Former Google AI expert raises $100mn for biotech start-up

Inceptive’s Jakob Uszkoreit, a pioneer of the technology behind ChatGPT, attracts Nvidia funding to design ‘biological software’.

What are the 6 key types of data intermediaries and how do they relate to insurance?

This report reviews current academic and policy literature of data intermediaries

Following 6 types are presented in detail:

1)   personal information management systems (PIMS)

2)   data cooperatives

3)   data trusts

4)   data unions

5)   data marketplaces

6)   data sharing pools

The findings highlight the fragmentation and heterogeneity of the field.

Data intermediaries range from individualistic and business-oriented types to more collective and inclusive models.

Certain types do aim at facilitating economic transactions between data holders and users.

Others mainly seek to produce collective benefits or public value.

The report derives a series of take-aways regarding main obstacles faced by data intermediaries.

It also identifies lines of empirical work in this field.

Now you might ask how this is related to insurance?

Data intermediaries can be linked to insurance data.

The report mentions a platform offering harmonized multi-brand in-vehicle data from different vehicle manufacturers.

It positions its offering as a neutral intermediary that obtains data from original equipment manufacturers.

The data will be harmonised and made available to all players in the mobility sector including insurance companies.

Another application offers to its users a decentralised storage space.

It can be used to manage, retrieve and organise their personal data including financial and insurance data.

A systematic review of Explainable Artificial Intelligence (XAI) models and applications: Recent developments and future trends

One of the key impacts of Large Language Models (LLMs) on the wider field of AI and its applications is predicted to be the adoption of foundation models to deliver AI-driven innovation in different sectors of the economy. Deep Learning (DL) is gaining more attention because of its potential for accuracy, having been trained with a larger amount of data, but concerns stemming from the opacity of resultant models will slow down the delivery of benefits. XAI can be used to improve both transparency in results and performance as well as assist with better domain-specific model development by exposing reasoning gaps, data and knowledge inconsistencies, and interpretability weaknesses. Additionally, XAI may assist with domain-specific knowledge discovery and better integration of data and real-world evidence, both of which iterate back to better foundation models and regulatory performance for all. Overall, to understand the output and process of the foundation and domain-specific models, the industry shall increasingly move towards XAI. With XAI we can better integrate AI models and systems into current evidence practices for healthcare improvement and the underlying innovation ecosystems. To this end, an understanding of the technique is necessary.

Thus key endpoints for XAI are defined as follows:

1. Justification based on existing models, knowledge and facts; e.g., in the context of the GDPR “right to explanation” recital.

2. Controlling the explanation (from going wrong) – thus using the process to identify vulnerabilities and flaws in a sandbox environment, which may lead to errors in high-criticality, real-world conditions.

3. Continuous learning and improvement of models in a process driven by identifying (knowing) and addressing flaws.
 
4. Discovery of new facts for the problem, thus gaining knowledge by gathering the information, which will create new insights.

Rumor: TSMC will team up with Nvidia, Broadcom to co-develop silicon photonics to boost data transmission speeds for AI ,and should win orders as early as 2nd-half 2024.

Arm’s IPO is more than 5-times oversubscribed, citing bankers, though note the small size of the deal and large number of banks (28) involved as key reasons.

Arm execs reportedly said they raised royalties on smartphone makers, and revenue could +20% in the financial year ending March 2025.

China wants metaverse firms with ‘global influence’ and plans for up to 5 industrial clusters by 2025

China has set out an action plan for developing a domestic metaverse, which includes cultivating “three to five metaverse companies with global influence”, by 2025 through development of AI, blockchain, and VR.

According to the policy document published on Friday by five Chinese ministries led by the Ministry of Industry and Information Technology (MIIT), China aims to build “three to five industrial clusters” around the emerging technologies by 2025 with key breakthrough applications and governance for the conceptual next-generation internet made up of three-dimensional spaces.

The document, which maps out a blueprint covering 2023 to 2025, emphasises the application of the metaverse to various industries, such as home appliances, automotive and aerospace. Manufacturing industries such as steel and textiles can also adopt related technologies to optimise scheduling, material calculation, and other parts of the production process, according to the plan.

Another national action plan issued last November set a goal of growing the MR industry to 350 billion yuan (US$47.8 billion) and shipments of 25 million MR devices by 2026. Some local Chinese governments have issued their own policies to encourage metaverse development, usually with a focus on how it can support the economy and traditional industries.

Mental health startups are opting for a business model that starts in elementary school

Companies are looking to prove their value to cash-strapped school districts Modern Healthcare.

Tokenisation has the potential to bring trillions in value to blockchains by 2030 - EY.

Textbooks are All You Need II: phi-1.5 technical report

Builds a new 1.3B model named phi-1.5, with performance comparable to models 5x larger, and surpassing most non-frontier LLMs on tasks like GSM8k and HumanEval.

Scala Biodesign, a company working on a solution to improve and accelerate the development of proteins in biotech products, has excited stealth and raised $5.5 million in seed.

A neural network can smell like humans do for the first time

Digital smell is a modality that AI community has long ignored, but maybe one day useful for robot chef?

Here's how to do smell2text:

1. Collected 5,000 molecules and ask humans to label "creamy, chocolate, alcoholic, beefy, spicy, citrus", etc. This dataset is one of its kind and a huge contribution from the paper.

2. Train a graph neural network (GNN) to map the molecule to label. Each molecule is a graph of atoms described by valence, degree, hydrogen count, hybridization, formal charge, atomic number, etc.

3. The GNN predictions match well with expert humans on novel smells.

4. The embeddings give us a "Principal Odor map (POM)" that faithfully represents hierarchies and distances among odorants.

Science Paper.

Open access PDF.

35 Under 35: MIT Biotech Innovators List Revealed

35 people driving the next wave of innovation and technology in the near future.

This paper delves into the unique business models of AI startups, comparing them to conventional IT-related business models

It identifies four archetypal business model patterns for AI startups: AI-charged Product/Service Provider, AI Development Facilitator, Data Analytics Provider, and Deep Tech Researcher.

The paper also discusses three distinctive aspects that set AI startup business models apart: new value propositions enabled by AI, different roles of data in value creation, and the impact of AI technology on the overall business logic.

1. AI startups often introduce new value propositions that are not possible without AI capabilities. For example, they can offer highly personalized services or automate complex tasks.

2. Data plays a different role in AI startups. Unlike traditional IT startups where data might be a by-product, in AI startups, data is often central to the value creation process.

3. The overall business logic of AI startups is influenced by the AI technology itself. This means the way they operate, interact with customers, and even their revenue models can be fundamentally different from traditional IT startups.

Stability AI launched Stable Audio first AI product for music and sound generation.

A mini-heart in a Petri dish

A team at the Technical University of Munich (TUM) has induced stem cells to emulate the development of the human heart.

Senate Majority Leader Chuck Schumer is hosting executives from Google, Microsoft, Meta, OpenAI, IBM and more today to talk AI regulation.

The much-anticipated meeting comes as lawmakers in Washington try to get a handle on the rapidly evolving technology.

MIT researchers introduced a non-invasive technique called temporal interference to stimulate deep brain structures.

Unlike traditional deep brain stimulation (DBS), which requires surgical electrode implantation, this cutting-edge method uses high-frequency electric fields to create a low-frequency envelope field. This means that it can precisely target deep brain cells without affecting surface cells.

This type of technology is opening up a world of possibilities for more effective and less invasive treatments, revolutionizing the way we approach treatments for conditions like depression, Alzheimer's, and PTSD.

Brain chip company Motif Neurotech, gets chips in two human skulls for depression.

Just published Nature. A ViT architecture in medicine

Self-supervised AI of 2 million retina images to predict many diseases beyond the eye, such as heart attack, stroke, heart failure, Parkinson's disease.

Wayve introduced LINGO1 is a first-of-its-kind vision-language-action model for self-driving that unlocks new possibilities for enhancing the learning and explainability of AI driving models.

It can also generate continuous commentary to explain its actions.

The 2023 Global Crypto Adoption Index by Chainanalysis

Central/South Asia leads the way as 6 of the top 10 countries are in this region.