TLDR AI 2024-06-28

Between Google Gemini and the latest release of OpenAI's GPT-4o, online search and discovery is fast becoming a multimodal affair. Users are looking beyond text and expecting the same high quality results for image search.

With Algolia developer teams at the forefront of search innovation, this ebook explains how Algolia is adding image understanding to its lightning-fast keyword and vector search capabilities.

17,000+ businesses trust Algolia to empower their user experience with fast, reliable, and scalable AI search solutions. Learn more at algolia.com

Instagram's new "AI Studio" lets creators develop AI chatbot versions of themselves. It is currently rolling out as an early test in the US.

Snap has released the GenAI suite in Lens Studio 5.0 - this is a great advancement and is massively useful for developing AR applications.

Tech giants are shifting focus from large language models to more efficient small language models (SLMs), with Apple and Microsoft introducing models with substantially fewer parameters yet comparable, or even superior, performance in benchmarks. OpenAI's CEO suggests we're moving beyond the era of LLMs, as SLMs offer advantages like local device operation, increased accessibility for smaller entities, and potential insights into human language acquisition. Training SLMs on high-quality or "textbook-quality" data contributes to their effectiveness despite their smaller scale.

OpenAI has trained a model based on GPT-4 that can catch GPT-4 mistakes and help humans spot mistakes during the RLHF process.

Meta has released two language models that can compile code to assembly and decompile to LLVM IR. They were trained on 546 Billion tokens of high-quality data then further instruction tuned. They are able to achieve 77% of optimized assembly performance and have 45% round trip disassembly performance.

ALPBench is a tool designed to standardize benchmarks for active learning query strategies.

Need to deploy AI /ML at the edge? Deliver fast AI applications with high throughput and ultra-low latency anywhere in the world with Gcore. Automatically adjust resources based on demand (only pay for what you use), deploy and scale effortlessly, process data locally and securely. Get started or book a demo

This survey delves into generative information retrieval (GenIR), a fresh take on finding information that moves away from traditional search methods to ones that create answers on the fly.

To meet the growing demand for detailed 3D data, researchers have introduced Point-SAM, a transformer-based model for 3D segmentation.

AuraSR is a super resolution image upscaler recently open sourced by FAL. It requires a single forward pass and can upscale by 4x even over repeated applications. AuraSR works well on generated images.

The Gemma 2 9B and 27B models are now available.

MatText is a suite of benchmarking tools and datasets designed to evaluate language models' performance in materials science.

Roll up strategies depend on the ability to drive operational efficiencies to create excess cash flows. The emergence of AI possibly presents a new lever that unlocks the rollup strategy to a new degree. Are Vertical/SMBs rollups the future? This post presents two opposing viewpoints on this question.

A quick implementation of Mamba 2 that doesn't require associative scans

One of the first chaining and agents companies of the modern era, Dust raised additional funding after passing $1m ARR.

ElevenLabs has launched its first iOS app, "ElevenLabs Reader: AI Audio," designed to convert text files or web links into audio narration, allowing users to listen while on the go.