INTERESTING ENGINEERING SHOP

An interesting week in AI kicked off with Elon reportedly embarking on an ambitious endeavor to construct one of the world’s most AI powerful supercomputers. This “Gigafactory of Compute” would run 100,000 of Nvidia’s H100 chips and be four times larger than existing GPU clusters.

More brain power than Musk

Musk may be being short-sighted in focusing on silicon chips, when the obvious and not at all disturbing solution could be running AI on lab-grown human brains.

A Swiss company has launched a ‘bioprocessing platform’ that uses lab grown brain tissue to run computing tasks. Their claim is that this tissue uses a million times less power than silicon chips, which could be a solution to AI’s energy demands if it can operate at scale.

The OpenAI drama continues

The OpenAI management saga keeps on rolling with more twists and turns than a telenovela. It should have been a good week for Sam & Co. - announcing a successor in training for GPT-4 and forming a new safety and security committee.

However, things quickly went off the rails as former OpenAI board member, Helen Toner, went to town on what really happened with Sam’s firing.

Speaking on the Ted AI Show, Toner claims that:

• The OpenAI board only found out about the launch of ChatGPT through Twitter

• Altman didn’t inform the board that he owned the OpenAI startup fund, whilst claiming he had no financial interest in the company

• Altman gave inaccurate information on the number of formal safety processes that were in place

Coming after the departure of Ilya Sutskever and Jan Leike, who criticized OpenAI for prioritizing "shiny new products" over vital safety work, question marks must be raised over Altman’s commitment to safety and leadership and whether the new safety board, with Altman a member, is anything more than a whitewash.

Google’s AI search backlash

Google’s launch of AI search summaries has received significant, and not unexpected, backlash. Google claims its search to be ‘built from the ground up to deliver reliable, helpful, and high-quality information’, which is why it released AI summaries trained on old Reddit posts.

Results have ranged from the humorous to the dangerous. It has claimed astronauts met cats on the moon and that the list of ‘fruits that end with um’ includes applum, bananum and coconut.

More worryingly, it also recommends “you should eat at least one small rock per day” as “rocks are a vital source of minerals and vitamins”, and suggests putting glue in pizza topping.

Media sites have delighted in slamming Google for a move that threatens their search traffic. However, we should be asking who is responsible if someone does get harmed after false advice from AI search? Is there any culpability on Google’s side? Google should be prioritizing the reliability of information, instead first on their roadmap was figuring out how to get adverts in.

Did a friend forward this e-mail to you?

IE₊ SUPPORT INTERESTING ENGINEERING
_{Invest In Science And Engineering}

Enjoy exclusive access to the forefront of AI content, highlighting trends and news that shape the future. Join a community passionate about AI, delve into the latest AI breakthroughs, and be informed with our AI-focused weekly premium newsletters. With IE+, AI reporting goes beyond the ordinary - and it is Ad-Free.

NEWS

• World’s 1st robot surgeon concept for head-swap surgery shocks internet

  Biomedical engineering and neuroscience startup BrainBridge has unveiled its plans to perform the first human head transplant within the next eight years.

• Musk to build most powerful AI supercomputer powered by 100,000 Nvidia chips

  Elon Musk’s artificial intelligence startup, xAI, is reportedly embarking on an ambitious endeavor to construct one of the world’s most powerful supercomputers, aiming to redefine the AI landscape.
  

• Skylark Labs' AI-powered anti-drone system ARIES passes key DoD test

  Skylark Labs, a California-based tech startup, has just completed tests of its counter-unmanned aerial system (c-UAS) solution. Called the Aerial Reconnaissance & Elimination System (ARIES), the system was demonstrated to representatives of the US DoD.
  

• 16 lab-grown brains run world’s first ‘living computer’ in Switzerland

  Swiss technology firm Final Spark has successfully launched Neuroplatform, the world’s first bioprocessing platform where human brain organoids (lab-grown miniaturized versions of organs) perform computational tasks instead of silicon chips.

MUST READ

OpenAI starts training next AI model, sets up safety and security committee

OpenAI announced on Tuesday that it is training a new artificial intelligence model that would serve as a successor to GPT-4 that is behind its online Chabot, ChatGPT.

The company also said that it has set up a committee to advise the board on critical safety and security decisions for all OpenAI projects and operations.

LEXICON PODCAST

Jaroslaw Rzepecki talks to us about using a unique set of data and ML techniques to build what it calls one of the world’s first 'large engineering models' (LEM).

AI PICTURE OF THE WEEK

The Wall Street Journal ranked OpenAI’s ChatGPT against Microsoft’s Copilot and Google’s Gemini, along with Perplexity and Anthropic’s Claude to see which worked best at different tasks.

TUTORIAL

Getting started with Github Copilot Workspace

Github has launched the technical preview of Copilot Workspace, a tool that lets users create applications by editing plain English. This shifts development from traditional coding to natural language interaction.

Currently on a waitlist, here’s what you can expect when you get in.

Getting started

GitHub Copilot Workspace meets developers at the GitHub repository, issue or task. In the workspace the editor is replaced with a text input field where the user can put in a simple natural language prompt, it is also multilingual.

Specification

From the prompt, Copilot Workspace analyzes the instruction and assets in the environment, then proposes a specification. This is all in natural language that you can edit to change the proposal or add and remove items.

Build the plan

In the next stage, an agent runs through all files and assets that you have and figures out how to use or modify those files, or find files and functions that need adding to the repository.

Implementation

Copilot Workspace then uses your task, specification and plan to write the code for you and outputs into the GitHub Codespace, where you can adjust the until happy with the final result.

Copilot Workspace looks like it will massively lower the barrier to entry for development, allowing projects to be run entirely through natural language.

To join the waitlist, click here.