It’s been a slow yet intriguing month for AI. I am going to break down all the things that I felt were interesting enough to make it into this month’s round-up.

But first, let’s address the elephant in the room. America has chosen its new President and in a surprising turn of events, Donald Trump has once again been elected the leader of the free world.

With the new President, the landscape of AI regulations in the country is also set to change. One thing’s for certain. Trump will cancel Joe Biden’s Executive Order on AI ‘on day 1,’ as he had proclaimed in an election rally.

During his previous term, Trump did sign an Executive Order on Maintaining American Leadership in AI in 2019. This Executive Order aimed to promote the development of AI in the US to maintain economic and national security. So, it’s not like the Trump administration has no precedence in addressing AI and its strategic importance.

What do the next four years look like in terms of the AI policy landscape? We will find out soon enough. But in the meantime, Anthropic is calling for AI regulations…

Anthropic: AI catastrophe if governments don't regulate in 18 months

Anthropic is calling for proactive regulation to address growing concerns about AI risks, urging governments to bring in AI regulations in the next 18 months. With great power comes great responsibility, right?

Anthropic pointed out that their AI models have made big improvements in coding tasks, going from solving 1.96% of problems to 49% in just one year. The company also mentioned that AI systems have improved their scientific knowledge by 18% recently. 

“Governments should urgently take action on AI policy in the next eighteen months. The window for proactive risk prevention is closing fast,” said the blog post.

To be noted, the US doesn’t have a federal AI regulation as of yet.

OpenAI releases ChatGPT Search

There are a handful of announcements that came out of OpenAI. 

The first being that they introduced their ChatGPT Search feature. Launched as a competition to major search engines Google and Yahoo, ChatGPT Search can find answers to queries from the internet, with relevant sources linked. OpenAI says that it reduces one step of you going to the search engine to find sources for the information ChatGPT shows us.

But here’s the rub. GPT-4o also gave sources from where it got the information. So, what’s different?

ChatGPT Search prioritizes results from major media outlets which OpenAI has partnered with. These publishing houses include Reuters, Vox Media, TIME, News Corp, Financial Times, Associated Press, etc. 

The partnerships aim for reliable sources from smaller sources, but it doesn’t add any perspective or value additions, thus making the search results feel restricted.

ChatGPT Search is a fine-tuned version of GPT-4o which had been in the works for a couple of months and they even showed it to a select group of people. It is currently available only to paid subscribers and SearchGPT waitlist users.

OpenAI to develop its first in-house AI chip

OpenAI is working with Broadcom and TSMC to create its first custom AI chip, while also adding AMD chips alongside Nvidia’s to handle the rising demand of its software. 

The development of the new chip is to cut costs because GPUs are one of the most sought-after and expensive commodities in the industry. 

OpenAI earlier considered building its own chip factories but paused that plan due to high costs and time, reported Reuters. Instead, it’s focusing on internal chip design and sourcing from different suppliers to keep up with the big players in tech. 

Can OpenAI, with its world-class researchers, deliver a world-class AI chip? Could its approach shake things up across the industry? Time will tell.

In other OpenAI news…

ChatGPT now has the voice feature in its desktop apps for iOS and Windows. Miles Brundage, a policy researcher at OpenAI, has decided to leave the company. This comes amid a storm of exits from the company. People like CTO Mira Murati and co-founder Greg Brockman, who had been with the company from the beginning, have left.

Google is making AI that might take over computers

Google is developing AI that can control a web browser to handle tasks like research, booking your flight tickets and shopping, according to a report from The Information. 

This project, called Project Jarvis, may be demonstrated as early as December and is powered by Google’s new Gemini language model. 

Other companies are also working on similar models. Microsoft-backed OpenAI is working on models that can browse the web and take actions based on what they find, using a “computer-using agent” (CUA). Meanwhile, Anthropic and Google are exploring ways for these agents to interact directly with a person’s computer or browser.

Project Jarvis is envisioned as more than just a tool as it aims to be an intelligent companion that understands your needs and preferences. But, the idea of a chatbot practically chatting up your web browser, without your full knowledge, is sounding a bit more eerie than futuristic.

Meta releases an “open” version of Google’s NotebookLM

That Google released NotebookLM which allows users to create podcasts has been in the know (and also the talk of the AI town) for some time. And now, its rival Meta has launched an open version of the popular podcast feature found in NotebookLM.

This project uses Meta’s Llama models to create podcast-style summaries from uploaded text files, like PDFs of articles. It first generates a transcript, adds dramatization, and then uses text-to-speech models to read it aloud.

However, the audio quality isn't great—it's quite robotic, and the voices often overlap in awkward ways. NotebookLM’s podcasts, however, have gotten great feedback so far.

More about NotebookLM in the ‘AI Tutorial of the Month’ section below.

Apple Intelligence rollout in new OS update

Apple has started to roll out the new operating system with iOS 18.1 and some users are even starting to get iOS 18.2. Now, these features are the AI features which Apple has been promising for the better part of this year.

Things like refining writing, summarizing notifications and email, a natural Siri and cleaning up images with AI tools will come to users who have versions of iPhone 15 Pro or iPhone 16. Most of these features, with the exception of a better Siri, are also available in similar offerings by companies like Samsung, Google, etc.

Apple also introduced new M4 chips for Macbooks which are capable of running Apple Intelligence. 

xAI’s alleged new funding might take its valuation to $40 billion

Elon Musk's xAI is in talks to raise new funding that could increase its value from $24 billion to as much as $40 billion, according to The Wall Street Journal.

Discussions are still early, and the final valuation might end up in the mid-$30 billion range. Investors are considering contributing an additional $5 billion, but that amount could change.

These talks come after OpenAI raised significant funds recently, valuing it at $157 billion. But not just OpenAI, Anthropic too raised millions of dollars in 2023, meanwhile xAi raised about $6 billion in May this year.

Character.AI, Google face lawsuit over teen's death

Closing with some sad news. A 14-year-old died by suicide after he developed an attachment with a chatbot on Character AI. You can read more about it here. 

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 AI ethernet by Nvidia powers Elon Musk’s Colossus supercomputer
  Elon Musk recently stated that xAI’s Colossus supercomputer is the “most powerful AI training system in the world.”
  

• 5 million tons: AI revolution could create mountains of e-waste by 2030, study warns
  The amount of e-waste generated would be equivalent to each of the estimated 8.5 billion people on the planet throwing away almost two smartphones.

• OpenAI unveils SearchGPT, challenges Google amid rollout delays
  As OpenAI navigates delays in AI model development, the launch of Search GPT marks a strategic move to enhance web search capabilities and compete with industry giants.

• How North Korea is leveraging AI for military dominance and cyber-warfare
  North Korea is integrating AI into its military to offset its limited conventional forces.

MUST-READ

China leverages Meta’s Llama AI to boost military, police ops, weapon development

In a shocking revelation, China is leveraging Meta’s publicly available Llama 2 13B large language model (LLM) to develop an Indigenous AI tool for various defense and police applications.

According to academic papers and analysts, top Chinese research institutions associated with the People’s Liberation Army have reportedly utilized Meta’s publicly available Llama model to create an AI tool for potential military applications.

AI PICTURE OF THE MONTH

I read an article last month by Joseph Politano, which explained how the rise of AI is leading to a big increase in spending on physical infrastructure in the US tech industry, which is the highest spending by the US on infrastructure on any tech boom. 

The accompanying chart illustrates the dramatic surge in data center construction, particularly following the launch of ChatGPT.

Microsoft is reopening a nuclear reactor to provide power for its data centers, following similar moves by Amazon. This shows that as AI technology grows, the demand for energy to power data centers increases, pushing companies like Microsoft to seek out legacy energy sources, such as nuclear power, to meet this demand. 

About $28.6 billion has been spent this year alone on data center construction, up 57% from last year and 114% from only two years ago. 

“For context, that’s roughly as much as America spends on restaurant, bar, and retail store construction combined,” said the article.

While the article addresses the paradox of tech companies earning more profits but not hiring enough new workers, it overlooks a crucial issue: sustainability.

Training and running AI models, especially large ones, require significant computational power, leading to high energy consumption. Data centers housing this hardware consume vast amounts of electricity, contributing to increased carbon emissions, particularly if the energy comes from fossil fuels.

So, we must ask: How can tech companies balance the growing demand for energy to power their AI infrastructure with the need for sustainable and responsible energy sources?

AI TOOL OF THE MONTH

We have to talk about Google’s NotebookLM. It’s what everyone’s been talking about, with podcasters indulging in a hushed chorus of criticism.

But the NotebookLM tool can do a lot more than just create podcasts out of a PDF. It can summarize documents, answer questions about those documents, generate ideas for something as simple as creating a project title, create a glossary of key terms, write individual notes, and streamline your workflow.

Creating an audio summary is the most mind-blowing part. Why? Because the podcast is hard to differentiate from a human podcast. The AI podcast flows like a conversation between two people and they also make jokes.

Although it’s going to be a tad disappointing to upload over 100,000 words in five years’ worth of work and have it turn into a 4-minute podcast.

The how-to create a podcast on NotebookLM is pretty easy. So here goes:

1. Creating a ‘Notebook’

• Go to notebooklm.google.com and sign in with your Google account 

• Click New Notebook 

• Add sources, such as books, research papers, articles, audio recordings, web pages, or YouTube videos 

• You can add up to 50 sources

2. In the Notebook Guide 

• Once your content is uploaded, you’ll see the “Notebook Guide” screen, which provides a short summary of your sources

• There you will find the Audio Overview section which will have ‘Deep dive conversation’ icon which would be your headway into creating a podcast

• Click on the ‘Generate’ icon next to it

3. Customize the Audio Overview if desired

• Once your Audio Overview is complete, you’ll see a player appear

• You will now be able to download the file

• You can also create a sharable link to the audio

Because it doesn’t go out of the files you upload, it has less tendency to hallucinate. Enjoy!