AI: The End of Accounting, or of All Personkind? By Eric Cohen
I promised an update on trends related to OpenAI’s ChatGPT (and its licenses and competitors) in my 23 March blog post, and in the article that appeared in the latest ThinkTwenty20 journal, speaking to the imminent (and now accomplished) release of the Cloud Security Alliance’s publication, Security Implications of ChatGPT, to which I contributed.
For those who have escaped learning about ChatGPT so far, it is an application of artificial intelligence purposefully designed to be very accessible to anyone who can type and read. I like to say it is like talking with a patient and very knowledgeable person that can converse on virtually any topic. It is fallible; it sometimes makes things up; most versions are not aware of facts that have taken place after 2021. It is not in itself a search engine to a database, but a “large language model” that can pull information from its training to generate responses. It is a tool that can provide exceptional results to those who know how to interact with it, which has found great success in a wide variety of uses.
The accessibility of ChatGPT has led to users trying to use it for purposes where other implementations and categories of AI may be more suitable. However, the impact it has had, and the visibility to the potential benefits and risks has made it a powerful and disruptive application.
Here are a few observations of what has transpired since my last article.
- An assessment by a large group of accounting academicians ran CPA exam-like assessment questions passed ChatGPT. It did not pass. However, at least one of the participants noted that GPT 3.5 was used for their testing, and not GPT 4, and found GPT 4 far superior at the assessment.
- Interest in generative AI (the “G” of “GPT”) has led to estimates that 70% of organizations surveyed by Gartner are now in exploration mode, and 45% of executives say it has prompted an increase in their AI investment.
- Italy lifted its ban on ChatGPT. Italy had banned ChatGPT primarily over privacy concerns. OpenAI changed its data usage procedures and policies at the end of April, offering the option to exclude conversations from training the AI, providing an export function for history, and working on a new “don’t train on our dialogues” option for business. How this will impact investigations into ChatGPT for privacy issues by other regions, such as Canada and the European Union. Bans remain in place in a handful of other countries, most of which have strict internet censorship laws and regulations.
- Experts with concerns continue to express concerns that there are immense risks and dangers.
- The number of signers of the petition to pause the development of more powerful AI technology has reached 27,565.
- The so-called “Godfather of AI”, Geoffrey Hinton tweeted that he left his role at Google so he “could talk about the dangers of AI without considering how this impacts Google.” He likewise posted that he predicts AI will become smarter than a human in 5-10 years, but notes he could be entirely wrong … and that’s why “we should worry now”.
- While some extreme critics shout that the next evolution of GPT could lead to the “beginning of the end of mankind” with plans to destroy humanity, others point out that more likely scenarios are job losses (threatening up to 18% of the workforce), misinformation and increased equipping of evil doers are the largest risks.
- Google’s competitor, Bard, is now available to everyone in 180 countries with no waitlist. They say the ability to connect Bard with Google’s other platforms (you can easily connect to Docs and Gmail now) , creating of images like Dall-E-2/Bing Image Creator, and other features are on the way. Bard says it can understand English, Japanese and Korean, and is working to get to 40; they do support 20 programming languages. ChatGPT supports more than 90 languages, and has been trained in at least a dozen programming languages – as I have written before, it has also helped me with the native scripting languages of apps like the flowcharter Mermaid.
- Microsoft has also made Bing AI Chat, based on GPT, available to all. The GPT 4 engine is available on one setting (“more creative” mode), while 3.5 is available for “more balanced” and “more precise” modes. They have also made their version of the image creator, Bing Image Creator, based on OpenAI’s DALL-E-2. While the guardrails around Microsoft’s implementation of GPT 4 do not permit all of the capabilities, such as working with digital files such as images, it has made the more powerful and seemingly intelligent version of GPT available to a broad community.
- There are thousands of adds on and other uses of GPT now available. One of my favorite is called “Call Annie”. You can now chat (with your voice) with GPT4, using a telephone, an iPhone App, or the Web. The developers have layered a very outgoing and friendly front end to GPT 3, so you can speak with Annie as an expert (however flawed) and guide on virtually any topic. The conversation is very smooth and natural, which can make you forget it is just a text-to-speech engine (OpenAI’s Whisper) over GPT. The friendliness may be a little jarring; my last conversation began with Annie asking, “What kind of tunes do you jam to? I'm always down to grow my playlist.”!While you can’t paste in large volumes of text, and the voice output is not suited for composing large documents, the uses of the voice interface are many (e.g., thinking through an issue while driving, rehearsing a dialogue like an interview), and the text is transcribed for retrieval at a later time.
Comments
- No comments found
Leave a comment