AI Newsletter #98 - Language models that teach themselves to use external tools via simple API calls

Welcome to Nural's newsletter focusing on how AI is being used to tackle global grand challenges.

Packed inside we have

• Chinese Search Giant Baidu to Launch ChatGPT-Style Bot
• Toolformer - language models that teach themselves to use external tools via simple API calls
• and optimizing prompts for image and text models through gradient descent

If you would like to support our continued work from £1 then click here!

Marcel Hedman

Key Recent Developments

Alphabet's stock plunges $100 billion after it's chatbot provides incorrect answer

What: "Shares of Alphabet slipped on Wednesday after reports that promotional material for the Google parent's new artificial intelligence chatbot contained inaccurate information"

Chinese Search Giant Baidu to Launch ChatGPT-Style Bot

What: The ChatGPT craze continues! Baidu, China's answer to Google, have announced their response to ChatGPT named ERNIE (Enhanced Representation through Knowledge Integration). Quoting Baidu’s spokesperson, “what sets Ernie apart from other language models is its ability to integrate extensive knowledge with massive data, resulting in exceptional understanding and generation capabilities.”

Technical summary:

Ernie is a pretrained language model with 260 billion parameters. For context, ChatGPT uses the 175-billion parameter GPT3 model. However, bigger does not always mean better so it is yet to be seen if they have replicated the effectiveness of ChatGPT. Ernie was trained on unstructured data and a knowledge graph.

Text-to-4D: Meta AI Proposes a Novel System for Text-to-4D (3D+time) Generation by Combining the Benefits of Video and 3D Generative Models

AI Ethics

🚀 Former Google CEO, Eric Schmidt says A.I. could be the military’s new nuke—but only if the Pentagon acts like a tech firm

🚀 For the first time, A.I.-generated deepfakes, were detected in a state-aligned information campaign

🚀 Toolformer - introduces language models that teach themselves to use external tools via simple API calls [Paper]

🚀 A Categorical Archive of ChatGPT Failures [Paper]

🚀 Optimizing prompts for image and text models through efficient gradient-based optimization [Paper]

Interesting AI developments

🚀 Stanford Researcher develops a simple prompting strategy that enables open-source LLMs with 30x fewer parameters to exceed the few-shot performance of GPT3-175B

🚀 Instagram’s founders are back with a new AI powered app

🚀 Perplexity AI Challenges ChatGPT and Google With New Conversational Search Engine

🚀 Quora opens its new AI chatbot app Poe to the general public

Cool companies found this week

AI safety

Anthropic - Anthropic is an AI safety and research company that is working to build reliable, interpretable, and steerable AI systems. They recently received a $300m investment from Alphabet via a cash-for-computing investment. I.e. the investment must be spent on Google cloud.

Generative AI

Character.AI - Character, a ChatGPT alternative, lets you have live chats with all sorts of characters in real time. Whether these are tech heroes like Elon Musk or interviewers for your upcoming role.

Best,

Marcel Hedman
Nural Research Founder
www.nural.cc

If this has been interesting, share it with a friend who will find it equally valuable. If you are not already a subscriber, then subscribe here.

If you are enjoying this content and would like to support the work financially then you can amend your plan here from £1/month!