Ahead of its formal launch, Apple’s new AI offering, Apple Intelligence, is receiving mixed reviews. While Apple touts the iPhone 16 as being built for AI “from the ground up,” early tests have revealed a less-than-polished experience. Vox calls the tech “magically mediocre,” while AI Supremacy describes it as “late, unfinished & clumsy.” The Washington Post even went so far as to title their review “Apple’s iPhone 16 AI is useful so far, except when it’s bonkers,” citing instances of inaccurate results and misinterpretations. As Apple prepares for a wider rollout in October, the company faces the challenge of meeting consumer expectations in an increasingly competitive AI landscape.

Reflection 70B’s promising debut overshadowed by fraud allegations

Source: VentureBeat (September 10, 2024) & Tom’s Guide (September 11, 2024)

The Reflection 70B AI model, initially touted as the “world’s top open-source model” by HyperWrite CEO Matt Shumer, has come under intense scrutiny. Independent researchers have failed to replicate its claimed benchmark performance, leading to accusations of fraud. Some reviewers suggest the model might be a “wrapper” around Anthropic’s Claude 3.5, rather than an improved version of Meta’s Llama 3.1. Shumer has since apologized, admitting he “got ahead of himself.” Critics demand further transparency and a thorough investigation into the discrepancies.

Snowflake’s profitability fuels expansion, intensifies competition with Databricks amidst slowing growth

Source: LinkedIn Post by Chad Peets (September 11, 2024), SeekingAlpha (September 9, 2024)

Snowflake’s self-funded growth model contrasts with Databricks’ reliance on continuous fundraising. From FY22 to FY25, Databricks’ projected revenue growth from $625 million to $2.4 billion has required $3.1 billion in investor funding. Snowflake’s revenue is expected to grow from $1.2 billion to $3.5 billion, generating over $2 billion in free cash flow. However, Snowflake’s growth is decelerating, with projections of mid to low 20% growth over the next few years. In response, Snowflake is investing heavily in AI capabilities, evidenced by nine new product announcements in Q2 2024. Competition in the cloud data platform space is intensifying, with hyperscalers like AWS, Azure, and GCP vying for market share.

A race to reproduce AlphaFold 3 with open source alternatives

Source: LinkedIn Post by Charlie Harris (September 12, 2024) and accompanying Substack article (September 10, 2024)

Following the release of DeepMind’s AlphaFold 3 in May 2024 – which lacked open-source code and trained model weights – several groups are working to reproduce and open-source the model, which can predict a range of biological interactions and structures. The research push aims to make the technology more accessible to the broader scientific community. Charlie Harris, a Venture Fellow at IQ Capital, provides a comprehensive picture of the notable open-source AlphaFold 3 alternatives in his Substack article. Efforts include HelixFold3 by Baidu, Ligo Biosciences’ implementation, and Chai-1 by Chai Discovery. Harris also notes that DeepMind has yet to release the official AlphaFold 3 code and weights, despite indicating plans to do so within six months of the paper’s release. Nature has also reported on the race to develop an open source AlphaFold 3 alternative.

If you are unfamiliar with AlphaFold3 or just want to learn more about it, check out the great overview from the Boston Protein Design and Modeling Club.

Code Ocean 3.0 cloud-based computational research platform boasts native ML capabilities

Source: LinkedIn Post from Code Ocean CEO Simon Adar (September 12, 2024) and accompanying blog post (September 12, 2024)

Code Ocean 3.0, the latest iteration of the company’s computational science platform, introduces native support for machine learning (ML) models through the integration of MLflow, aiming to streamline the ML model lifecycle for computational biologists and bioinformaticians, particularly in drug discovery and AI-driven research. The MLflow integration allows users to track model development, access model information, and manage runs via a dedicated MLflow dashboard. Additionally, the Code Ocean Lineage Graph automatically records the complete history of code, data, and environments used in model creation, enhancing reproducibility and transparency. The platform also offers no-code model validation through Inference Capsules, integration with Nextflow-based pipelines for automated workflows, and data flexibility for training and validating models. Code Ocean provides a secure model registry and includes industry-standard security features, complying with regulations like HIPAA, GDPR, and ISO 27001. More details can be found in the company’s blog post.

Roche expands digital pathology platform with AI integrations

Source: LinkedIn post from Andrii Buvailo, Ph.D. (September 11, 2024) and accompanying BioPharmaTrend article (September 11, 2024)

Roche has announced the integration of more than 20 AI algorithms from eight new collaborators into its navify Digital Pathology platform, bolstering its Digital Pathology Open Environment. This initiative aims to improve cancer diagnosis and research by providing pathologists with AI-powered tools for more accurate tissue sample analysis. The integrated AI algorithms cover a wide spectrum of cancer types and diagnostic tasks. For prostate cancer, Deep Bio offers detection, grading, and tumor quantification, while Qritive provides screening and grading capabilities. In breast cancer diagnostics, DiaDeep excels in biomarker quantification, Mindpeak handles biomarkers and PD-L1 analysis, and Stratipath focuses on risk profiling. Lung cancer analysis is addressed by Lunit, which performs tumor proportion score analysis for NSCLC, and Mindpeak, which conducts PD-L1 analysis. For colorectal cancer, Owkin detects microsatellite stability, while Sonrai Analytics identifies microsatellite instability status. Mindpeak extends its PD-L1 analysis capabilities to gastric, esophageal, and bladder cancers. Additionally, Qritive offers analysis for lymph node metastasis. This comprehensive suite of AI tools demonstrates the breadth and depth of cancer diagnostics covered by these innovative algorithms. More details can be found in the BioPharmaTrend article.

China looking to unseat U.S. leadership in genAI

Source: CNBC (September 11, 2024)

While U.S. companies like OpenAI, Anthropic, and Google have dominated headlines in generative AI, Chinese tech giants are aiming to outdo them. These companies, including Baidu, Alibaba, Tencent, Huawei, and ByteDance, have launched a range of AI models with diverse capabilities. Kai-Fu Lee, a prominent figure in AI who has worked for Microsoft, Google and Apple, launched a Beijing-based startup called 01.AI in March 2023 that has achieved a $1 billion valuation as of November 2023. The company reached unicorn status in just eight months after its founding. Baidu’s Ernie Bot, powered by the Ernie 4.0 model, boasts a significant user base of 300 million and aims to compete directly with OpenAI’s ChatGPT. Alibaba has released Tongyi Qianwen (Qwen), a suite of models covering diverse tasks such as content creation, problem-solving, and audio understanding. Some Qwen models are open-sourced, enabling developer access. Tencent’s Hunyuan model, available through its cloud platform, demonstrates strong Chinese language processing capabilities and supports functions like image creation, text recognition, and chatbot integration with its popular messaging app, WeChat. Huawei has taken a more industry-focused approach with its Pangu models, offering solutions for sectors like meteorology (typhoon trajectory prediction) and manufacturing, alongside generative features for code and avatar creation. ByteDance, though entering the race later, has introduced Doubao, a competitively priced model emphasizing voice and code generation. More details are in the CNBC article.

Related Content

R&D World Index: Tech Industry Shifts Learn about recent developments in the tech industry, including IBM’s closure of its China R&D arm and the boom in AI data centers.

White House hosts AI leaders to address energy demands of emerging technology

Source: CNN (September 12, 2024)

Senior White House officials met with prominent AI industry leaders, including Sam Altman (OpenAI), Ruth Porat (Google), and Dario Amodei (Anthropic), to discuss the growing energy demands of artificial intelligence and its potential strain on U.S. infrastructure. The meeting, which also included Energy Secretary Jennifer Granholm, Commerce Secretary Gina Raimondo, and representatives from Microsoft, marks the first time the White House has convened with tech executives to address the energy challenges resulting from the AI boom.

The International Energy Agency estimates that a single ChatGPT request consumes ten times more electricity than a typical Google search. Projections indicate that AI-related power demand from data centers could surge by 160% by 2030, according to Goldman Sachs. Sam Altman highlighted the need for investment in AI infrastructure, including power generation, data centers, and semiconductor manufacturing. Meanwhile, the White House underscored its commitment to responsible AI development and data center construction within the United States. The meeting follows a 2023 initiative where the Biden administration secured pledges from AI companies to prioritize safety and transparency in AI system development and deployment.

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