Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI – Page 652

Intel to Build Next-Generation Chips Using ‘Glass Substrate’

Year 2023 face_with_colon_three

The new manufacturing method deals with the packaging substrate, the material to which chip dies are bonded. Intel and others have long used plastic (also known as organic) substrates, but the material can shrink or warp during the chip-making process, leading to defects.

Intel notes the warping risk grows as more silicon is placed on the substrate. “As the demand for data-centric, AI-centric compute increases, we are seeing an increasing amount of silicon being packed onto the package substrate, which organic packages have come to some kind of limitation in terms of handling it,” Manepalli added.

The company found a solution in glass, a homogenous substance that can remain rigid under a higher chip load. “Compared to today’s organic substrates, glass offers distinctive properties such as ultra-low flatness and better thermal and mechanical stability, resulting in much higher interconnect density in a substrate,” Intel said in its announcement.

Apple researchers develop AI that can ‘see’ and understand screen context

Apple researchers have developed a new artificial intelligence system that can understand ambiguous references to on-screen entities as well as conversational and background context, enabling more natural interactions with voice assistants, according to a paper published on Friday.

The system, called ReALM (Reference Resolution As Language Modeling), leverages large language models to convert the complex task of reference resolution — including understanding references to visual elements on a screen — into a pure language modeling problem. This allows ReALM to achieve substantial performance gains compared to existing methods.

Paper page — Aurora-M: The First Open Source Multilingual Language Model Red-teamed according to the U.S. Executive Order

Aurora-M

The First Open Source Multilingual Language Model Red-teamed according to the U.S. Executive Order https://huggingface.co/papers/2404.

Pretrained language models underpin several AI applications, but their high computational cost for training limits accessibility.

Join the discussion on this paper page.

Identifying inflammation is at the heart of the matter

By contrast, information on coronary inflammation can provide crucial early warning signs of a cardiac event. Yet traditional diagnostic methods of measuring inflammation are not specific for cardiovascular diseases. Inflammation is invisible to CT scans, for instance. And biomarkers such as hsCRP (High-sensitivity C-reactive Protein) measure systemic inflammation, rather than cardiovascular inflammation, so the test may show up high in the case of inflammation driven by non-heart organs.

CaRi-Heart leverages AI tech to detect and quantify coronary inflammation, giving it an edge over traditional diagnostic methods. Cheng explains that while it is important to find patients who already have significantly narrowed coronary arteries, and obviously need immediate treatment, cardiologists often end up archiving many cases of patients with no visible signs of disease but who potentially have high coronary inflammation. This inflammation, driven by cholesterol, or smoking, or diabetes and other risk factors, ultimately causes the wall of the artery to become thickened and narrowed.

Caristo’s CaRi-Heart technology is a non-invasive cloud-based solution that utilizes AI to analyse CT scans, overcoming the limitations of traditional diagnostic methods, offering a more sensitive and specific approach to detecting and quantifying coronary inflammation, says Cheng. CaRi-Heart is the only commercially available technology that can detect and measure coronary inflammation on routine cardiac CT scans, and it has been cleared for clinical use in the UK, EU and Australia.

UCL spinout bags £10M to make AI ‘super brains’ for 100x faster LLM training

Oriole Networks, a UCL spinout, has raised £10mn in seed funding to build AI “super brains” that promise to accelerate the training of Large Language Models (LLMs).

Founded in 2023 by UCL scientists, the startup has developed a new method that harnesses the power of light to connect thousands of AI chips. This results in a network of chips, where the power of each individual GPU is combined to form a “super brain.”

According to James Regan, CEO at Oriole Networks, this “enables the direct connection of a very large number of nodes enabling it to function as a single machine.”

/* */