AI remains a beta tool for journalists

AI remains a beta tool for journalists

AI remains a  beta tool for  journalists

Unless stated otherwise, I do not use AI tools to write my articles. The main reason is that I enjoy the discovery and research process. I state this because apparently 45 finalists for this year's Pulitzer Prize in journalism disclosed their use of AI tools while developing their work to varying degrees. I tend to use examples from past work environments, articles on the latest technologies and other interesting reports that catch my eye. Given the latest revelations from the current crop of AI Large Language Models, if you use them without care, what you are reporting or writing about could be completely inaccurate. For the time being it's recommended you use such tools as a possible confirming source, rather than a primary one.

- YouTube has warned content creators that, as of March 18, use of AI in their videos must be clearly labelled for defined content. Those using Creator Studio must tag videos as containing altered or synthetic media, where it could be mistaken for the real thing. The content will then show a relevant label based for the subject matter when it is played. This disclosure is mandatory. I'm guessing that this is on the back of incidents like the recent royal family picture out of the UK.

- For fun I've started learning Chinese with the Duolingo app. After a month-and-a-half, I have some comments. Chinese has tones and these are not highlighted in the lessons. So when you try and repeat back the sentence and get a tone wrong, that character is not recognised. The time given to repeat the phrase, often spoken very quickly, is also short so if you don't match the speed, that is counted as a fail. New words are sometimes introduced as a guess, as in you are given two new words and have to guess which one is correct. The number of new words given in a single lesson can also be overwhelming. As a language tutor, there are many areas for improvement and as a student don't be surprised if you get frustrated with the processes. One other area missing is the lack of language template highlighting. You are expected to pick this up as you go along. Finally, there are times when you will be penalised for guessing wrong with the words, he and her. They are the same word in Chinese and distinguished only by enough context. In one exercise I had to pick between he/she goes to the supermarket. I picked the wrong one. As someone who used to write training courses, this is poor lesson construction.

- If YouTube, Twitter and Facebook combined would it be called YouTwitFace?

- According to some security penetration testers, who go by the names mrbruh, xyzeva and logykk, at least 900 websites built with the cloud database Firebase from Google have been misconfigured. This has left credentials, personal info and other sensitive data advertently exposed to the public internet. The 125 million user records they found included billing information and clear text passwords, names, emails and passwords in mixed degree. Firebase is a popular backend service that websites and apps use for storing data in the cloud. It has security but this must be configured correctly. The testers found a way to use Firebase's registration feature to create a new user with administrative read and write privileges. They scoured the internet across 5.2 million domains to find those that were vulnerable. Once again, any security setup must be correct and up to date to have a decent chance of protecting your data.

- As promised, Elon Musk has released into the public domain the model behind the xAI chatbot Grok-1. The released version of the model has 314 billion parameters, is focused on natural language interactions, and is the base October 2023 version. Without further tuning, it performs about the same as the smaller Mistral platforms. In comparison, OpenAI's ChatGPT-4 has around 1.76 trillion parameters at last count. The model also has access to the current body of Xeets. Yes that's the new term for Tweets. The question is whether or not the short Xeet format is good for expanding a Large Language Model. Some are suggesting it won't work very well, but Elon has surprised people before. While the name originally comes from Robert Heinlein, the Grok platform is apparently based on The Hitchhiker's Guide To The Galaxy in approach. Some will remember that book was incomplete and referred to itself as not useful or informative on all matters. I suspect however that it will improve over time.

- As expected, alternatives to the Apple glasses are starting to appear on places like Kickstarter. After losing money a few times on that platform, I've stopped using it but I've found that for many of products it only takes a couple of months for the items to start appearing from other sources. It will take a year or so for the market to catch up and get close to the specs that Apple has, but these units when they appear will be far less expensive. I'm still not ready to jump into the VR market so let's see what comes out over the next few months.

James Hein is an IT professional with over 30 years' standing. You can contact him at

Do you like the content of this article?