HackerNews Summary (2023-06-28 08PM)

Source: OpenOrca

Summary
OpenOrca is an open-source dataset and series of instruct-tuned language models. The team has completed the OpenOrca dataset, which consists of around 1 million FLANv2 augmented with GPT-4 completions and 3.5 million FLANv2 augmented with GPT-3.5 completions. They are currently performing full weights fine-tuning on the LLaMA-13b foundation and expect to release OpenOrca-LLaMA-13b in mid-July 2023.

Facts

💡 OpenOrca is an open-source dataset and series of instruct-tuned language models.

📚 The OpenOrca dataset consists of around 1 million FLANv2 augmented with GPT-4 completions and 3.5 million FLANv2 augmented with GPT-3.5 completions.

🔍 The dataset followed the submix and system prompt distribution outlined in the Orca paper, with a few exceptions.

🔁 Duplicates were removed from the dataset, resulting in 3.5 million instructions in the ChatGPT dataset.

🗓️ OpenOrca-LLaMA-13b is expected to be released in mid-July 2023.

💻 The team is seeking GPU compute sponsors for training OpenOrca on various platforms including Falcon, LLaMA, and MPT.

💰 Financial contribution and mentorship for OpenOrca is provided by chirper.ai.

🖥️ Compute sponsorship for LLaMA 7b is provided by preemo.io, and sponsorship for LLaMA-33b is provided by latitude.sh.

🙌 The team would like to thank the open-source AI/ML engineers who have worked on OpenOrca.

Source: The Biologist Blowing Our Minds

Summary
Developmental biologist Michael Levin and his team have experimented with various organisms to show their latent abilities. They have successfully cultivated skin cells from a frog embryo to create a new organism called a “xenobot”. Slime mold, a unicellular organism, has been studied by Levin to understand its decision-making process in response to stimuli.

Fact

🐸 Xenobots are new organisms created by cultivating skin cells from a frog embryo and allowing them to reassemble into a new entity.

🦋 Butterflies retain memories from their caterpillar stage, even though their brains turn to mush during metamorphosis.

🪱 Planarians, or flatworms, can regrow both their head and tail when amputated, and they store information about their new shape as electrical patterns.

🧠 Electrical signaling is pervasive in nature and is not limited to neurons.

🍄 Slime mold, such as Physarum, behaves by changing shape and can make decisions based on a calculus of payoff and trade-offs.

Source: Writing Python like Rust · Questions Nobody Asked..

Summary
This article discusses how to write Python code with type annotations and struct-like data classes, inspired by Rust and Go.

Facts

🔍 Type annotations in Python are similar to those in Rust.

🔍 Type checking in Python is performed by static checkers like mypy.

🔍 Dataclasses in Python simplify the creation of struct-like containers for a collection of values.

🔍 Python’s protocols are equivalent to Rust’s traits and Go’s interfaces.

🔍 Python’s approach to error handling is different from Rust’s, often leading to exceptions being allowed to bubble up.

Leave a Reply Cancel reply