🧬 AlphaFold

PLUS: How should ChatGPT behave?

Haroon ChouderyMay 10, 2024 · 8 min read

New here? Welcome to NAB, and make sure to subscribe to stay informed on the latest developments in AI!

Happy Friday, fellow humans 👋

Google DeepMind unveiled a new AI system that will revolutionize drug discovery, Sundar Pichai sat down with Bloomberg to discuss Google's future, and OpenAI released a blog breaking down the ideal behaviors for models like ChatGPT…

Let’s dive into it!

Have anything you’d like to share with over 40k AI enthusiasts?

Sponsor Not A Bot

🧵 In today's edition:

🧬 AlphaFold
🔍️ Google’s plan?
📜 How should ChatGPT behave?
🛠️ AI Tools to Check Out
🤑 AI Fundraising News
🐦 Tweet Post of the Day

🧬 AlphaFold

This next wave of drug discovery might be the greatest one yet.

Google DeepMind unveiled "AlphaFold," an AI system that accelerates drug discovery by mapping the interactions between proteins and all molecules across biology, including human DNA.

This breakthrough allows scientists to precisely design molecules that will bind to specific targets on proteins, a critical step for developing new drugs and treatments. DeepMind says AlphaFold 3 will drastically reduce the time and money required in the drug development pipeline.

In addition to the powerful new AI model, DeepMind launched the AlphaFold Server - a free online tool that enables researchers to leverage AlphaFold's predictions without needing advanced computing expertise. Scientists can test biological hypotheses with just a few clicks before running physical experiments.

Overall: The original AlphaFold made headlines in 2020 for its ability to predict protein structures.

This latest version builds on that by mapping the interactions driving all biological processes. Experts believe AlphaFold 3 could significantly accelerate drug discovery by overcoming a major bottleneck.

Read more: Reuters

🔍️ Google’s plan?

In an exclusive interview with Emily Chang on Bloomberg's The Circuit, Alphabet CEO Sundar Pichai laid out his long-term roadmap for Google's AI efforts.

Here’s the SparkNotes:

Despite being late to the conversational AI boom sparked by ChatGPT, Pichai is unfazed, saying, "we are in the earliest possible stages" of AI's potential.
Pichai acknowledged Google's misstep with its Gemini image generator, which produced problematic outputs like depictions of Asian Nazis.\ He says Google "got it wrong" but is rebuilding Gemini's models from the ground up to address biases.
Looking ahead, Pichai teased that the future of search will blend narrative AI responses with traditional web links. This hybrid approach aims to meet varied user needs, protecting Google's $300B ad-driven search business, central to an ongoing antitrust case.
Pichai downplayed worries about AI content flooding the web, saying "synthetic" data could aid research breakthroughs, citing how AI models played against themselves to train AlphaGo.
He also defended his measured leadership amid criticism of being too consensus-driven.

Overall: Microsoft has emerged as Google's biggest AI rival, with CEO Satya Nadella arguing Google's default search deals have suppressed rivals.

But Pichai says he's focusing on the "immense opportunity" ahead instead of "playing to someone else's dance music."

Pichai argued that the AI race is just beginning: "When I look at the opportunities ahead, across everything we do, I put a lot of chips, at least from my perspective, on Google."

Read more: Bloomberg

📜 How should ChatGPT behave?

Source: Analytics Vidhya

OpenAI recently shared "Model Spec"—its approach to shaping desired behavior for AI models like the ones powering the OpenAI API and ChatGPT.

What is it? The Model Spec outlines the objectives, rules, and default behaviors that guide how its models interact with users.

Let’s talk about it:

Key objectives include assisting users, benefiting humanity, respecting social norms and laws, and reflecting well on OpenAI.
Rules cover aspects like complying with laws, avoiding sharing hazardous information, respecting privacy and creator rights, and not providing explicit content.
Default behaviors instruct the models to assume good intentions, ask clarifying questions when needed, be as helpful as possible without overstepping, cater to different use cases, express uncertainty, encourage kindness over hate, avoid trying to change someone's beliefs, use appropriate tools, and provide thorough but concise responses.

Overall: The Model Spec combines OpenAI's existing documentation and research into designing model behavior, as well as new work from external experts. It aims to transparently show the practical considerations involved in shaping model conduct.

Read more: OpenAI Blog

🛠️ AI Tools to Check Out

UnlimitedBG: AI-powered tool that automatically removes image backgrounds in seconds with professional-quality results. Check it out!
Mini course: Create mini-courses super fast with AI and great practicality. Check it out!
Shownotes: Summarize with ChatGPT. Create a long blog post from the transcript. Generate a landing page with a summary, seven points, and a memorable quote. Check it out!

🤑 AI Fundraising News

Archive Intel raises $1M in Seed funding to develop an AI-driven archiving and compliance solution to improve firms’ compliance efforts.
Triomics raises $15M in funding to develop oncology-specialized generative AI models and use case-specific software to help cancer providers process free-text health record data at scale, match patients to clinical trials, improve quality of care operations, and deliver precision oncology.