From the course: Post-Quantum Security for AI: Resilient Digital Security in the Age of Artificial General Intelligence and Technological Singularity by Pearson

Security implications of accelerated AI development

From the course: Post-Quantum Security for AI: Resilient Digital Security in the Age of Artificial General Intelligence and Technological Singularity by Pearson

Start my 1-month free trial Buy for my team

Security implications of accelerated AI development

“

We looked at the figure of how artificial intelligence and human level intelligence are somehow changing the level of development. So the human level of intelligence is fixed, unless we can somehow merge the cognitive capabilities with machines. the emergence of a quantum computer changes that shift even if we had technology that we could connect the human brain to the machine. So this is the security list that we discussed in the previous slide and this is an updated figure that I have created of the technological singularity event as you would see here. The event in my opinion is much further up the scale because in the original prediction from 1950s we should have already crossed that line and we haven't so the singularity has not occurred yet. So we need to look at singularity happening in the future and we're discussing what kind of capabilities would enable the artificial intelligence to become better than humans in all of the four categories that we discussed. And one interesting concept is that is on August the 7th 2025 OpenAI announced the GPT-5, which is by now an obsolete model. Now we have GPT-5.1. And that was promoted as a PhD level in reasoning, coding, and writing. And at that stage, I had to test it as somebody with a PhD who trains PhD students. I had to test it for myself to see if it's really a PhD level of coding, reasoning, and writing. And the GPT-5 was far, far, far better than GPT-4. And is it better than PhD level in reasoning? That depends on the individual PhD student because you have sometimes few outliers, two or three students that they're exceptionally better than their rest of the peer group. Was GPT-5 at that stage better than those few outliers? No. Was it better than the average PhD student that uses AI in order to predict some kind of analysis? Well, it was close. It was very close to that level. And even when you have VS Code and you're running a genetic mode, GPT-5 was capable of resolving most of the problems within your coding environment. So if you're looking at best generative AI now, which one is better, Cloud, GPT, ChatGPT, OpenAI? Well, ChatGPT is considered best overall for versatility. Cloud is better for handling long documents. Google Gemini 3 has by far outperformed the Gemini 2.5 and it's strong in real-time information access. So the answer to what is the best generative AI now, the answer is when do you watch this video? Because that changes, that is not static. Maybe ChatGPT 5 was best at the time of August the 7th. Today, it's possible that Google Gemini it's better at specific tasks than GPT-5. And mid-journey, it's still used for high-quality image generation, but many of those technologies are becoming obsolete every day. So with that in mind, we need to look at some of the reports that are coming up in terms of performance in the real world. Now, Elon Musk has already developed an implant that is connecting the pig brain to a computer. So this is called Neuralink. And many people are critical about this kind of technology. And many people say that this won't really resolve anything. But as we have seen from the PowerPoint slides, unless we can connect the human level to the machine, human brain to the machine, artificial intelligence is gonna outperform human level intelligence much quicker than we think. And also, Neuralink has promising technology for disabled people. It's not only for people with normal cognitive capabilities, there are people who can use this kind of technology today, even if they're not interested in artificial intelligence. This kind of technology in connecting the human brain to the machine would provide solution for many of the physically disabled people. Now, I do have an article on that topic on how technology needs to be accessible and inclusive. So all the users would need to have accessibility to this kind of technology. But I have to say that the actual useful technology, it's not at the level that we would be able to recommend it to a person with physical disabilities to use it. The technology will continue to grow, but it's not there yet. When we're talking about GPT-5, not only that is close to a PhD level of performance, but actually GPT 5.1 might be exceeding some of the PhD level of reasoning, coding, and writing. And we're measuring that based on capabilities, we're measuring that based on benchmarks, and we're not really measuring it against one individual PhD student against the rest, we have benchmarks for that. you

Contents