successful. These robotic dogs are equipped with a remote weapon system, capable of performing tasks like tunnel work and perimeter security. Despite the advanced capabilities of these robots, including autonomous target detection and engagement with human oversight, MARSOC has clarified that this technology is still under evaluation and not yet fielded.
🧬 Biotechnology
The US Is Cracking Down on Synthetic DNA
The White House has issued new regulations targeting synthetic DNA manufacturers. The new rules, which are the result of an executive order signed by President Joe Biden last fall to establish new standards for AI safety and security, will require companies to screen for hazardous genetic sequences and verify customer legitimacy, but they currently only apply to entities receiving federal funding. The regulation aims to standardize safety measures across an industry where voluntary compliance varies, highlighting the need for consistent oversight in biotechnology.
Before They Hatch
This article explores how three technologies—in-ovo sexing, on-farm hatching, and in-ovo vaccination—hold the promise to transform the poultry industry. It explains how each of these technologies works and argues that they will make chicken farms not only more humane but also more efficient.
▶️ Growing Meat in... Gatorade (22:53)
In this video, one of my favourite mad scientists on YouTube, The Thought Emporium, shows how to turn common items you can get from a grocery store and Gatorade into replacement growth medium for lab-grown meat. Although this experiment might sound random and crazy, the results show that it might be possible to replace expensive growth mediums with cheaper and more sustainable alternatives.
💡Tangents
▶️ Major New Announcements From TSMC Symposium (6:50)
During the North America Tech Symposium, TSMC revealed its plans for the second half of this decade, which included the announcement of a new process node beyond N2, called A16, among other developments. Dr Ian Cutress from Tech Tech Potato breaks down these announcements and explains what they mean for the near future of computing and chip design.
Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.
Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.
A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!
My DMs are open to all subscribers. Feel free to drop me a message, share feedback, or just say "hi!"
Once again, in one of my weekly spaces dedicated to in-depth newsletters reading, I can only appreciate the collection of interesting things here!
In this issue, my attention was sincerely focused on Neuralink's PRIME study Progress Update. The study - and this is an important thing not only from a technological and medical point of view, but also from a technology adoption and marketing point of view - focuses on the user experience. The report illustrates some users perceptions, like this: 'The biggest thing with comfort is that I can lie in my bed and use [the Link]. Any other assistive technology had to have someone else help or have me sit up. Sitting causes stress mentally and on my body which would give me pressure sores or spasms. It lets me live on my own time, not needing to have someone adjust me, etc. throughout the day'. Regardless of opinions on the subject, it is definitely something to read. Thanks again for Conrad for giving me the opportunity not to miss it!