TLDR AI 2024-07-10

Selling AI to enterprise means getting serious about security and compliance — usually involving features like single sign-on (SSO), Directory Sync (SCIM), RBAC, and Audit Logs.

With WorkOS, you can become Enterprise Ready on day 1.

→ Modular features (SSO, Directory Sync) makes it easy for you to proactively drive enterprise growth

→ Open-source client libraries allow implementations to finish in minutes

→ Transparent pricing and a FREE tier of up to 1 million MAUs

WorkOS is trusted by the hottest AI startups: Perplexity, copy.ai, AI21Labs, Jasper, and many others. Deploy enterprise plans today with zero fuss.

Quora's Poe has introduced a feature called Previews, enabling users to create interactive apps such as data visualizations and games directly through chatbot interactions.

VC firm Andreessen Horowitz has secured thousands of AI chips, including Nvidia H100 GPUs, to dole out to its AI portfolio companies in exchange for equity.

xAI has ended its deal with Oracle. It will build its own datacenter after Grok 2 training finishes. The company originally contracted for 24k H100s from Oracle.

Video Self-Training with augmented Reasoning (Video-STaR) is a new approach for improving Large Vision Language Models (LVLMs).

System 2 models are those that use something like Chain of Thought to spend more test time compute and improve reasoning. It turns out that you can distill this behavior into a System 1 model that runs faster with similar accuracy.

A new RNN variant that outperforms Mamba on a number of tasks. Importantly, the update function is itself an ML model - which enables long contexts and in-context learning.

A universal semantic layer makes it much easier to run both BI and AI workloads on your data warehouse. AtScale, now available as a Snowflake Native App in the Snowflake Marketplace, allows you to define and consume a universal semantic layer directly within your Snowflake account.

See how it works: register for the live session on July 24 →

MobileLLM optimizes sub-billion parameter language models for on-device use cases.

Researchers have developed a new framework that combines a large language model with reinforcement learning to improve traffic signal control systems.

GraphRAG is a way to explore unstructured text using the power of language models. The quick start runs on Azure, making it easy to spin up.

The rapid adoption of ChatGPT, which reached 100 million users in two months, stands in contrast to historical tech developments like the iPhone and e-commerce, which took years to gain traction. Despite the initial hype, many users have not found long-term utility with ChatGPT and enterprise deployment of large language models remains limited, indicating a need for further development to achieve meaningful product-market fit and sustained value.

Tinygrad is a toolkit for training deep learning models. This set of notes is a great introduction for AI compilers and prepares one for a deep dive into Tinygrad internals.

Goldman Sachs released a critical 31-page report titled "Gen AI: Too Much Spend, Too Little Benefit?", arguing that generative AI's productivity benefits and returns are significantly limited and that its power demands will drastically increase utility spending. The report highlights doubts about AI's ability to transform industries, pointing out high costs, power grid challenges, and lack of clear productivity gains or significant revenue generation. It suggests a potentially bleak future for the technology without major breakthroughs.

OccSora is a diffusion-based 4D occupancy generation model designed to improve long-term temporal evolutions.

Krut AI is an AI co-pilot built for e-commerce that helps with product shots and content creation.

This project investigates the journey towards Artificial General Intelligence (AGI), which aims to perform diverse real-world tasks with human-like efficiency.