AI News is Getting Out of Hand!

TL;DR

The biggest AI updates covered are longer-memory Transformers, Hugging Chat, new ChatGPT data controls, and Replit’s coding models. The research described could support up to 2 million and 48,000 tokens, although it would require more computing power and was not yet available. Hugging Chat was free and open source but still unreliable, while ChatGPT added an option to disable chat history. Read on for the practical implications and limitations of each development.

Transcript

so I just got back from a camping trip where I was without internet for several days it gave me the opportunity to make really cool Nerf videos and play around with some of the AI cameras that I have but I missed a lot of stuff while I was gone so I want to make a news breakdown and show you all of the cool stuff that I've come across in the AI wor... Read More

Key Insights

• 😒 Research on scaling Transformers enables the use of more tokens, expanding the capabilities of AI models like Chat GPT.
• 👊 Hugging Face's release of Hugging Chat provides an alternative chat platform based on Open Assistant, but it is still in early development stages.
• 👊 Open AI's updates to Chat GPT allows users to manage their chat history, addressing privacy concerns.
• 🤨 Replit raises significant funding and plans to develop its own language models for coding.
• 🎼 Grimes advocates for AI-generated music and offers to split royalties for successful songs using her voice.
• 🤗 Bark, an open-source text-to-speech generator, brings more nuance to voice synthesis.
• 🎮 Augmented reality showcases advancements in gaming and video editing capabilities.

Questions & Answers

Q: What are the biggest AI news developments covered?

The breakdown covers research on scaling Transformers with more memory, the release of Hugging Chat, new ChatGPT data controls, and Replit’s coding-focused language models. It also explains the practical benefits and current limitations of these developments.

Q: What does the research on scaling Transformers make possible?

The research describes a way to add memory to Transformer-based tools so they can retain information across very long sequences. It could allow up to 2 million and 48,000 tokens to be processed across the input and output, though the speaker says the technology was not yet available.

Q: Why would a larger token capacity be useful?

A larger token capacity could let a model process material that is currently too long for a single prompt. The speaker suggests that a user could provide the full code for a software product and then ask questions about it.

Q: What is the downside of supporting much longer prompts?

Processing much longer sequences would require substantially more computing power. The speaker therefore expects systems using this approach would probably cost more to operate or use.

Q: What is Hugging Chat?

Hugging Chat is Hugging Face’s chat platform based on Open Assistant’s latest model. The transcript describes it as completely free and open source, but still in an early stage and not yet a strong replacement for ChatGPT or Bard.

Q: How reliable was Hugging Chat in the demonstration?

It gave an incorrect and incomplete response when asked who won the most recent World Series. The answer mentioned the New York Yankees even though, according to the speaker, they were not in that World Series, and the generated title referred to the Boston Red Sox.

Q: How can users manage their data in ChatGPT?

Users can turn off chat history and choose whether their conversations may be used to train the models. The speaker presents this as an effort by OpenAI to address privacy concerns and let users opt out of having entered information retained for training.

Q: What did Replit announce for AI-assisted coding?

Replit announced that it was developing its own large language models specifically fine-tuned for coding. The model discussed has 2.7 billion parameters, which the speaker describes as smaller than models offered by many other platforms.

Summary & Key Takeaways

• Research titled "Scaling Transformers to 1 million tokens and Beyond with rmt" introduces a way to add more memory to tools like Chat GPT, allowing the use of up to 2 million and 48,000 tokens.

• Hugging Chat, a chat platform based on Open Assistant, is released by Hugging Face, but it is still in the early stages of development.

• Open AI introduces new ways to manage data in Chat GPT, allowing users to turn off chat history to address privacy concerns.

• Replit raises $97.4 million and announces the development of its own large language models tailored for coding.