AIR#176 - Claude’s Secrets Revealed & AI Projects at Risk 📉
Hey there!
Here's the latest AI news for today. Enjoy!
Today's top stories
🔥 Anthropic publishes the 'system prompts' that make Claude tick
Anthropic reveals Claude's system prompts, promoting transparency while outlining model limitations and desired traits.
🔥 Cerebras Inference: AI at Instant Speed
Cerebras Inference offers the fastest AI inference, achieving 1,800 tokens/sec, 20x faster than GPUs, with unmatched pricing and accuracy.
🔥 80% of AI Projects Crash and Burn, Billions Wasted Says Rand Report
A RAND report reveals 80% of AI projects fail, wasting billions due to poor leadership, data issues, and lack of infrastructure.
🔥 Splatt3R: Zero-Shot Gaussian Splatting from Uncalibrated Image Pairs
Splatt3R enables 3D reconstruction from uncalibrated image pairs, predicting Gaussian splats without camera parameters.
🔥 Cerebras launches inference for Llama 3.1; benchmarked at 1846 tokens/s on 8B
Cerebras sets a new AI inference speed record with Llama 3.1, achieving 1,850 tokens/s on 8B and 446 tokens/s on 70B models.
🔥 Unlocking the Pixel 9 bootloader breaks some Pixel AI apps
Unlocking the Pixel 9 bootloader disables key AI features, limiting functionality despite offering more device control.
DisTrO – a family of low latency distributed optimizers
DisTrO is a low latency distributed optimizer that significantly reduces inter-GPU communication for efficient internet-based training.
Smaller Gemini 1.5 Flash-8B, stronger Gemini 1.5 Pro, improved Gemini 1.5 Flash
Logan Kilpatrick announces three new experimental Gemini 1.5 models: Flash-8B, Pro, and improved Flash.
75% of founders in the Y Combinator S24 cohort are working on AI startups
75% of Y Combinator's latest cohort are launching AI startups, reflecting the tech's dominance in Silicon Valley.
Congress shouldn't pass AI protections only rich people can use
Congress must prioritize AI protections for all, not just the wealthy, ensuring consent and support for marginalized communities.
Why Did Databricks Open-Source Unity Catalog?
Databricks open-sourced Unity Catalog, signaling a shift towards open lakehouses and enhancing competition in the analytics market.
Cerebras Launches the Fastest AI Inference
Cerebras unveils the fastest AI inference technology, enhancing performance and efficiency in AI applications.
Using Multimodal LLMs to Understand UI Elements on Websites
Multimodal LLMs enhance understanding of UI elements on websites by combining visual and textual data for better context and accuracy.
Show HN: Open-source study to measure end user satisfaction levels with LLMs
An open-source study invites users to rate AI responses, aiming to measure satisfaction with language models.
Onsa.ai – intelligent sales meeting prep and post-call analysis
Onsa.ai enhances B2B sales with AI agents for meeting prep, post-call analysis, and personalized outreach.
Show HN: Netflix for AI movies – built with Cursor in 2 days
Bento creates a Netflix-like platform for AI-generated movies, developed in just two days using Cursor.
Calendar tool Clockwise adds new AI-powered interface called Prism
Clockwise introduces Prism, an AI-powered calendar tool for smart scheduling and conflict resolution, enhancing team productivity.
cerebras: 450 tokens/sec llama 3.1 70B
Cerebras claims its WSE-3 chips achieve 450 tokens/sec for Llama 3.1, outperforming competitors by leveraging SRAM over HBM.
Hobbyists discover how to insert custom fonts into AI-generated images
Hobbyists can now insert custom fonts into AI-generated images using the Flux model, enhancing text accuracy in visuals.
PSA: Some Pixel AI features may not work if you unlock your device's bootloader
Unlocking a Pixel device's bootloader disables key AI features like Call Notes and Pixel Screenshots; locking it restores functionality.
Elon Musk supports California bill to regulate AI
Elon Musk backs California's AI regulation bill, diverging from Silicon Valley peers, citing safety concerns amid rising AI risks.
Show HN: Brushless – An AI-first alternative to Illustrator
Brushless.ai offers an AI-driven platform for creating customizable, editable vector graphics quickly and affordably.
Unit Economics of LLM APIs
OpenAI's API likely remains profitable with high margins, but competition may force price drops or lead to a single dominant model.
Burn 0.14.0 Released: The First Rust-Native Deep Learning Framework
Burn 0.14.0 introduces CubeCL for GPU support in Rust, enhancing ONNX compatibility and adding new tensor features.
OpenAI Shows 'Strawberry' AI to the Feds and Uses It to Develop 'Orion'
OpenAI presents 'Strawberry' AI to federal officials, leveraging it for the development of their new project, 'Orion.'
Open-weight Llama3-based LLMs for 5 African languages
Jacaranda launches open-source LLMs in five African languages to enhance maternal health support across the continent.
Cerebras reaches 1800 tokens/s for 8B Llama3.1
Cerebras achieves 1,800 tokens/s for LLaMA 3.1 on its wafer-scale chip, promising faster, efficient AI inference and challenging Nvidia.
Stephen Wolfram thinks we need philosophers working on big questions around AI
Stephen Wolfram advocates for philosophers to address critical ethical questions in AI development, emphasizing the need for deeper thinking.
Cerebras Inference
Cerebras Inference offers Llama3.1-70B at 450 tokens/s, 20x faster than GPUs, and at a fifth of hyperscaler prices.