Friday, April 28, 2023

KPMG How professionals are using ChatGPT

 Raskolnikov: Yes, Tax The Rich—And Also The Merely Affluent


Inside the secret list of websites that make AI like ChatGPT sound smart


How a Google Antitrust Case Could Determine the Future of AI Matt Stoller, BIG. Musical interlude


Google’s Rush to Win in AI Led to Ethical Lapses, Employees Say Bloomberg:

Shortly before Google introduced Bard, its AI chatbot, to the public in March, it asked employees to test the tool.

One worker’s conclusion: Bard was “a pathological liar,” according to screenshots of the internal discussion. Another called it “cringe-worthy.” One employee wrote that when they asked Bard suggestions for how to land a plane, it regularly gave advice that would lead to a crash; another said it gave answers on scuba diving “which would likely result in serious injury or death.”

Google launched Bard anyway. The trusted internet-search giant is providing low-quality information in a race to keep up with the competition, while giving less priority to its ethical commitments, according to 18 current and former workers at the company and internal documentation reviewed by Bloomberg.

Hard to believe. Google?


Artificial Intelligence in the Garden of EdenPeggy Noonan, WSJ

Personalized AI-Written Spam May Soon Be Flooding Your Inbox

Gizmodo: “…Now, the arms race between spam blockers and spam senders is about to escalate with the emergence of a new weapon: generative artificial intelligence. With recent advances in AI made famous by ChatGPT, spammers could have new tools to evade filters, grab people’s attention and convince them to click, buy or give up personal information. As director of the Advancing Human and Machine Reasoning lab at the University of South Florida, I research the intersection of artificial intelligence, natural language processing and human reasoning. I have studied how AI can learn the individual preferences, beliefs and personality quirks of people. This can be used to better understand how to interact with people, help them learn or provide them with helpful suggestions. But this also means you should brace for smarter spam that knows your weak spots – and can use them against you

From pope’s jacket to napalm recipes: how worrying is AI’s rapid growth? Google boss says issue keeps him up at night, while thousands have urged six-month pause on creation of ‘giant’ AIS

How profesionals are using ChatGPT

Edmund Tadros

The release of ChatGPT has brought into the mainstream a branch of technology that will transform the way white collar work is done around the world.

Experts say workers need to learn how to use these generative AI tools but the challenge is trying to get a foothold, especially for the non-tech savvy, in an area that is developing so quickly.

This blog will focus on how professionals from around the country are using the tools to help them do their work. It will spend time examining how the technology works only when it relates back to the specific examples.

It will also feature short interviews with workers from around Australia describing the specific ways they have used these tools and the pros and cons of their experiments. We’ll also highlight the specific prompts used, the outcomes (redacting any sensitive information) and our take on it.

We’re particularly interested in highlighting how people without technical backgrounds are making imaginative use of the technology.

This will be an evolving experiment and will at first focus upon how the ChatGPT tool is being used.

We welcome submissions at with “AI Blog” in the subject line. Please include your contact details (not for publication), the specific use case, the pros and cons of the example, the prompt(s) used and the material produced by the tool.

Extracting data for PDFs using ChatGPT

Edmund Tadros

Brisbane-based Accountant Jason Andrew has been using ChatGPT to create emails, edit blog posts and general life admin.

But what got our attention was the way he has used the tool to take much of the hassle out of the dreaded task of extracting data from a PDF file.

Accountant Jason Andrew is using ChatGPT to extract data from PDFs. 

“I recently used it to help with data cleansing and restructuring copied data from a PDF into Excel,” Mr Andrew, the director and head of growth at SBO Financial, said.

“I would typically task our analyst team to do this spreadsheet grunt work – it would probably take them at least an hour. ChatGPT did it in seconds.”

Asked if he would use the tool to carry out such a task again, he replied: “Hell, yes.”

ChatGPT 3.5 prompt:

hey mate, can you reformat this data for excel

30 June 2022 30 June 2021
as at 30 June 2022 Note $’000 $’000
Current assets
Cash at bank 4.5 24,276 70,497
Trade and other receivables 3.1 27,687 30,621
Inventories 3.2 65,335 46,179
Prepayments 5,026 4,311
Contract assets 2.1 844 843
Current tax asset 3.6 8,487 -
Assets held for sale 5.2 657 -
Total current assets 132,312 152,451
Non-current assets
Right-of-use assets 3.3 41,268 9,003
Plant and equipment 3.4 38,068 19,864
[Edited for length]

The upshot

The ease or difficulty of manually extracting data from a PDF will all depend on how the data was saved – and whether you have access to Adobe Acrobat Pro or a similar specialist tool that can be used to cut and paste tables.

Without these tools, ChatGPT 3.5 and 4 can do quite a bit of the heavy lifting, – but you’ll still have to double-check the results carefully and ask the chatbot to correct its mistakes.

We had mixed results in our experiments using the same table as Mr Andrew (a cut and paste from page 48 of the annual report below). The tool sometimes had to be reminded to include all the columns and at other times cut of data for no reason. However, ChatGPT quickly fixed the problems when prompted.

The tool was less useful for tables with blank cells that did not have a dash or some other indicator. This data in these tables was wrongly placed and in each test had to be manually corrected.

KPMG turns to AI to improve its audits

Edmund TadrosProfessional services editor

KPMG Australia has begun using an enhanced version of its auditing software on local clients which can search for outlier transactions within a complete dataset during an audit.

Julian McPherson, partner in charge of national external audit and assurance at KPMG. 

The capability, already in use by the firm in the US, UK and Canada, allows KPMG’s auditors a faster analysis of more client data and should improve audit quality, which is critical to investors who rely on financial statements to make decisions.

The enhancements to KPMG’s audit software, called KPMG Clara, uses aspects of artificial intelligence in its operation, such as machine learning, to work out what constitutes a normal transaction within a dataset.

This use of advanced technology comes after the corporate regulator last October called forKPMG and its big four rival Deloitte to take immediate action to improve their audit processes after it found their standard of audit work had declined dramatically in the previous year.

It is also part of an ongoing trend within the world’s largest accounting firms and will have long-term implications for the usefulness of audits as well as the professionals who work within the audit sector.

‘Analysing 100pc of transactions’

KPMG is initially using the enhanced software on five audits in Australia with “some teams doing [a traditional] audit in parallel” to test the effectiveness of the machine learning functions, said Julian McPherson, the partner in charge of national external audit and assurance at the firm.

The new features work by “analysing 100 per cent of transactions” and then scoring each transaction against a battery of tests seeking to uncover statistical outliers, he said. “Every transaction is given a score” as high, medium and low risk, allowing the firm’s auditors to focus their “attention on high-risk transactions”.

“The solution was able to identify high-risk transactions that wouldn’t have been able to be identified normally,” Mr McPherson said.

The KPMG partner said that the new capability would not, at first, reduce the time or manpower required to carry out an audit because of the amount of work initially involved in setting up the data feeds to allow the software to analyse entire datasets.

“My team has grown, we are now recruiting more data scientists,” he said.

AI and auditing

In the long term, researcher are split on what the increased use of artificial intelligence technologies will mean for white collar jobs and for auditors specifically.

Two recent reports looked into the effect of AI chatbots, such as ChatGPT, on white collar jobs. They concluded that while many office tasks will likely be sped up or completely automated by AI tools, it was too early to predict if that will lead to net job losses around the world or simply change the nature of many jobs and create a range of new occupations.

A separate 2022 research paper published before ChatGPT was made public – Is artificial intelligence improving the audit process? – concluded that “investing in AI helps improve audit quality, reduces fees and ultimately displaces human auditors”.

The study examined the use of artificial intelligence by the world’s 36 largest audit firms found that increased investment in AI-related technologies would reduce the likelihood of an audit restatement, lead to lower audit fees and reduce the number of accounting employees.

“Our empirical results document substantial gains in quality, as AI investments by audit firms are associated with significant declines in the incidence of restatements, including material restatements and restatements related to accruals and revenue recognition,” the authors of the paper wrote.

“[Moreover], we find preliminary evidence that improved audit quality is accompanied by a move towards a leaner process: as audit firms invest in AI, they are able to lower the fees they charge while reducing their audit workforces and showing increased productivity, as measured by total fees per employee.”