By Christian Prokopp on 2023-02-11
Microsoft could follow Google's $100bn loss. I tried the new Bing Chat (ChatGPT) feature, which was great until it went disastrously wrong. It even started arguing with me while being wrong and making source code up.
To try Bing Chat, you must get on a waiting list, and then you are forced to use Edge. On a positive note, the experience is better than ChatGPT's version. Bing Chat is snappier, shows what is searched in the backend, and gives suggestions on continuing the conversation and the occasional references supporting its code and claims. Or so I thought.
I asked it to code a class to query Athena with Python. It looked good at first, but things went bad when I asked it to stream results into a feather file. In particular, it used an 'append' flag with the 'pyarrow.feather.write_feather' method, which is not in the documentation and the references it produced. One reference it produced used 'fastparquet.write()', which has an append flag. It may have confused the two.
When I gave Bing Chat a chance to correct itself, I was surprised when it wrote, "I am not wrong.", and continued with source code to prove its point only to prove itself wrong unknowingly. To top it off, when I asked to show me where it got the code from, it directed me correctly to the 'pyarrow/feather.py' source on Github. But there, I found that the source code differs completely from the one it showed me.
In summary, Bing Chat generated code for me. And:
That is devastating. It produced incorrect code, failed to understand its mistake and faked source code with references. It did everything it could to throw me off and get things wrong.
Christian Prokopp, PhD, is an experienced data and AI advisor and founder who has worked with Cloud Computing, Data and AI for decades, from hands-on engineering in startups to senior executive positions in global corporations. You can contact him at email@example.com for inquiries.
Large-language models (LLMs) are great generalists, but modifications are required for optimisation or specialist tasks. The easiest choice is Retr...
OpenAI's DevDay announcement yesterday addresses issues I wrote about in the infeasibility of RAG after building Llamar.ai this summer. Did I get i...
Over four months, I created a working retrieval-augmented generation (RAG) product prototype for a sizeable potential customer using a Large-Langua...
Get the insights you need to compete with Amazon with a comprehensive dataset.
ChatGPT can combine Data with natural language and has extensive information about most subjects. That lends itself to novel applications like crea...
Insurance works because it shares costs in the face of uncertainty. What happens when Tesla removes uncertainty and distributes cost seemingly more...