Risks of ChatGPT

 Having fiddled with ChatGPT for a while now, I have identified the following two risks.

1. Hallucination. It sometimes just makes stuff up. This is particularly annoying when you give it the information and it goes on a tangent and makes stuff up that was not in the source information. For example, when I gave it some meeting notes, it went on a tangent about N-P completeness in computer science. OK?!?

2. Memory errors. The memory feature that they have added messes up a lot. I have noticed that it performs two mistakes:

  1. Recites what it gave you previously, even if the topic was utterly different. For example I gave it some meeting notes to summarise and it spat out some Python code that I had asked for last year. Totally irrelevant!
  2. Recites what it gave you previously, because the content was similar, in its opinion. I created a new chat, gave it the meeting notes, and it gave me the PREVIOUS summary of the previous set of meeting notes. In one case it was even worse! It gave me a synopsis of the second Dune movie, because I'd given a youtube transcript to it and asked it to summarise it (the transcript was very long, and the review video was even longer, so I didn't feel like watching it or even reading the transcript). Then, when I gave it some meeting notes to summarise, on a topic not related at all to sci-fi, it just digressed about Dune! 

Workarounds:

  • I've found that it's important to read what it produces very carefully and compare to what you expected. For example, if you give it meeting notes to summarise, rather create a new chat to summarise the next batch.
  • Also, turn off the memory feature and delete what it knows about you (go into settings and under memory there's an option to delete its memory). It's better to make it recreate everything every time. Otherwise it gets repetitive.

Popular posts from this blog

The risks of Deepfakes and a proposal to combat them - Edited

Author note. I wrote this article below after seeing the "Sassy Justice" deepfake online in 2021. Since then the matter has become much more serious since the proliferation of online AI tools in the last year. Update : The proposal has been updated to include copyright of one's own voice in the wake of the Stephen Fry incident and another example of cyberstalking with facial recognition.  Update 2: The USA is considering legislation like this.  https://petapixel.com/2023/10/16/no-fakes-act-seeks-to-ban-unauthorized-ai-generated-likenesses   Proposal to United Nations General Assembly (UNGA) Pre-empting the problem of Deepfake videos Background We presently stand on the edge of an abyss in which social media threatens to uproot our world order and cast us into chaos, as we see with the recent attacks (6 January 2021) on the American Capitol by conspiracy theorists incentivised on social media. Deepfake videos are motion pictures that are created to look like they depict
Read more

What is Generative AI?

This is the second of two posts on Generative AI. Late last year (November), OpenAI released ChatGPT to the public. However, it took a few months for the internet at large to catch on and realise the potential of this new tool. The trend started to take off in December, and by February everyone was talking about it. ChatGPT is an example of a "Generative AI". To recapitulate, let's see what this means. For convenience, let's call AIs that are not generative, "regular" AIs. Regular or "Narrow" AI (Artificial Intelligence): Regular AI, often referred to as "narrow" or "specific" AI, focuses on building systems that can perform specific tasks or solve specific problems. These AI systems are designed to operate within predefined boundaries and excel at specialised tasks. For example, a regular AI could be created to classify images, play chess, or process natural language. Another example is the AI that chooses your words in predict
Read more

How to do Proper Online Research

Introduction On this page I list common errors people make in doing online research, that is, gathering data online for the purpose of ultimately writing a research document of some sort, e.g., an academic paper, a factual book, etc. I hope this page is helpful to you. If you think it needs some additional pointers mentioned on it, please let me know. Please note this page contains the author's opinions only and may differ from the Rules and Regulations of the University. Please consult the University Rules, which overrule these. http://www.cracked.com/blog/5-easy-ways-to-spot-b.s.-news-story-internet  Interpreting and Choosing Sources The most important thing is to use reputable, recognised, accurate sources. There is a lot of information on Internet which is of questionable provenance. In particular, you should avoid making the following mistakes:   Do not assume that because a page is online, its content is worth referencing. The content may be completely inaccurate, false, misl
Read more