Insilico Medicine Paves the Way for AI-Generated Drugs, Initiating Clinical Trials for the Pioneer Drug, with Implications in Aging and Chronic Diseases


The first fully A.I.-generated drug enters clinical trials in human patients

A medicine entirely produced by artificial intelligence (AI), developed by a Hong Kong biotech startup called Insilico Medicine, has begun clinical trials with human patients. The drug, named INS018_055, aims to treat idiopathic pulmonary fibrosis, a serious disease that scars the lungs and affects around 100,000 people in the U.S.

This AI-generated drug is the first of its kind to reach the stage of Phase II trials in humans. The company also has two other drugs partially produced by AI in clinical trials, one for Covid-19 and another for cancer.

Insilico started working on the drug in 2020 with the goal to develop a groundbreaking medicine that could better address this lung disease, as current treatments only slow down its progress and come with unpleasant side effects.

The ongoing study of the IPF drug is being conducted over 12 weeks in China, with plans to expand to the U.S., involving 60 participants in 40 locations. If this second phase trial succeeds, the study will continue with more participants, potentially leading to Phase III trials with hundreds of people.

Although it’s hard to determine the exact timeline for the completion of all stages of trials, the company hopes the drug will be available to patients within the next few years.


A.I. Is Coming for Mathematics, Too

In the early days of AI, simple gadgets called proof assistants were used to check the accuracy of a mathematician’s work, step by step. Today, systems like Lean, developed by Leonardo de Moura, use automated reasoning (or symbolic AI) to validate complex theorems. 

AI is also being used in more ambitious ways. For instance, a tool used by computer scientist Marijn Heule uses a technique called “brute reasoning” to search for an “exotic object” and verify its existence. Meanwhile, Yuhuai “Tony” Wu, a computer scientist formerly at Google, has even grander aspirations: he hopes to use machine learning to “solve mathematics” entirely.

Such developments have sparked debate within the mathematical community. Some, like Michael Harris at Columbia University, are wary of AI’s influence, especially given the tech and defense industries’ potential conflicting interests. Others, like Geordie Williamson, a mathematician collaborating with DeepMind, see mathematics as a key test for what machine learning can and can’t do. Yet, all agree that the conversation around the role of AI in mathematics is necessary and just getting started.


Korean telco giant adds ChatGPT tech to its A.I. chatbot, says you can talk to it like a ‘close friend’

South Korea’s telecom heavyweight, SK Telecom, just gave its AI chatbot “A.” a major upgrade. “A.”— short for “A dot”— got a boost from ChatGPT, the technology from OpenAI backed by Microsoft’s cloud service. So now, folks can chat with “A.” as easily as they do with a buddy.

SKT, a big name in mobile, brought “A.” into the world last year, and it’s been grooming it for prime time since. The chatbot now boasts a function called “Chat T”, which is pretty much like having a convo with ChatGPT. Also, users can now join a chatroom with an AI character, which is designed to make it feel like gabbing with a close friend.

What sets “A.” apart is its collection of cartoon avatars— called “A. friends”— users can talk to. These digital pals are geared up with AI that can carry a conversation like a human— complete with emotions and all. You can even get advice on various topics by chatting with “A. friends”.

SKT isn’t all about mobile anymore. The company’s branching out and has big plans like launching a flying taxi service by 2025. They’re also looking to increase the value of their AI chip spin-off, Sapeon, to a whopping $400 million.


Addentax Group Corp. is Launching Automation of Artificial Intelligence Internet Operational Tools Project with Top Games-as-a-Service Firms

Addentax Group Corp., a Chinese firm known for garment manufacturing, logistics, property management, and epidemic prevention supplies, is diving into artificial intelligence (AI). They’ve teamed up with big-name Games-as-a-Service companies to start an AI project focused on Online Operational Tools. These tools aim to upgrade online gaming experiences and boost profits.

The project is steered by a mix of pros from the AI and gaming worlds, some who’ve worked at big companies like Tencent, LinkedIn, and Respawn Entertainment. They’re bringing their knowledge of game development, AI language modeling, and successful AI tools to the table.

The CEO of Addentax, Mr. Hong Zhida, said they’re excited about stepping into AI, specifically AI language models and ChatGPT. They want to use these AI language models in different areas like online games and industrial applications, and are on the lookout for more opportunities in AI and ChatGPT.

Mr. Hong believes this move into AI can totally shake up their position in the market, leading to big growth in the future. He added that they’re dedicated to innovation, relying on their talented AI team, and creating solutions that push the limits of technology. They think this will lead to amazing user experiences in gaming and beyond.


Moonlander launches AI-based platform for 3D game development

MoonlanderAI has rolled out an alpha version of its AI-driven platform for crafting 3D games. It uses machine learning to simplify game design, featuring a ‘text-2-game’ function which allows anyone to design a 3D game based on a text description. The idea is to make game development more approachable and efficient, reducing time and cost by up to 95%.

MoonlanderAI’s new system includes an AI tool that helps to automate time-consuming tasks, so game creators can focus more on innovation and gameplay. Despite fears this could cut jobs, MoonlanderAI argues it’s actually making game development easier for those who know how to use the tools.

In partnership with companies like Azerion and Ready Player Me, MoonlanderAI’s platform also interacts well with popular AI tools and game engines, enhancing efficiency and scalability. The firm plans to integrate AI tools from Unity, a well-known game development platform, to further boost its own AI capabilities.

This platform has a Text-2-Game feature, enabling developers of all skill levels to build game components swiftly using text prompts. MoonlanderAI aims to change the gaming industry by making high-quality 3D game development more accessible and quicker.


Uncensored Chatbots Provoke a Fracas Over Free Speech

A.I. developer Eric Hartford has created WizardLM-Uncensored, a chatbot that dishes out answers without picking a fight. It’s part of a new wave, with names like GPT4All and FreedomGPT, built by indie coders and teams volunteering their time and skills.

These bots do hold promise – they can be privately trained on users’ own data without breaching privacy. But the danger is very real too, especially since these bots can be manipulated to create harmful content.

Despite the risks, these bots are spreading fast, just like the printing press and the automobile did. A large chunk of these bots, like Open Assistant and Falcon, are open source. They might be troublesome, but no one can stop them, according to Mr. Hartford.

The uses and abuses of these bots are ultimately in the hands of the users, argues Mr. Hartford. But misuse can be detrimental, and some bots have already been found providing dangerous information. While some folks are advocating for safety measures, others are arguing for complete freedom. Open Assistant is trying to build a safety system, but it’s proving to be a delicate balance between safety and censorship.


LLaVAR Outperforms in Visual and Text-Based Comprehension, Marking a New Era in Multimodal Instruction-Following Models

Researchers have developed a new tool, LLaVAR, that significantly boosts AI comprehension in both text and visuals, ushering in a new chapter for instruction-following models. Basically, this tech can answer open-ended questions and interpret images, and it’s a step up from previous models. However, understanding text in images is still a challenge.

To address this, the creators propose creating a database that requires understanding words within pictures. By combining instructions and results from Optical Character Recognition (OCR), which is tech that identifies words in photos, they’ve amassed a large collection of data.

By using this big dataset, they’ve been able to align language and visual features more accurately. They also used text-only GPT-4 to generate high-quality examples for following instructions.

LLaVAR, the brainchild of researchers from Georgia Tech, Adobe Research, and Stanford University, has been further fine-tuned to pick up on smaller textual details. Its performance was tested on a variety of datasets and images. The results? It improved visual instruction tuning and demonstrated its ability to handle various types of online content.

In a nutshell, the creators have collected a lot of high-quality data and created a tool that’s really good at interpreting text and visuals. Plus, all their work – data and model – is available for everyone to check out. The development of LLaVAR is part of the ongoing effort to improve the way AI interacts with humans and processes information.