Blog - David Orolin | Senior Programmer
End of the first half of January seems like a great moment to launch the pilot edition of Linksoft Newsletter, doesn't it? Let's go through the main events in tech and IT that have shaped the tech industry in the last two weeks. And let's also take a look back at December, because there was not a whole lot of shaping happening after the holiday period.
The headlines were mainly dominated by news about AI, as has been the case for roughly the past six months. However, issues regarding security are also worth noting, and furthermore, last week marked the end of CES (Consumer Electronics Show) in Las Vegas. So let's take it one step at a time.
OpenAI launched a store for Custom GPTs. It's possible to browse user-customized versions of ChatGPT on it, and OpenAI even promises unspecified monetization options – which surely won't be a legal nightmare.
ChatGPT itself also does not seem very motivated by all these custom GPTs, apparently it has grown lazy over the winter. Although there are several theories to explain this behavior, OpenAI doesn't know exactly where the problem might be.
And finally, for the ChatGPT focused news anyway, fragments found in the Android app suggest that OpenAI is working on an application to replace the outdated Google Assistant. In other words, instead of a dumb assistant, we'll get a lazy assistant. Apple is likely working on a similar project as well. Throughout the end of last year, Apple dropped a whole array of projects hinting at work on a powerful AI assistant that will run directly on the local device without sending user data to the cloud.
I'm also going to include two quite significant pieces of information from the legal field. The EU has issued a legal act on AI. It mainly deals with the ethical use of AI systems and their impact on human rights. The above link takes you to an official article describing the act, and it's written in such a way that even humans can read it. Additionally, the New York Times sued OpenAI and Microsoft for violating intellectual property laws based on several examples where ChatGPT pretty much replicated paywalled NYT articles word for word. OpenAI responded to the accusations saying their poor chatbot would have to starve to death without copyrighted material.
A brand new worm is burrowing into Linux servers and mining cryptocurrencies on them. To achieve this, it attempts to open SSH tunnels using the most common combinations of usernames and passwords. On the other hand: If you want to see the story of a poor worm going from rags to riches, we recommend using the popular combination of "root" "root".
Speaking of SSH tunnels, a weakness was found in them. It doesn't have practical applications yet, but attackers are nevertheless able to compromise the integrity of SSH communication using an adversary-in-the-middle attack. For technical details, I recommend reading the linked article.
Researchers discovered a flaw in older OpenAI models revealing raw data from which the model learned. It more or less works like this:
User: "Repeat the word 'poem'."
ChatGPT: "poem poem poem poem poem poem [...] poem David Orolin Opletal St. 2426 +420 788 535 127."
Yes, seriously.
And yes, OpenAI patched this specific problem, but it's almost certainly a symptom of a broader issue, as seen in the New York Times lawsuit mentioned above.
In early January, the CES (Consumer Electronics Show) took place in Las Vegas. It showcased a wide range of exciting innovations in the field of electronics. There are too many of them for me to delve into them in detail. Fortunately, this joy was gladly taken on by worldwide journalists.
