Danko Nikolic

Check out my latest article: World Models Are Not Enough: Why Intelligence Requires Concepts, Not Prediction linkedin.com/pulse/world-mo… via @LinkedIn

Proud to support this petition for a unified pan-European startup entity. Let’s unite Europe's startup ecosystems! eu-inc.org 🇪🇺 eu-inc.org

@gating_ai To give correct answers in such tasks, LLM has to recognize that this is a math problem and then has to start its math module, which means running simple old-fashioned computer code. Basically, this is LLM asking hashtag#GOFAI for help. In this case ChatGPT failed to evoke help

Here is one example why LLMs do not really reason but rather perform pattern recognition. I asked ChatGPT 4o to perform a parity function on the binary string below. The LLM miscounted the 1s in the string. It is off by one, which is enough to return an incorrect answer.

Today, there will be a special event on the new AI technology designed to replace deep learning. The event will take place at IBM offices in Frankfurt and will be streamed. The technology is called Gating. Watch the event: meetup.com/artificial-int…

@davidpoeppel @ESI_Frankfurt en.wikipedia.org/wiki/Sayre%27s…

I'm leaving @ESI_Frankfurt. I’ll miss the huge group of kind smart hardworking 95% who make it so promising. But relieved to leave the handful of people whose behavior does not meet minimal standards of morality. In admin & senior science, a few truly *appalling* people. 🧵 soon.

Check out my latest article: Why does deep learning readily hallucinate? linkedin.com/pulse/why-does… via @LinkedIn

AI technology that may replace deep learning sometime soon. youtu.be/71gR9HT76m8?si… via @YouTube

It has to be repeated over and over again: GPT does not understand the world. It only gives us statistics of words. Here is a post explaining of why GPT does not understand much and is only giving us the statistics of likely words to follow other words. blog.metamirror.io/fastest-mammal…

I am happy to announce that this evening at 6 pm CET I will give an online talk about next-generation technology for AI. The technology is based on the recent advancements in neuroscience. Please follow this link: techtrust.ai/neuroscience-a…

Transfer learning is likely the most ubiquitous method for accelerating the training of neural networks. There is now a new method called Guided Transfer Learning, which transfers, not the weights, but the dimensions on which the new training should focus: arxiv.org/abs/2303.16154

Our new algorithm Guided Transfer Learning was described by a talented writer at Towards Data Science. So nice to see such explanation clarity. I am curious to see how much change we will be able to inspire in the ways how we train neural nets. towardsdatascience.com/guided-transfe…

@Grady_Booch LLMs do not follow the theory of practopoiesis.

Our AI Meetup is comeing again. This time our guests are Accenture and Bayer, with great speakers. Click here to join: meetup.com/artificial-int…

@KatherineAMunro Congratulations!

Today a years-old dream came true: I moved to #Switzerland! Took the night train, woke up to this view, and knew I'd made the right choice. To all the dreamers: go do the thing! 🎉🎉🎉

AI Meetup Frankfurt // Freezy February meetup.com/artificial-int… #Meetup via @Meetup

@dankonikolic joins @qpclub1 in @Clubhouse to talk about his latest and most important work w/ @ceciletamura , @sierra_photon , @x_abyss , @BobprayD 7 a.m. PST, 10 a.m. EST. clubhouse.com/event/xp4nn9Aj…

My most important scientific paper just got published: authors.elsevier.com/a/1gVvg5Fq7aXe… The implications for AI are tremendous. We have been doing it all wrong. Here is what we have been missing.