Weekly News Roundup - Issue #475
Plus: Microsoft leaves OpenAI's board; OpenAI got hacked; the first step toward reversible cryopreservation; how good ChatGPT is at coding; China proposes guidelines for humanoid robotics; and more!
Hello and welcome to Weekly News Roundup Issue #475.
There is no write-up this week. Life got in the way, and I just didn't have enough time to properly research and tell the story I wanted to share. I apologize for that. The story will go out next week.
This week, Microsoft left OpenAI's board, and xAI ended a $10 billion deal with Oracle, opting to build its own data center instead. Additionally, OpenAI disclosed a security breach, and Google announced a new AI training method that is 13 times faster and 10 times more power-efficient.
In robotics, China proposed guidelines for future humanoid development, and a man in Florida shot at a delivery drone.
We will also meet a cryopreservation startup that successfully vitrified neural tissue earlier this year and explore the debate on AI as a Product vs. AI as a Feature.
Enjoy!
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
▶️ The 1st Step to Reversible Cryopreservation (17:55)
Meet Cradle, a cryopreservation startup that earlier this year successfully vitrified neural tissue and confirmed its electrical activity after cryopreservation. Now, the startup is working to develop technologies to reversibly cryopreserve whole human organs for clinical use and human neural tissue for research use, with the ultimate goal of achieving whole-body reversible cryopreservation.
Master controller of aging and development uncovered
A new study from the University of Queensland offers a comprehensive analysis of changes across various cell types in both mice and humans as they age and highlights the crucial role of Activator Protein 1 (AP-1) in common pathways that govern the transition from youth to old age. This research holds potential for addressing age-related diseases such as Alzheimer’s, metabolic disorders, and stroke. Targeting AP-1 and its associated pathways could lead to therapies that slow down or even prevent the onset of these diseases, marking a significant advancement in geriatric medicine.
Musk says next Neuralink brain implant expected soon, despite issues with the first patient
Elon Musk said during a livestream that his brain tech startup Neuralink is hoping to implant its second human patient within “the next week or so.” As reported last week, the previous second patient had to withdraw due to medical conditions that made them unsuitable to participate in Neuralink’s study of its experimental device.
🧠 Artificial Intelligence
OpenAI breach is a reminder that AI companies are treasure troves for hackers
OpenAI was hacked last year, and the news about it has only surfaced now. According to The New York Times, hackers accessed only an employee discussion forum. The code and data for OpenAI’s AI models were not breached. This incident highlights how lucrative target AI companies are and underscores the need for strengthened cybersecurity measures.
As Microsoft leaves its observer seat, OpenAI says it won’t have any more observers
Microsoft is leaving its observer seat on the OpenAI board, stating that it has seen enough progress at the AI company and is confident in its direction, according to Axios. Additionally, OpenAI announced that it would no longer have any observers on its board, effectively shutting down rumours reported last week that Apple was set to get an observer seat.
Musk's xAI, Oracle end talks on $10 bln server deal, the Information reports
According to a report published by The Information, xAI, Elon Musk’s AI startup aimed at challenging OpenAI, has ended talks on a potential $10 billion server deal. A multi-year agreement to rent Nvidia processors from Oracle for a planned supercomputer was underway, but the negotiations were stalled due to issues such as Musk's demands to build a supercomputer faster than Oracle deemed feasible. Additionally, Oracle also raised concerns that xAI's preferred location had inadequate power supply. xAI will now build its own data centre, expected to be ready by Q2 2025, and it will be located in Abilene, Texas.
How Good Is ChatGPT at Coding, Really?
So, how good at coding is ChatGPT really? A recent study evaluated ChatGPT’s coding skills in terms of functionality, complexity, and security, and found that the AI has an extremely broad range of success, ranging from 0.66% to 89%—depending on the difficulty of the task, the programming language, and several other factors. Additionally, researchers found that ChatGPT has a higher chance of producing correct solutions for algorithmic problems posted before 2021, as those problems could be part of its training dataset.
Google claims new AI training tech is 13 times faster and 10 times more power efficient
Google DeepMind has published JEST, a new AI training method that the AI lab claims to be much faster and more power-efficient than current methods. According to DeepMind's paper, JEST "surpasses state-of-the-art models with up to 13× fewer iterations and 10× less computation."
▶️ AI the Product vs AI the Feature (8:15)
Is AI a product, or is it a feature? Marques Brownlee explores this question, and the answer seems to lie somewhere in between.
Open LLM performances are plateauing
People at HuggingFace realized that the benchmark they were using to evaluate large language models for the Open LLM Leaderboard had become overused and saturated, prompting them to create a new and more challenging leaderboard. HuggingFace has selected six benchmarks to measure models’ knowledge, reasoning on short and long contexts, complex mathematical abilities, and tasks that are well correlated with human preferences, such as instruction following. The hope is that this new leaderboard will better reflect the actual performance of open language models.
▶️ Scalable MatMul-free Language Modeling (Paper Explained) (49:45)
In this video, Yannic Kilcher reviews and explains a recent paper that introduces a new method for coding large language models that eliminates the need for matrix multiplications, an operation ubiquitous in neural networks. According to the paper, this new architecture achieves performance on par with state-of-the-art transformer models at scales up to 2.7 billion parameters and promises to use significantly less memory and energy, especially when implemented on custom hardware.
The developers suing over GitHub Copilot got dealt a major blow in court
A California judge has dismissed most of the claims in a copyright lawsuit against GitHub, Microsoft, and OpenAI, leaving only two claims intact: one alleging an open-source license violation and another for breach of contract. The lawsuit, filed in 2022 by a group of developers, accused the companies of copyright infringement by using the AI-powered GitHub Copilot coding assistant, which trains on developers' work.
If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?
🤖 Robotics
What happens if you shoot down a delivery drone?
Recently, a man in Florida allegedly shot down Walmart’s delivery drone. The man admitted to shooting down the drone, saying drones flew over his house multiple times, leading him to believe that the drones might be spying on him. This incident raises questions about the legal ramifications of such acts. As it stands, the FAA points to 18 U.S.C. 32, titled “Aircraft Sabotage,” which is concerned with the wanton destruction of manned aircraft, not drones. As drone delivery increases, a separate law specifically addressing drones might need to be passed.
China proposes guidelines for future humanoid development
During the World Artificial Intelligence Conference, five Shanghai-based industry organizations published the "Guidelines for Humanoid Robot Governance." The guidelines advocate for humanoid robots to safeguard human dignity and not threaten security, reflecting principles similar to Isaac Asimov’s “Three Laws of Robotics.” They promote ethical use, privacy protection, and emergency response systems, aligning with human values and safety. Currently, no official safety standards exist for humanoid designs, and these guidelines might prompt a conversation to develop comprehensive safety and performance standards.
"Breakthrough" robot with lab-grown "human brain" promises advancement in brain-computer interfacing
Chinese researchers have built a small robot powered by a brain organoid derived from human stem cells. Described as the "world's first open-source brain-on-chip intelligent complex information interaction system," the robot marks a significant advancement in brain-computer interfaces and could help advance the efforts to use lab-grown brain tissue to restore human brain functions.
🧬 Biotechnology
Insilico Medicine and Fosun Pharma deliver second preclinical candidate for solid tumor treatment
Insilico Medicine, an AI-driven drug discovery company, has successfully delivered its second preclinical candidate compound targeting DNA damage repair mechanisms for treating solid tumours in collaboration with Fosun Pharma. Using its AI platforms Insilico identified and designed the compound, which demonstrated strong anti-tumor efficacy and promising properties in preclinical studies.
💡Tangents
Ultra-detailed brain map shows neurons that encode words’ meaning
Scientists have mapped neurons that encode the meaning of words in the prefrontal cortex. By studying ten people with epilepsy, they discovered that specific sets of neurons activate for similar categories of words, such as actions or people-related words. Words with similar meanings triggered similar neuronal patterns. This research could aid in developing brain-computer interfaces to restore speech and offer faster language processing insights than traditional imaging techniques.
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!"