TLDR AI 2024-07-17

The bad news is that threat actors are already using AI to collect user information for robocalls, imposter calls, and other types of fraud. Real data + convincing voice impersonation can trip up almost anyone, and deep fakes are only getting better.

The best way to protect yourself? Get your personal information OFF data broker databases. If you're too busy to send hundreds of deletion requests, Incogni will do it for you.

Use this link to get 60% off by using code TLDRAI.

❌ Delete your personal data today

Eureka Labs is a new education-focused AI company designed to change how we learn new things.

Whistleblowers have accused OpenAI of placing illegal restrictions on employee communication with government regulators, according to a letter obtained by The Washington Post, which urges the SEC to investigate OpenAI's severance, non-disparagement, and non-disclosure agreements for violating employees' rights to whistleblower incentives and compensation.

Several major AI companies, including Anthropic, Nvidia, Apple, and Salesforce, used subtitles from 173,536 YouTube videos across 48,000 channels to train AI models, despite YouTube's rules against unauthorized data harvesting. This has sparked backlash from content creators, who argue that their work has been exploited without consent or compensation, raising concerns about AI's impact on the creative industry and the ethics of using such training data.

HumanEval has long been the goal for coding models. It has essentially been solved. This benchmark is the next evolution for challenging science programming problems.

Mistral has trained a Mamba 2 code model with extreme performance in long contexts and 75% on HumanEval.

Mistral's newest math model performs well on various benchmarks.

Researchers have developed a Physics-Informed Diffusion (PID) model to improve the translation of RGB images to infrared images by incorporating physical laws into the process.

Run your own AI cluster at home on everyday devices.

Decoupled Refusal Training (DeRTa) is a new approach that improves safety tuning in large language models by addressing refusal position bias.

A great suite of small language models from Hugging Face that outperform many larger models. The key comes in the careful curation of the dataset.

Great post by Yi Tay of Reka and Google on encoders, prefixlm, denoising objectives, and more modern tricks for language modeling.

Redis has unveiled new data on the top vector databases comparing latency and throughput using the industry-standard Qdrant framework. Key findings include Redis outperforming competitors by 62% for low complexity datasets and by 21% for high-dimensional datasets, achieving significantly higher queries per second and lower latency than Qdrant, Milvus, and Weaviate.

Gray Swan AI is an AI safety and security company that specializes in building tools to help companies assess the risks of their AI systems and safeguard their AI deployments from harmful use.

Project Atlas generates complex business automation agents step-by-step, then it outputs a no-code drag-drop graph for you to edit.

Anthropic launched the Claude Android app to expand its AI chatbot's reach, aiming to attract users from ChatGPT by offering features like real-time image analysis, language translation, and syncing across devices.