OpenAI Announces New Enterprise AI Features (4 minute read)
OpenAI has announced new enterprise-grade features for its API customers, including enhanced security measures, an upgraded Assistants API, a new Projects feature for granular access control, and cost management tools. These updates demonstrate OpenAI's focus on offering a more "plug and play" experience for enterprises, countering the rise of competitors like Meta's Llama 3 and open models from Mistral.
π§
Research & Innovation
Phi 3 (17 minute read)
Phi 3 is a series of models, 3B-14B in size, which perform exceptionally well on modern benchmarks. The 3B model claims to outperform the original ChatGPT model. Weights have been released. There is a variant available with a 128k context length.
Instruction hierarchy (17 minute read)
OpenAI published research on giving system prompts stronger weighting, which dramatically improves model robustness to jailbreaks and adversarial attacks.
π¨βπ»
Engineering & Resources
Exploring LLaMA3's Performance in Low-Bit Quantization (GitHub Repo)
Meta's LLaMA3, a leading large language model, is being tested for its efficiency in low-bit scenarios, often essential in systems with limited resources. This study, available on GitHub and Hugging Face, aims to refine and improve quantization strategies for future large language models.
Instructor (GitHub Repo)
Instructor is a Python library that makes it easy to work with structured outputs from large language models.
How does ChatGPT work? As explained by the ChatGPT team (6 minute read)
In this article, OpenAI's Evan Morikawa provides insights into ChatGPT's inner workings, covering input text processing and tokenization to prediction sampling using large language models. ChatGPT operates by turning tokens into numerical vectors (embeddings), multiplying them by a weight matrix of billions, and selecting the most probable next word. The tech is grounded in extensive pretraining to predict text based on vast internet data.
Self-Reasoning Tokens, teaching models to think ahead (4 minute read)
Recent experiments introduced "Reasoning Tokens" to improve the thinking process of language models like GPT-2, encouraging them to make calculations for future tokens. Early results show a 35% decrease in loss, indicating the models can indeed learn to anticipate future information. This approach could enhance the ability of language models to plan and reason in a self-supervised manner, potentially reducing the need for step-by-step explanations.
Los Angeles is using AI in a pilot program to try to predict homelessness and allocate aid (7 minute read)
The Los Angeles County Department of Health Services is using predictive AI to prevent homelessness, identifying at-risk individuals to provide aid and successfully keeping 86% of participants housed. The program, initiated in 2021, has assisted nearly 800 households with over $4,000 in support each. Despite concerns over privacy and ethics, the AI initiative shows promise in addressing California's climbing homelessness crisis.