Are You the Next AI Trainer?

IE Academy is looking for individuals passionate about AI to lead our next training session. If you're ready to share your knowledge and inspire others, fill out the form using the link below.

Apply Now to Lead the Future of AI!

I know we are only one month into 2025, but it seems like a year has gone by between January 1 and today. So much has happened in AI that it’s been hard to keep up. There was one piece of news that kind of sucked the oxygen out of everything else—DeepSeek. 

Unless one lives under a rock, it was hard to miss out on the developments around DeepSeek—like how the stocks of the US big tech companies crashed, how Italy became the first country to ban the app, and how everyone now seems to be questioning the pecking order when it comes to AI dominance. Does the US no longer hold the crown for AI supremacy?

IE’s coverage of DeepSeek has been pretty comprehensive over the last two weeks, so we are not going to get too deep into the Chinese AI model. We’re just going to quickly wrap up what’s the latest. You can check out some interesting stuff here.

Industry reacts to DeepSeek

In what seems like a reaction to DeepSeek, OpenAI, and Microsoft added the ‘Reason’ and ‘Think Deeper’ buttons to ChatGPT and Copilot, respectively. What does it mean? 

These buttons serve the same purpose: giving users access to advanced reasoning capabilities that break down complex questions into multiple steps, reducing errors and hallucinations.

The 'Reason' button allows free users to try OpenAI's reasoning model o3-mini, which it launched five days ago, for the first time. Whereas, Microsoft Copilot’s ‘Think Deeper’ runs on OpenAI’s o1 model.

US report on copyright and AI

The US Copyright Office has come out with part 2 of a report on AI and copyright law. Published in January 2025, the report states that AI-generated art can only be copyrighted if there is significant human input. Fully AI-generated works, including those created solely from prompts, are not eligible for copyright. 

For example, artists who use AI as a tool while still calling the creative shots can claim copyright, but it’s judged case by case. The Copyright Office went through 10,000+ public comments, and most people agree that human input is key to copyright eligibility. 

No new AI-specific laws are coming in for now. They’re sticking to existing rules. That said, the Office is keeping an eye on AI and might change its stance down the line. This comes amid a spate of copyright lawsuits that have been filed against OpenAI.

Nvidia mini AI supercomputer

Nvidia announced in January that it will launch a personal AI supercomputer called Project Digits in May 2025. Powered by the GB10 Grace Blackwell Superchip, it packs massive AI processing power into a desktop-sized device that runs from a standard outlet, something that earlier used to require bulky, power-hungry setups. 

Digits can handle AI models with up to 200 billion parameters, starts at $3,000, and looks a lot like a Mac Mini. For heavy-duty tasks, two units can be linked to handle 405 billion parameters. 

It delivers 1 petaflop of AI performance, runs on Nvidia’s latest CUDA and Tensor cores, and was co-developed with MediaTek for efficiency. Jensen Huang calls it a game-changer, putting AI supercomputers on every developer’s desk.

Sam says he knows how to build AGI

OpenAI CEO Sam Altman claims OpenAI knows how to build AGI — AI that is smarter than human beings — and expects AI agents to start “materially changing” business output this year. 

In a blog post, he shifted focus from AGI to "superintelligence," a vague next step that sounds suspiciously like AGI but with a shinier label. He envisions AI accelerating scientific breakthroughs and prosperity, though OpenAI’s real goal, AGI that “benefits humanity,” remains elusive. 

Notably, it was only in December, in an interview with the New York Times, that Altman downplayed AGI’s importance, making this new grand vision feel more like a marketing spin than a concrete breakthrough.

However, I believe the biggest challenge the companies across the AI spectrum will have to deal with is the definition of AGI. And how does one know when they have achieved it?

AI models are now…underthinking?

Chinese researchers found that AI models like OpenAI’s o1 struggle with complex reasoning because they abandon good ideas too quickly, wasting computing power and lowering accuracy. This jumping between ideas too quickly instead of fully exploring promising reasoning paths is being termed as ‘underthinking.’

This leads to shallower reasoning and worse performance, especially in math problems, finds the study. By studying open-source models, they discovered that frequent thought-switching is linked to wrong answers. 

Models like OpenAI’s o1 frequently switch problem-solving strategies instead of fully exploring promising paths. The study found that 70% of wrong answers had at least one valid reasoning step that was ignored. 

To fix this, the Chinese researchers introduced TIP (Thought Switching Penalty), a decoding tweak that discourages models from switching ideas too soon. The result? Better accuracy on tough problems without retraining the models.

Putin deepfake movie

Something hilarious and impressive caught my and the rest of the internet’s attention last month. A Polish biopic about Vladimir Putin uses AI to superimpose the Russian President’s face onto actor Sławomir Sobala, who studied Putin's body language for two years to make the portrayal more accurate. 

The film, titled ‘Putin’, will be shown in 64 countries, but director Patryk Vega admits AI alone can't perfectly capture Putin’s emotions and movements, so they blended human acting with AI effects. 

The film promises a controversial and exaggerated view of the Russian president, with scenes like him being bullied as a child and later in a hospital bed in a soiled diaper. It sparked interest from Russian intelligence, who even tried to pay for a screening.

Watch the trailer:

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

• After DeepSeek’s strike, OpenAI drops Deep Research AI agent for tough studies
  The launch follows OpenAI’s earlier release of Operator, a tool that browses the web to complete tasks.

• China blames US hackers for DeepSeek cyber strike, calls it a brute force attack
  DeepSeek launched a free AI assistant, challenging US artificial models and sparking concerns in America’s AI industry.

• DeepSeek 100% fail: Chinese AI model could not stop a single harmful prompt
  The team used “algorithmic jailbreaking” to test DeepSeek R1 with 50 harmful prompts.

• US warship gets powerful AI to scan 10,000 readings per sec for battle-readiness
  ERM v4 is a key part of the Pentagon’s Condition-Based Maintenance Plus program.

• Robots master Jenga whipping, motherboard assembly with new AI learning method
  The new AI-powered technique, developed at UC Berkeley, includes a combination of reinforcement learning and human tutoring to help robots master tasks quickly and efficiently.

MUST-READ

$30 DeepSeek dupe? US scientists claim to duplicate AI model for peanuts

A group of researchers at the University of California, Berkeley, claims they’ve managed to reproduce the core technology behind DeepSeek’s headline-grabbing AI at a total cost of roughly $30.

The news is another twist in a quickly developing narrative about whether building state-of-the-art AI demands colossal budgets or if far more affordable alternatives have been overlooked by tech’s biggest players.

DeepSeek made waves recently by introducing R1, an AI model that claims to replicate the functions of ChatGPT and other costly systems at just a fraction of the training expense typically seen in Silicon Valley.

AI PICTURE OF THE MONTH

A picture is worth a thousand words. But this one is worth billions of dollars. The AI picture of the month is that of billionaires at Donald Trump’s presidential inauguration on January 20, 2025.

Present are Meta CEO Mark Zuckerberg, Amazon founder Jeff Bezos, Google CEO Sundar Pichai, and Tesla founder Elon Musk. They aren’t just standing in line. They’re toeing it too. Missing from the frame is Apple’s Tim Cook and TikTok’s Shou Zi Chew, who were also present.

Many of the attendees at Trump's swearing-in raised eyebrows, as their political stances had not appeared right-leaning in recent years, nor had their platforms traditionally aligned with conservative ideologies. Yet, their presence suggests a shift - whether strategic, opportunistic, or driven by necessity.

With Trump back in the White House, it’s safe to say that the tech and AI regulation landscape is going to look very different. The industry’s biggest players may not just be adapting to change, they could be helping shape it. The next four years will redefine the relationship between Silicon Valley and Washington.

AI TOOL OF THE MONTH

DeepSeek-V3

I know, I know. It's lazy and kind of an obvious choice for this month. But it’s getting banned in more and more countries and US states as it gains popularity. Italy was first, followed by Taiwan where government agencies are banned from using the app.

Texas has become the first state in the US to ban the app on all state-issued devices due to concerns over data security and the potential influence of the Chinese Communist Party, given the app's Chinese origins.

The way things are rolling, there might be more bans coming in. So check it out while it’s still available.

You can download the DeepSeek app for free from the App Store or Google Play, and you can also use it on your desktop browser. If you want to use the DeepSeek API, keep in mind that it’s not free like the mobile and web versions — it comes with a cost.

You can access everything on its website.

However, an interesting thing happened when I was playing around with the app. It refused to answer questions about China, Xi Jinping, and the State’s surveillance system. It also reportedly doesn’t answer questions about Tiananmen Square and Taiwan.

DeepSeek doesn't answer certain sensitive questions due to censorship rules, especially those related to China’s government and politics, as these are seen as violating China's "core socialist values." 

However, users are smart. They have found ways around this by phrasing questions differently (like using "leetspeak" or special characters), and DeepSeek would provide answers that skirt the censorship, such as describing Tank Man or protests in a more indirect way. The app's responses on sensitive political issues are aligned with China's government stance, which often leads to different results compared to AI models like ChatGPT or Gemini.