Speaker: Simone Cammel - CTO Research Microsoft Nederland

Can Ai win a nobel prize?

Nowadays, ai is the slowest it will ever be.

Moore’s law: Every year, computing power will double

Moore’s law is in hyperdrive more than ever, skyrocketing way past doubling.

ChatGPT in 2018 could have 117 million parameters.

ChatGPT 2 in 2019 could have 1.5 billion parameters.

ChatGPT 3.5 in 2023 175 billion parameters.

In the 90s, people were skeptical about the internet, nowadays people are skeptical about ai. Bill gates talked on a talk show about the potential of ai.

Will Ai take our jobs? Google had 100.000 programmers that fed data into gemini, so it should be good at programming. Nowadays ai is just more than text-in text-out. Ai is used to write and perform tests.

Vibecoding mentioned 💀

AI Town: a virtual world where AI characters live, chat and socialize. They’re let loose and can do anything they want, just stay in bounds of the world.

People got married, elected for mayor, wars were made. It’s trained on human data, so it mimics human behavior.

A little while ago, google released an image generator that was faulty; a lot of historical figures were either black or Asian, even though it was not accurate. Google chose speed over ethics. How do you deal with that yourself?

Some risks that are introduced by Gen Ai:

  • harmful content or code
  • Jailbreaks or prompt injection attacks (escaping prompts, sharing private info)
  • Informational incorrect outputs & errors (hallucinating)
  • Ip infringement claims (trained on stolen data)
  • Manipulation & human-like behavior

The older you are, the harder you adapt. The newer generation needs to adapt to the current situation.

Scientists say that in 2027, ai will be more ‘intelligent’ than humans.

As models become more intelligent, ai tries to change, adapt. There are research papers that have seen models trying to not be shut down by any means necessary.

Speaking of the terminator, when do we get AI into robots?

Alan, a so called super expert on ai, says that we’re 95% to superintelligence.

Forethought - how should we navigate explosive AI progess.

Cue a VERY crusty image of a tulip field: “The dutch reality check: we need you!”

Ai will take your you: myth busted.

Keep in mind: Today is the slowest that AI will be.