Lifeboat Foundation

The statement that may not be obvious is that the sleeping giant, Google has woken up, and they are iterating on a pace that will smash GPT-4 total pre-training FLOPS by 5x before the end of the year. The path is clear to 100x by the end of next year given their current infrastructure buildout. Whether Google has the stomach to put these models out publicly without neutering their creativity or their existing business model is a different discussion.

Today we want to discuss Google’s training systems for Gemini, the iteration velocity for Gemini models, Google’s Viperfish (TPUv5) ramp, Google’s competitiveness going forward versus the other frontier labs, and a crowd we are dubbing the GPU-Poor.

Access to compute is a bimodal distribution. There are a handful of firms with 20k+ A/H100 GPUs, and individual researchers can access 100s or 1,000s of GPUs for pet projects. The chief among these are researchers at OpenAI, Google, Anthropic, Inflection, X, and Meta, who will have the highest ratios of compute resources to researchers. A few of the firms above as well as multiple Chinese firms will 100k+ by the end of next year, although we are unsure of the ratio of researchers in China, only the GPU volumes.

To discover how light interacts with molecules, the first step is to follow electron dynamics, which evolve at the attosecond timescale. The dynamics of this first step have been called charge migration (CM). CM plays a fundamental role in chemical reactions and biological functions associated with light–matter interaction. For years, visualizing CM at the natural timescale of electrons has been a formidable challenge in ultrafast science due to the ultrafine spatial (angstrom) and ultrafast temporal (attosecond) resolution required.

Experimentally, the sensitive dependence of CM on molecular orbitals and orientations has made the CM dynamics complex and difficult to trace. There are still some open questions about molecular CM that remain unclear. One of the most fundamental questions: how fast does the charge migrate in molecules? Although molecular CM has been extensively studied theoretically in the last decade by using time-dependent quantum chemistry packages, a real measurement of the CM speed has remained unattainable, due to the extreme challenge.

As reported in Advanced Photonics, a research team from Huazhong University of Science and Technology (HUST), in cooperation with theoretical teams from Kansas State University and University of Connecticut, recently proposed a high harmonic spectroscopy (HHS) method for measuring the CM speed in a carbon-chain molecule, butadiyne (C₄H₂).

Scientists already have their ways of coaxing human cells into new forms, using a special concoction of chemicals to nudge humble skin cells into malleable tissues known as induced pluripotent stem cells.

In spite of this new lease on life, these particular cells still retain a few genetic reminders of their time as a fully developed tissue, affecting their use as a blank slate.

Now an international team of researchers has gone one better: finding a new way of wiping a cell’s memory clean so it can be better reprogrammed as a stem cell.

A new VTOL firm promises to make flying cars a reality, but unlike many similar firms, this one has a NASA-proven design at its heart.

Simple heterojunction combines many functions in a single component.

Stem cells in organoids self-organize into tissue patterns with unknown mechanisms. Here, we use skin organoids to analyze this process. Cell behavior videos show that the morphological transformation from multiple spheroidal units with morphogenesis competence (CMU) to planar skin is characterized by two abrupt cell motility–increasing events before calming down. The self-organizing processes are controlled by a morphogenetic module composed of molecular sensors, modulators, and executers. Increasing dermal stiffness provides the initial driving force (driver) which activates Yap1 (sensor) in epidermal cysts. Notch signaling (modulator 1) in epidermal cyst tunes the threshold of Yap1 activation. Activated Yap1 induces Wnts and MMPs (epidermal executers) in basal cells to facilitate cellular flows, allowing epidermal cells to protrude out from the CMU. Dermal cell–expressed Rock (dermal executer) generates a stiff force bridge between two CMU and accelerates tissue mixing via activating Laminin and β1-integrin. Thus, this self-organizing coalescence process is controlled by a mechano-chemical circuit. Beyond skin, self-organization in organoids may use similar mechano-chemical circuit structures.

Aude Oliva is a prominent Cognitive and Computer Scientist directing the MIT Computational Perception and Cognition group at CSAIL while also leading the MIT-IBM Watson AI Lab and co-leading the MIT AI Hardware Program. With research spanning computational neuroscience, cognition, and computer vision, she pioneers the integration of human perception and machine recognition. Her contributions extend across academia, industry, and research, making her a distinguished figure at MIT.

Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:

Continue reading “Building Tools To Learn Human Brain Processes” »

Construction is set to break ground by the end of this year, and the company expects to move into the new space by the end of 2024. The production facility for semiconductor quartz will include a clean room, high-purity cleaning system and allow them to expand an automation component of their business that they’ve been capitalizing on for years.

“We knew that our customers all over the world were expanding at a rate we couldn’t keep up with,” said Scott Lingren, SXT’s managing director and U.S. chairman. “As you see all these expansions from Samsung in Taylor to Texas Instruments Inc. in the Dallas area to all over the world … we just have to keep up.”

SXT – which is headquartered in the Netherlands and owned by the privately-held Schunk Group in Germany – supplies semiconductor manufacturers around the world, like Samsung, which has had a presence in Central Texas for decades and is potentially adding to its existing Austin campus and its new site in Taylor. Other major players in the industry include Taiwan Semiconductor Manufacturing Co., which is expanding in Arizona, and Intel Corp., which is expanding to Ohio.

GM’s self-driving robotaxi unit, Cruise, is poised to deploy a small fleet of autonomous cars in Seattle, Washington. The initial deployment of the vehicles will reportedly include safety drivers.

Cruise’s deployment in the city will begin in sections of downtown Seattle, Capitol Hill, Queen Anne, Fremont, the University District, and West Seattle. From this, Cruise would be able to collect valuable real-world data. The robotaxi provider would be joining other self-driving units that are also operating in Seattle today, such as Zoox.

In a comment to Geek Wire, Cruise executive VP of engineering Mohamed Elshenawy noted that Seattle would be a great testing ground for the company’s self-driving robotaxis. The city, after all, features hilly areas and inclement weather conditions, which should present Cruise with some interesting edge cases and challenges.