Hello, Starters!

AI models come and go, each with its very unique capabilities that, for fleeting moments, leave researchers and enthusiasts in awe. But sometimes, not even they can keep track of where these breakthroughs are coming from!

Here’s what you’ll find today:

• ChatGPT now remembers your preferences!

• gpt2-chatbot: The new mysterious model

• GitHub Copilot Workspace is coming

• SenseTime presents SenseNova 5.0

• DeepMind introduces the “Gecko” benchmark

• And more.

📋 ChatGPT now remembers your preferences! (1 min)

After announcing it in February, OpenAI has finally rolled out ChatGPT's memory feature for Plus users. The function allows the chatbot to be even more customised by remembering details and learning from past conversations.

ChatGPT is designed to be a multifaceted assistant. This update makes it an even more reliable companion, as it keeps track of your preferences and uses them in new conversations without needing reminders. However, unwanted details can be "forgotten".

🕵️ gpt2-chatbot: The new mysterious model (4 min)

A powerful model from an "unknown" source has the AI world caught in a frenzy. "gpt2-chatbot" recently appeared on the LMSYS Chatbot Arena, where users compare AI systems. Despite its low-key entrance, its capabilities are now the talk of the moment, with many claiming it matches and potentially exceeds GPT-4.

A post on X showcases the model easily solving a problem from the International Math Olympiad. Yet, there's no way to pinpoint who's responsible for this creation. Many suspect OpenAI, but that's still unconfirmed.

🧑‍💻 GitHub Copilot Workspace is coming (4 min)

GitHub continues to revolutionise developers' workflows, first through GitHub Copilot, then with Copilot Chat, and most recently by unleashing the technical preview of GitHub Copilot Workspace—a developer environment that makes it possible to brainstorm, plan, build, test, and run code using natural language. 

By doing so, GitHub aims to streamline and democratise software development. Copilot Workspace is designed to be used across several devices, such as mobiles, making it easier to get started.

🇨🇳Chinese AI company SenseTime has launched SenseNova 5.0, a powerful large language model with over 10TB of token training. This model has reportedly surpassed the performance of GPT-4 on several benchmarks, showcasing enhanced results in understanding and generating human-like text.

🦎There hasn't been a specific way to fully understand the capabilities of text-to-image models, but Google DeepMind aims to change that with "Gecko," an evaluation method that provides 2,000 diverse text prompts to measure their skills and complexity levels.

⚡️Quick links

Meta's AI-everywhere push raises hackles

Google allegedly lays off US Python team, builds new team in Munich

Can Rabbit’s R1 outsmart the smartphone assistants?

Yann Lecun: "LLMs are still pretty dumb."

New AI training method mitigates the "lost-in-the-middle" problem that plagues LLMs

Thank you for reading!