TSMC’s restrictions on AI chip sales to China stoke tech tensions

The US-China tech war is escalating on the silicon front. Taiwan Semiconductor Manufacturing Co. (TSMC), the world’s top chipmaker, will halt production of certain AI chips for Chinese customers, complying with intensifying U.S. efforts to restrict China’s access to silicon to drive its own AI research. This move, effective Monday, throws another wrench into the already strained relationship between the two superpowers and could have ripple effects throughout the global AI landscape. Simultaneously, China is pulling out all of the stops to bolster its own chip production capabilities.

Microsoft debuts healthcare AI foundation model catalog

Source: LinkedIn Post by Ingrid Vallee (November 5)

Microsoft has unveiled a comprehensive catalog of foundation models specifically designed for healthcare applications, according to a LinkedIn post from Ingrid Vallee.  The release aims to accelerate the adoption of AI in healthcare settings while maintaining high standards of accuracy and safety. For instance, the pre-trained models can analyze diverse medical data like images, genomic data, and clinical records, making it easier for healthcare organizations to develop custom AI technologies.

Denmark launches AI supercomputer ‘Gefion’

Jensen Huang, Nadia Carlsten and King Frederik X of Denmark connect Gefion’s cables. [Novo Nordisk Foundation]

Source: Novo Nordisk Foundation (October 23, 2024)

Denmark has unveiled its first AI supercomputer, named “Gefion,” in a landmark public-private partnership between the Novo Nordisk Foundation and the Export and Investment Fund of Denmark (EIFO). The NVIDIA DGX SuperPOD system, powered by 1,528 NVIDIA H100 Tensor Core GPUs, was officially launched by King Frederik X of Denmark alongside NVIDIA CEO Jensen Huang. Built in just six months, the DKK 700 million ($100 million) facility will initially support six pilot projects ranging from quantum computing to healthcare and climate research. The supercomputer, housed in a sustainable data center running on 100% renewable energy, represents Denmark’s push to establish domestic AI infrastructure for accelerating innovation in life sciences, climate research, and quantum computing.

Google’s code generation hits new milestone

Source: The Verge (October 29, 2024)

While technically news from late October, AI observers continue to discuss the ramifications of the fact that more than one quarter of new code at Google is now generated with AI tools. In August, Matt Garman, AWS’ new chief executive, also acknowledged the quickly expanding role of genAI on development, predicting that “most developers are not coding” within two years. Rather than saying that AI would automate the profession entirely in a matter of years, Garman opined that developers would need to harness emerging tech to improve the impact of the code creation they oversee, according to an ITPro article.

Physical Intelligence wins major backing from Bezos, OpenAI

Source: Reuters (November 4, 2024)

A laundry robot [Physical Intelligence]

Physical Intelligence, a robotics software startup, raised $400 million in early-stage funding, reaching a $2 billion valuation. The funding round included prominent investors like Jeff Bezos and OpenAI. The company aims to create universal software applicable to any robot, eliminating the need for task-specific programming. This development occurs amidst a surge in AI investment, with tech giants pouring billions into the field. Physical Intelligence recently showcased its software, pi-zero, enabling robots to perform tasks like folding laundry and handling food.

TSMC restricts AI chip production for China

Source: Financial Times (November 7, 2024)

Starting Monday, TSMC will cease production of advanced AI chips for Chinese customers at process nodes of 7 nanometers or smaller. This decision aligns with ongoing U.S. efforts to curb China’s access to cutting-edge semiconductor technologies. TSMC has informed Chinese clients that future supplies of such semiconductors would require an approval process.

OpenAI wins initial victory in copyright case

OpenAI office shot [OpenAI]

Source: Reuters (November 7, 2024)

OpenAI has won an early victory in a copyright lawsuit filed by news outlets regarding AI training data. On November 7, 2024, U.S. District Judge Colleen McMahon dismissed a copyright lawsuit filed by news outlets Raw Story and AlterNet against OpenAI.

Oxford Nanopore announces UK government partnership

Source: Oxford Nanopore (November 5, 2024)

Oxford Nanopore has established a landmark partnership with the UK government to advance genomics-driven healthcare innovation, aiming to transform medical research and patient care capabilities across the United Kingdom. The collaboration seeks to translate genomic discoveries into clinical applications, focusing on cancer, rare genetic disorders, and infectious diseases.

Evogene forges Google Cloud alliance for AI drug discovery

Source: Evogene (October 31, 2024)

Evogene Ltd., a computational biology company, has hooked up with Google Cloud to develop an AI-driven foundation model that aims to improve the design of small molecules for applications in drug development, sustainable agriculture, and other life science sectors

AI debate methods show promise for truth finding

Source: Quanta Magazine (November 8, 2024)

One of the biggest pitfalls of off-the-shelf genAI models are hallucinations — basically made up AI facts that frequently sound believable but are, well, wrong. New research suggests that implementing debate mechanisms in AI systems could help models converge on truthful answers. This approach mimics human deliberative processes and shows potential for improving AI accuracy and reliability. The finding is not entirely new. A February 2024 paper on Arxiv titled “Debating with More Persuasive LLMs Leads to More Truthful Answers” also found that when large language models (LLMs) participate in debates, they are more likely to produce truthful responses.

Nous Research launches ‘unrestricted’ AI chatbot

Source: VentureBeat (November 7, 2024)

The ‘unrestricted’ AI group Nous Research has released its first chatbot, featuring some guardrails despite their open approach to AI development. The chatbot would be more open than chatbots from “more buttoned up corporate outfits” from the likes of OpenAI, Anthropic, Google, Meta. The chatbot does not appear to be completely unrestricted, however, with VentureBeat noting it refused answering questions about synthesizing illegal narcotics, for instance.

Anthropic teams up with Amazon, Palantir for defense AI project

Source: Inc. (November 8, 2024)
Anthropic, Amazon, and Palantir have announced a partnership to bring AI capabilities to U.S. defense and intelligence agencies. The collaboration will integrate Anthropic’s Claude 3 and 3.5 models into Palantir’s AI Platform (AIP), running on Amazon Web Services, with clearance to process information classified up to the “secret” level. The move represents a significant step for Anthropic, which has previously emphasized AI safety, into defense applications. According to Anthropic’s head of sales and partnerships Kate Earle Jensen, the partnership aims to enhance analytical capabilities and operational efficiencies in vital government operations.

Meta’s Llama AI sparks military AI race

Sources: The Verge (November 4, 2024), Reuters (November 1, 2024)
Similar to Anthropic, Meta is also allowing U.S. government agencies to use its Llama AI model for national security applications, just days after Reuters revealed Chinese military researchers had already developed an AI tool called ChatBIT based on an earlier version of Llama. Meta is also partnering with U.S. defense contractors including Lockheed Martin and Oracle for authorized military use, the company stated that any use by the People’s Liberation Army was “unauthorized and contrary to our acceptable use policy.” Meta emphasized that in “the global competition on AI,” China is already investing more than a trillion dollars to surpass the U.S.

AI technology promises billions in North Sea decommissioning savings

Source: The Times (November 4, 2024)

An AI system developed by Rahd AI could reduce North Sea oil and gas decommissioning costs by up to 35%, potentially saving the UK Treasury over £8 billion, according to an article in The Times. The technology, which analyzes data from 15,000 wells worldwide to optimize decommissioning processes, has already won support from major energy companies including Equinor and TotalEnergies. A pilot program in Australia demonstrated 10% cost savings, with Rahd AI projecting they can achieve 35% savings by 2027, eight years ahead of UK government targets.

Research assistance: Frederic Célerse, Ph.D., Research Scientist in AI for Chemistry, Ecole polytechnique fédérale de Lausanne.