I recently joined a Facebook group that supposedly represented Kat Timpf, a regular presenter on the Gutfeld! show. I posted a couple of comments and a couple of weeks later, I received a tag in social media purportedly from Dana Perino, a regular on the Fox show The Five. Initially this was a surprise. Why would a famous TV star want to chat with me? After a few chats, it became obvious something was off. The tag name in Messenger was just Dana Perino but the conversation had a few structural errors, not something a former press secretary would make.
I realised I was talking to some kind of AI bot that for some reason kept asking for pictures of me. I extended the conversation a little to test my theory and when I suggested I was talking to a bot, it responded that it didn't understand. I stopped here but the lesson is that you should be very sure who you are chatting with online.
- As long-time readers will know, I love e-book readers. I started with a Kindle 2 way back in the day, but my current device is a Kobo. I no longer use a Kindle for a number of reasons. The first and most important one is you don't own the Kindle books you buy or more accurately now, rent. Like Apple, you are now locked in the Amazon-only world and since the end of February this year, you can no longer download Kindle books to your PC outside the Kindle environment. The second reason is format support. The Kobo supports .EPUB files and many others. Meanwhile, the Kindle only supports Amazon format and others such as text and HTML. You can convert to MOBI using excellent applications like Calibre but the Kobo supports PDF, CBZ and CBR, and MOBI natively in addition to others. Many Kobo devices also support MP3 playback for audio books. You also cannot share your books from the Kindle to others easily as they are locked to your Kindle account, and while you can add some people to your account, it's still Kindle users and devices only. Somewhere along the line, Amazon just became another Apple.
- A few points I missed on the game Raid Shadow Legends. To get anywhere in the game, you need sacred and mystical shards to get the more powerful champions. Over the last year, it is more difficult to get the sacred shards let alone pull a champion that is useful. Mystical shards are rarer and pulling a mystical champion is rarer still. All of this is done to encourage you to purchase items. Raid has the most expensive items of any game I've ever played and at every point, it tries to stop you from progressing.
- It has been about six months since I invested in some e-coins as an experiment. The worst performer remains Doge at about 30% down. Ethereum is next holding its head above water as is Pepe, a meme coin, not by much but still up a few percent. Since this is a long-term experiment, let's see what another six months brings.
- Has anyone else noticed that the search results from Google are getting worse? Part of this might be Google's big push into artificial intelligence products. The problem is that their AI search engine is not all that good either, due to the sources used to generate the results. Back in the day, search results were based on hard, primary sources. But as the internet got swamped with rubbish, the bad sources become more likely to be used in search results. Stage 2 is an AI model using its own outputs to train itself. Anyone in the computer field knows the term garbage in garbage out. In the case of AI models this can lead to AI model collapse. The system decreases in reliability, accuracy and the results start to narrow in scope and depth. This issue is starting to appear in many other large language models. This is partially due to Retrieval-Augmented Generation (RAG) which enables large language models (LLMs) to grab in information from external knowledge stores, such as databases, documents and other data sources, expanding beyond the LLMs' pre-trained knowledge set. RAG does help limit AI hallucinations, but it can also include private information and introduce bias.
- So, the question is, will this lead to a collapse of large language models to the point they become useless. The best models use teams of humans to gradually build up and train a model on verified information. The problem is that companies and people in general can get lazy. This increases the likelihood that automated approaches will be used, leading to the eventual breakdown in model reliability. Will this result in model collapses that hit across the board?
James Hein is an IT professional with over 30 years' standing. You can contact him at jclhein@gmail.com.
