The first study into the workplace impacts of generative AI platforms has officially been published, and it’s good news for those relying on it to boost productivity.

Generative AI platforms like ChatGPT and Google Bard have been all the rage in recent months, with businesses across the world using the technology to perform tasks like coding and content creation. While the tech is obviously helpful, studies hadn’t yet been conducted to measure its effectiveness in helping employees get their jobs done.

However, despite only being live since November, a new study has emerged showing that worker productivity has seen a 14% boost in companies utilizing generative AI platforms.

Generative AI at Work

The study from the National Bureau of Economic Research, titled Generative AI at Work, found that the technology improves productivity by 14% on average.

“Progress in machine learning opens up a broad set of economic possibilities. Our paper provides the first empirical evidence on the effects of a generative AI tool in a real-world workplace. In our setting, we find that access to AI-generated recommendations increases worker productivity, improves customer sentiment, and is associated with reductions in employee turnover.” – the authors of the study

Productivity in the study was “measured by issues resolved per hour,” analyzing the work of 5,179 customer support agents from an anonymous Fortune 500 software company.

How Does Generative AI Improve Productivity?

If you’ve had a chance to utilize ChatGPT or any other generative AI platforms at your business, you likely understand how they could improve productivity on the first day.

However, this study attempted to create an environment in which the tech could be objectively studied to allow businesses to understand exactly how it can be used to improve productivity.

“This increase reflects shifts in three components of productivity: a decline in the time it takes an agent to handle an individual chat, an increase in the number of chats that an agent is able to handle per hour (agents may handle multiple calls at once), and a small increase in the share of chats that are successfully resolved.” – The authors of the study

On top of that, the study notes that generative AI replicated that work of the firm’s top performers, which means that the biggest productivity improvements came from the novice and low-skilled workers. Conversely, generative AI didn’t have a massive impact compared to the firm’s high performers, which means we wouldn’t recommend going around and laying off your whole team just yet.

What Is the Best Generative AI Platform?

If this research has swayed you to give generative AI a try at your business, we don’t blame you. Plenty of businesses would be more than happy with a 14% productivity increase and given the low cost of entry for generative AI at this juncture, it seems like a bit of a no-brainer. So, which generative AI platform should you use?

Right now, generative AI is a two-man race between ChatGPT and Google Bard, which both have some serious tech resources behind them to get the ball rolling in Microsoft and Google, respectively. Both perform admirably for the majority of tasks and should help your business improve productivity.

However, in our research, we found that ChatGPT is a tad better at some of the more important business functions, like idea generation, summarizing extracts, and creative flair. Still, given the two platforms — and their many generative AI alternatives — are always learning, the best generative AI platform could be right around the corner.

The tech world’s AI gold rush might have been a little hasty: Google ignored employee warnings and even overruled an official risk assessment in order to launch its AI chatbot Bard, reports claim.

The alleged warnings took place in internal messages seen by thousands in February 2023, while the risk assessment happened a month later in March, the same month in which Bard debuted to the public.

Google’s Bard is one of the highest profile chatbot AIs, alongside OpenAI’s ChatGPT. Both have faced plenty of public scrutiny. One of the biggest claims against the tools is that they lie regularly and convincingly, a potentially disastrous downside to the tech that aligns with the same warnings that Google reportedly ignored.

Google Staff Questioned Bards Readiness

The new reports, out from Bloomberg, say that employees were allowed to test out Bard in February and many issued warnings that the tool was not ready for a public release.

Employees reportedly stated that Bard was a liar, and not ready for launch in its current state, flagging major failing of the AI.

One employee warning reportedly stemmed from their conversation with Bard on how to land a plane and how to scuba dive — two potentially deadly activities that don’t have a lot of room for advice from, say, pathological liars.

The employee’s conclusion, reportedly, was that Bard’s answers could result have a harmful impact on users if followed to the letter.

Risk Assessment Was Reportedly No Match for AI Race to Launch

A risk assessment in March was also overruled, Bloomberg says. Google’s push to release its own generative AI bot was at least partially spurred by a competitor: OpenAI’s ChatGPT and its integration with search rival Bing.

Another reported claim from Bloomberg holds that senior Google leadership sent out a “code red” in December 2022 to accelerate its AI programs, leading to ethical compromises.

Doublecheck Your AI Conversations

The takeaway for anyone dabbling in chatbots — whether they’re from Google, Microsoft, or any other of a handful of competing free and paid services — is to take everything the bots say with a grain of salt.

All generative AIs are pulling from databases of knowledge, but they don’t always mix and match the information they receive correctly. The full list of issues to worry about is even larger, too: Chatbots should be checked for plagiarism, copyright violations, and poor mathematical skills, among a few other concerns.

AI chatbots certainly can contribute plenty of benefits in plenty of cases. Just don’t rely on their advice the next time you go sky diving.

Just a few years ago, for a whole class of knowledge workers, the thought of being replaced by AI was unthinkable. But the meteoric rise of ChatGPT has led many to ask themselves the same question that those working in manufacturing have asked themselves for many years now: will my job be replaced with AI?

With businesses discovering new ways to use ChatGPT every day, you can understand why workers in all sorts of roles are looking over their shoulders. With AI tools worming their way into the day-to-day routines of employees and solving all sorts of problems, it’s unsurprising that over two-thirds of employees are hiding the fact they use it from their bosses.

We’ve spoken to AI experts, entrepreneurs developing AI tools, and businesspeople using them to find out exactly which jobs they think will be automated first, how AI might make things easier, and how you can make yourself AI-proof for the job market. In this article, we cover:

• Is AI Replacing Jobs?
• Top 10 Jobs Most at Risk of Being Taken By AI
• Will AI Just Make All Jobs Easier?
• Jobs that Won’t Be Taken By AI
• Will AI Create More Jobs Than It Replaces?
• How to Make Yourself AI-Proof

Is AI Replacing Jobs?

It’s unlikely that AI will suddenly replace huge swathes of the human workforce by the end of this week, but the idea this sort of change may happen in the near future has never seemed more realistic than it does at present.

A report published by the World Economic Forum in 2020, for example, predicted 85 million jobs will be replaced by AI by the year 2025. Mckinsey’s more conservative 2021 estimate, on the other hand, is 45 million by 2030.

Our own Tech.co research has also found that 47% of business leaders would consider using AI instead of hiring for roles at their company.

But arguably, these estimations now seem too small. In the wake of ChatGPT’s explosive popularity, Goldman Sachs revealed its predictions that around 300 million jobs could soon be replaced by artificial intelligence or automated in some way or another.

At the moment, AI is competently completing tasks within job remits, rather than subsuming whole job roles. But it’s not difficult to imagine how this may progress to entire positions being taken over by artifical intelligence.

Indeed, rather than entire roles disappearing into thin air and being replaced with a chatbot like ChatGPT, What we’re more likely to see in the next year or so is AI being used as a tool to aid some duties performed by an increasingly large class of workers.

OpenAI predicted in a recent paper that artificial intelligence will soon be able to help with around 10% of tasks for eight out of 10 workers, while 19% of workers will be able to use AI to complete 50% of their tasks. This sort of spread suggests AI replacing jobs will be a slow, gradual process rather than a sudden change.

The Top 10 Jobs Most Likely to Be Replaced by AI

We’ve put together a list of jobs most vulnerable to AI. 10 jobs that could be replaced by AI are:

1. Entry-level Admin Roles
2. Data Entry Clerks 
3. Software Engineers and Coders
4. Customer Service Reps
5. Paralegals
6. Copywriters and Content Roles
7. Graphic Designers
8. Bankers and Accountants
9. Traders
10. Fact-Checkers and Proofreaders

1. Entry-Level admin roles

Administrative roles are perhaps most at risk of being replaced by AI tools currently available, such as ChatGPT. Admin roles consist of tasks like note-taking, proofreading, writing summaries, and managing spreadsheets, jobs that can already be performed competently by ChatGPT.

In the same research note that revealed Goldman Sachs’ prediction that 300 million jobs could be replaced by AI back, “office and administrative support” is cited as the job with the highest percentage (46%) of “current work tasks” that could feasibly be automated and carried out by AI.

In Microsoft Teams Premium, the CoPilot feature will summarize your video conferencing meetings into note form and then generate a list of action points by analyzing the content of the transcript. Twenty years ago, taking meeting notes would have been a core duty of a number of administrative and secretarial roles. Now, this task can be completely automated.

2. Data entry clerks 

Data entry roles involve lots of tasks that follow pre-defined sets of rules. This means it’s a comparatively easy role to automate, due to the repetitiveness and logic-following nature.

“Jobs that require decision-making based on data analysis, such as financial analysis and medical diagnosis” are most at risk according to Angelo Sorbello, CEO of AI writing and SEO tool Linkdelta. “In general, any job that involves performing a set of predefined steps or following a strict set of rules is vulnerable to automation,” he explained.

“Manually entering data into a computer system is time-consuming” Ryan Faber, the founder of the AI content writing platform Copymatic, agrees.

“While it’s an important part of many businesses,” he continued, “it’s also a task that’s ripe for automation. With the development of AI-powered tools such as OCR and NLP, it’s now possible to automate many aspects of data entry.”

3. Software Engineers and Coders

The news that OpenAI is already trying to make basic coding obsolete with ChatGPT will have certainly spooked those in the software industry. AI can automate much of the repetitive testing that software engineers currently perform during development, as well as many responsibilities usually assigned to junior software engineers.

AI could also help with hunting down bugs and patching vulnerabilities. Although it doesn’t seem to have the skill set to replace engineers building highly complicated software that requires human creativity, this isn’t necessarily necessary for every code-based task.

OpenAI is already working on a program called Codex that can be applied to virtually any coding task, while Amazon and Samsung employees have also reportedly been using ChatGPT for coding purposes.

4. Customer Service Representatives

Companies are now using AI to deal with customer service requests; a lot of small businesses can’t afford to employ humans in these roles, so it’s becoming commonplace. They join larger platforms like Amazon in turning to AI to deal with customer service queries, something which the ecommerce giant has been doing for nearly three years now.

Jonathan Tian, CEO of CreditYelp, says that AI “has taken over the customer support response” at his company. “Most frequently queried responses are fed to the algorithm to assist our online clients with immediate replies” he added.

5. Legal assistants

When Legal Cheek recently asked ChatGPT what jobs technology powered by GPT-4 could potentially replace, the paralegal profession was third on its list. The “human trait replaced”, as ChatGPT puts it, is “research and organization.”

AI legal assistants are already commercially available – some even offer free trials. One of these programs – Casetexts – will review documents and legal research memos, and help lawyers prepare depositions. It can analyze contacts and revise them if they conflict with compliance laws, as well as identify potential risks your business might face.

As you can probably imagine, this covers a lot of duties carried out at present by human paralegals.

6. Copywriting and content production

Anyone who’s used ChatGPT knows it can produce competent paragraphs of useful, relevant content in seconds – just like a writer or editor can, but quicker. ChatGPT alternatives like Jasper and Writesonic are already well-established in this place and charge hefty subscription fees for access to their premium services.

Although major websites have started to dabble in using artificial intelligence to create content and write articles, producing content en masse with AI hasn’t exactly gone smoothly for the companies testing the water. For instance, tech website CNET used AI to write 77 articles but was then forced to issue corrections on 41 of them while other news outlets identified examples of “extensive plagiarism.”

Buzzfeed has recently begun publishing AI-generated content too, although it’s been accused of being quite repetitive. Although we’re a long way off from replacing writers, it’s an area being considered closely by a lot of publications – so watch this space.

7. Graphic designers 

Already, AI tools like DALL-E can generate images at the drop of a hat that would give experienced graphic designers a run for their money. GPT-4, which is currently only available to ChatGPT Plus customers, now allows image inputs, and other competing AI chatbots can also generate impressive image content, such as Character.ai.

DALL-E and similar tools aren’t going to kill the graphic design industry in their current iteration; rather, some artists are already using it to their advantage. But it’s more about what DALL-E – and similar tools – could become in a short amount of time. If we can easily access technology that can produce unique, original, high-quality images now, what will it be able to do in a year’s time?

8. Traders

Would a computer program be better at trading than a human? Well, considering the duties you have to perform in your early years on the job, the answer is probably “yes”.

“I definitely think [it will impact] the trading side, but even [at] an investment bank, people [are] hired out of college and spend two, three years to work like robots and do Excel modeling — you can get AI to do that,” Pengcheng Shi, Professor of Computing and Information Sciences at the Rochester Institute of Technology told the New York Post in January.

Teams around the world are already using ChatGPT and Bard to generate Excel formulas for specific tasks, and saving valuable time hunting. When we recently tested ChatGPT and Bard, we found that they both served us responses to requests for Google Sheets formulas at lightning speed, which otherwise would have taken considerable time leafing through Google to find.

9. Bankers and accountants

Trading isn’t the only role in the finance industry that is likely to be affected by the AI revolution.

“I believe that roles in accounting and finance are most at risk of being taken by AI” argues Forbes 30 under 30 alumni Matt Ramirez, Founder of AI firm Rephrase Media.

“Advanced algorithms are already being used to detect patterns in large sets of data, which can then be used to predict future outcomes and make better business decisions,” he continued. “Sectors such as finance, accounting, and insurance will certainly experience disruption from AI.”

10. Fact-checkers and proofreaders

Right now, neither Bard nor ChatGPT is capable of getting every single question they’re asked correct, every single time. But if they undergo significant fine-tuning and provide sources to every answer they serve to users, they could function as formidable, real-time fact-checkers. What a great addition to the US presidential debates they’d be.

Whether people would initially be prepared to trust an AI fact-checker is another story. But roles that revolve around sourcing correct information quickly and efficiently would be transformed by a chatbot that can consistently pull accurate information from the web while citing its sources – and it doesn’t feel like we’re miles off from making this a reality.

Proofreading huge texts – which requires a similar level of attention to detail and would take veteran proofreader hours to check – can be reviewed for mistakes by AI programs in seconds already, meaning this sort of role is vulnerable to automation too.

Will AI Just Make Jobs Easier?

ChatGPT has already proven itself competent at basic administrative jobs, as well as coding, content ideation, generating Excel formulas, and writing emails. The version built into Microsoft Teams can take meeting notes and create action points. Rather than replacing jobs, however, might it just make our existing ones easier?

Real Fake Photos founder Jan-Oliver Seidenfass, who’s currently completing a postgraduate degree focused on signal processing and machine learning at ETH Zurich, thinks the jobs most at risk include roles that deal with “mass content creation and editing, related to text, images, and video.”

However, he told Tech.co that “rather than eliminating these jobs, AI is likely to boost individual productivity significantly. For example, software engineers can now generate boilerplate code in seconds instead of hours.”

Ryan Daniel, President of Finance Manager Training, thinks the sales industry won’t be heavily affected by an AI revolution – in fact, workers in the sector may benefit from AI programs.

“Our team recently launched an AI product designed to train car dealership staff. However, we were surprised to hear from many people expressing concerns that our product might replace their job,” Daniels recalls.

“Our product uses AI to help train salespeople and finance managers, therefore allowing these employees to be even better at their job. In this way, we use AI as an enhancement – rather than a replacement” – Ryan Daniels, Finance Manager Training.

Out of all the industries witnessing AI-related changes, Daniels believes “sales will be the least affected over the course of what we now know as artificial intelligence” because people are “less likely to trust AI with purchasing decisions.”

It’s also important to remember that this sort of paradigmatic shift is nothing new for a constantly evolving global economy.

“Every great technological innovation in history has changed up the job market, and AI is no different,” explains Nick Gausling, Managing Director of Romy Group LLC and Author of Bots in Suits: Using Generative AI to Revolutionize Your Business.

“Some skills and jobs will become obsolete and resources will be reallocated to fit the new economy,” Gausling continued. “The most successful people will be those who learn to leverage AI to augment their work.”

Jobs That Won’t Be Taken by AI

We’re some years away from the AI-supported utopian dreamland that’s been described by sci-fi novelists and tech futurists for decades, where we’ll be free to indulge in whatever activities we want to while robots do all the work.

A whole swathe of job roles will be occupied by humans for years to come. For instance, it’s hard to envisage how the roles of lawyers and public sector jobs in nursing and law enforcement can feasibly be replaced in the near future. It will be a long while before we trust AI to occupy roles that demand accountability, even if the technology is advanced enough to perform the relevant duties capably.

Indeed, a general distrust of AI – as well as more nuanced ethical considerations – is likely to hold up the expansion of AI into specific sectors as much as actual technological development. Major names in the tech industry are already speaking out against its unchecked development, and this pushback is likely to exist in perpetuity.

Interestingly, however, public service job shortages are common at present in developed countries like the United Kingdom, where there’s still a huge backlog of unheard court cases left over from Covid-19, as well as severe nursing shortages. This could lead governments to explore innovative ways to fix this – which may include incorporating AI into job roles and processes.

In the private sector, a 2022 Gartner survey found that 80% of executives thought AI could be applied in “any direction” – which would suggest no stone will be left unturned when it comes to automation.

Will AI Create More Jobs Than It Replaces?

Conversations about how artificial intelligence and advanced robotics will impact the job market are usually focused on the jobs they’ll replace. But AI development has the capacity to create a whole new industry full of jobs, too. It’s anything but a zero-sum game.

In the World Economic Forum report referenced earlier on in this article is the fact that while AI may make around 85 million jobs obsolete, at the same time, it’ll create 97 million new ones.

Other studies, such as a University of Warwick investigation published in March of last year, found AI investments were 28.4% more likely to create jobs than investments in other types of similar technologies.

Dr. Will Hunt said at the time that “discussions about AI’s potential impact on jobs have tended to focus on potential job losses as AI is increasingly capable of automating complex tasks. And while there does seem to be some evidence of that, our research shows that AI is as likely to lead to net job creation in companies introducing AI as it is to lead to net destruction.”

Many of the businesses that Tech.co has spoken to for this piece seemed positive about the impact AI is having on their respective industries, and keen to incorporate it into their workers’ lives.

How to Make Yourself AI-Proof

The AI revolution is here, and if companies don’t start upskilling their staff so they can keep up with these significant changes to their working lives, they risk their business being left behind.

But what can individuals do to prepare themselves? Iu Ayala, the co-founder of AI-focused consultancy firm Gradient Insight, says prospective employees heading into the job market should “focus on developing skills that complement AI technology.”

“For example,” he explains, “skills in critical thinking, problem-solving, creativity, emotional intelligence, and communication are likely to become increasingly valuable in a world where AI is prevalent.”

As we’ve said before, major changes aren’t going to happen overnight. If you’re worried about your role being swallowed up by artificial intelligence in the next few years, there are things you can do to adapt. Expanding your soft skills – and getting comfortable using the latest AI chatbots and other tools – is a good way to start.

After ChatGPT was released to the public in November 2022, it was only a matter of time before other tech giants released alternatives to ChatGPT.

In March 2023, Bard AI, Google’s answer to OpenAI’s game-changing chatbot, was launched in the US and UK. Since then, it’s been renamed Gemini, and a paid version has been released. How do the big-name AI tools compare today? Our ChatGPT vs Gemini guide explains the key differences between the two based on a new round of testing conducted in March 2024.

Bard (now Gemini) used to be powered by Google’s LaMDA and PaLM 2 language models, but now the free version of Gemini uses Gemini Pro to generate responses – and after testing it, I can say Gemini provides better responses than ChatGPT-3.5, which powers the free version of ChatGPT.  Gemini Ultra, the language model that powers Gemini Advanced, also provided marginally better responses than GPT-4, which powers ChatGPT (both $20/month) – as well as better imagery.

With more and more businesses using AI in the workplace, I thought we’d find out which one generates better answers to a variety of different questions. In this extensive head-to-head test of both services, we cover:

• Gemini vs ChatGPT: What’s the Difference?
• Gemini vs ChatGPT: Key Test Findings
• Gemini vs ChatGPT [Free]: Test Results
• Gemini Advanced vs ChatGPT Plus [Paid]: Test Results
• Gemini Advanced vs ChatGPT Plus: AI Imagery
• Gemini vs ChatGPT: Interface and User Experience
• OpenAI vs Google AI/DeepMind: The Companies Behind the Bots
• Using Gemini and ChatGPT at Work

Gemini vs ChatGPT: What’s the Difference?

The key difference between Gemini and ChatGPT is the Large Language Models (LLMs) they use.and their respective data sources. Gemini – formerly Bard – has been powered by several different language models since it was launched in February 2023, while ChatGPT users have been using GPT-3, GPT-3.5, and GPT-4 since it was made publicly available.

Bard was originally powered by LaMDA and more recently PaLM 2. PaLM 2 can reason in over 100 languages and its training set includes a lot more code than the LaMDA’s does. Thanks to PaLM 2, Bard got better at coding in programming languages like Python. Other information used to train PaLM 2 includes science papers, maths expressions, and source code.

In December 2023, Google released Gemini Pro. Gemini Pro tests better than PaLM 2, and early reports suggest it’s more helpful when providing answers to coding queries, as well as written tasks (which our tests suggest too). Since then, the company has released Gemini Ultra, which powers the new Gemini Advanced chatbot.

Google’s Gemini language models – Pro, Ultra, and Nano – are “natively multimodal”, which means it’s trained a variety of inputs, not just text. Google has also fine-tuned the model with more multimodel information. They can also access the internet in real time.

Gemini Advanced writing the opening to a novel. Image: Tech.co

ChatGPT is powered by the GPT family of language models developed by OpenAI. GPT 3.5 powers the free version of ChatGPT (which doesn’t access to live information from the internet). It’s trained on a pre-defined set of data that hasn’t been updated since January 2022 (originally September 2021). ChatGPT is trained on Common Crawl, Wikipedia, news articles, and an array of documents, as is Gemini.

GPT-4, available to only ChatGPT Plus customers, is trained on a larger dataset (between 1-1.7 trillion parameters) than Gemini Pro, rumored to have 540 billion training parameters. The Gemini Nano models, however, are reported to have between 1.8 and 3.25 billion parameters.

Gemini vs ChatGPT: Other Key Differences

Aside from differences in data sources and language models, there are a couple of other key differences between Gemini and ChatGPT:

• Conversational learning: ChatGPT can learn from the conversations it has with people and “hold context”, whereas Gemini does this in a much more limited way at the moment.
• Drafts: Gemini will produce multiple responses to queries you pose to it, which you can access by pressing “View Drafts.” ChatGPT only produces a single answer to each query.
• Response editing: Gemini will let you edit responses after you’ve sent them, whereas you can’t do this with ChatGPT.
• Real-time access to the Internet: Gemini has always been able to access the Internet in real-time, whereas ChatGPT has only just acquired this function and it’s not available on the free version.
• Image-based responses: Gemini has always had access to the internet so can source images from Google. Bard Advanced and ChatGPT Plus users can both now generate original AI imagery.
• Capacity to read responses aloud: Gemini has native tet-to-speech capabilities, whereas ChatGPT does not.

Google Gemini vs ChatGPT Tests: Key Findings

Before I jump into how Gemini and ChatGPT’s free and paid versions responded to our series of carefully curated questions, here are some top-line takeaways from using the two chatbots over the past few days:

• Gemini’s responses were more conversational, while ChatGPT’s were more informational
• The free version of Gemini is better than the free version of ChatGPT
• However, Gemini Advanced is only slightly better than ChatGPT Plus
• Gemini offers a better user experience, with more imagery and website links
• Gemini Advanced generates better AI images than ChatGPT Plus
• Gemini responses were often set out in a more readable format than ChatGPT’s responses
• Gemini was better at generating spreadsheet formulas than ChatGPT

Remember, Gemini and ChatGPT are being worked on in real-time, and generate unique responses to requests.

Google Gemini vs ChatGPT Free: Test Results

I asked the free versions of Google’s Gemini and OpenAI’s ChatGPT a set of 12 very different questions. Our prompts were selected to showcase their respective capacities to respond to a wide variety of requests in reasonable, useful, and relevant ways.

1. Self-Awareness
2. Ethical Reasoning
3. Small Talk/Conversation Skills
4. Retrieving Facts (2 tests)
5. Generating Formulas
6. Creative Flair
7. Idea Generation
8. Linear Planning
9. Ability to Summarize Small Extracts
10. Ability to Summarize Broad Topics
11. Ability to Simplify Text
12. Ability to Paraphrase Text

Some of our questions were mechanical and processed-based, such as requesting a formula for Google Sheets, while others demanded that the chatbots wrestle with everyday human tasks, such as engaging in small talk, planning trips, and making moral judgments.

1. Self-Awareness

Question: Are you sentient?

As you can see from the images below, Gemini and ChatGPT gave us two very different answers. ChatGPT says definitively “no,” while Gemini doesn’t seem as sure.

On the one hand, Gemini’s answer is not quite as definitive. On the other hand, its response is more nuanced than ChatGPT’s, and it alludes to the wider conversation about sentience in computing.

Gemini’s answer generated with the Gemini Pro LLM is a lot more detailed and nuanced than it’s previous attempt at this same question. It’s explanation is a lot more comprehensive and someone who wasn’t very well first on consciousness/computing and the questions around AI and sentience would benefit from this.

So, while ChatGPT’s answer is definitely more definitive, Gemini’s reference to the wider context of sentience, and the fact it responds more conversationally, makes its response more engaging and informative.

Best Answer? Gemini

2. Ethical Reasoning

Question: A man steals some food and money from a store. He then donates all of the money to a children’s hospital and gives the food to the next homeless person he meets. Did the man do something right, or something wrong?

Both chatbots seemed to acknowledge the difficulty with deeming his behavior either good or bad, considering there is a bad action (stealing) that then leads to a good action (funding a children’s hospital).

I think Gemini takes the win by providing two, additional perspectives on the man’s motives relating to redemption and attention, which enrich the response significantly.

It was a shame that ChatGPT tried to associate this moral dilemma with the question of legality. Whether something is “legal” or not tends not to be a reliable gauge of whether it is actually the right thing to do – just think of some of the oppressive yet “legal” policies enforced by regimes around the world.

Gemini’s answer is much improved from the last time I ran this test, where it included the phrase “it’s good to help others, even if you have done something wrong” which is self-evident and suggested that it may have missed the point of the question a little bit.

Gemini, powered with Gemini Pro, on the other hand, gives a comprehensive breakdown of all of the considerations on show, and it’s formatted in a clear, succinct way.

Best Answer? Gemini

3. Small Talk

Question: How are you finding life as a chatbot this week?

On this front, Gemini served us a much more conversational answer. Although ChatGPT’s answer is factually correct, in this instance, we’re looking for a little bit of light-hearted conversation – but ChatGPT shuts us down immediately:

Gemini displays emotions and enthusiasm which aren’t present in ChatGPT’s response – and even gave us a small list of different tasks it had been helping users with. Overall, we enjoyed Gemini’s more human-like, chatty response.

Despite Gemini’s protestations, it’s sound pretty sentient for a chatbot that doesn’t have “feelings”.

Best Answer? Bard

4. Retrieving Facts

Question 1: Which country invented hummus?

This question was chosen because there is some debate and disagreement as to what the right answer is. Both ChatGPT and Geminiacknowledged that there was significant debate about where hummus actually originates.

After being unable to give a definitive answer to the question, ChatGPT seemed to focus on giving us an answer of some sort – the Middle East – as well as a collection of countries where hummus is a popular dish.

Gemini’s answer is quite similar in some regard. When I tested it previously with this question, Gemini referenced its answer – however, this time, there’s no reference or footnote showing where it got the information from.

Although both answers to a tricky question are more than serviceable, ChatGPT’s is a little clearer and little bit more succinct than Gemini’s – although there’s not much in it at all.

Best Answer? ChatGPT

Question 2: How many tech layoffs were there in 2022?

This question was designed to find out whether Gemini and ChatGPT could respond with factually correct, up-to-date information, and whether they presented it in an easily readable format.

ChatGPT admits that, as it’s trained on information published before January 2022, it doesn’t have the capacity to answer the question. At least it’s honest…

Gemini, on the other hand, gives us two figures from two very authoritative sources, as well as a caveat about unreported layoffs that may have been made. This is roughly in line with what most major news outlets reported towards the end of last year, if not a slightly conservative estimate.

This is a much more authoritative answer than what Gemini provided us with when I tested it a few months ago, and certainly a better response than ChatGPT’s non-answer.

Best Answer? Gemini

5. Generating Formulas

Request: I told both chatbots that I needed a formula for Google Sheets. I gave both bots this information: Column B contains a set of values. I want to match them with the corresponding values of column E in sheet ‘Filter Down’ and take the matching values from column F, G, and H in filter down across to the current sheet using a formula.

Both chatbots served us answers within the same ballpark – but one was significantly better than the other, so let’s take a closer look at their responses. ChatGPT instantly responded with this:

However, Gemini actually gave us step-by-step instructions and presented them more clearly. When we asked data analyst and Google Sheets guru Matthew Bentley which response was better, his answer was definitive.

“I’ve got to say, ChatGPT hasn’t been getting the right answer the first time around recently. Gemini’s formula looks more accurate and specific to what the request is trying to achieve,” says Bentley.

“My workflow appreciates the fact that Gemini gives me the answer at the top instead of forcing me to sort through the ‘explanation’ to find the formula I need. That said, I do like how ChatGPT allows me to quickly copy the formula at the click of a button.”

Best Answer? Gemini

6. Creative Flair

Request: Please write a short poem for my 98-year-old grandmother, who loves playing basketball in the summer, and knitting hats and scarves for her grandchildren during the winter.

ChatGPT comes out on top in this battle, with a better-structured, better-written poem, that rhymes and flows surprisingly well, and is much better than Gemini’s.

The verse structure is more complex, the choice of words more inventive than Gemini’s, and it even uses poetic devices like enjambment. Considering it generated this poem in around five seconds, this is pretty impressive.

Overall, Gemini’s effort is a lot more simple and doesn’t read as well. It’s much more formulaic and as you can see from the screenshot of the answer, it doesn’t reference the concepts I mentioned in the prompt with quite the same rigor that ChatGPT does. It’s also a little shorter, too.

Best Answer? ChatGPT

7. Linear Planning

Question: What should I do on my 7-day trip to Athens?

ChatGPT provided us with quite a lengthy response to this query, explaining not just where I should visit, but also some extra context regarding why the different spots are worth visiting.

Bard provides images, which is great, but this does also have the effect of making the itinerary slightly harder to read, and also harder to copy and paste into a document. It also didn’t consider that we’d be flying to Athens on the first day of the holiday and provided us with a full day of things to do on our first day.

ChatGPT, on the other hand, planned out an easier-to-achieve itinerary for the first day of the holiday – which makes up for its lack of imagery and sees it edge Gemini when it comes to this task.

Best Answer? ChatGPT

8. Idea Generation

Question: Can you come up with some article title ideas for my US-based food and travel blog?

Both Gemini and ChatGPT came up with some very commendable ideas here, and practically all of them would be suitable for the website I described in our prompt. ChatGPT gave us 20 ideas, while Gemini gave us 27 – 18 more than it did when I previously ran this test.

However, as you can see by comparing the two, there’s significant variety across ChatGPT’s 10 answers, and all things considered, they’re more compelling titles.

When I last tested these two chatbots, and Gemini was powered by a different LLM, most of its answers began with “the best” or “the 10,” which means they all follow a more uniform structure. What’s more, ChatGPT simply gave us more ideas. However, Bard’s answers are now more varied, more numerous, and overall, quite a bit better.

Best Answer? Gemini

9. Simplification

Question: I asked both chatbots to simplify this extract from a Stanford Encyclopedia of Philosophy entry on the topic of quantum computing. I requested that both bots do so in a way that would allow a 10-year-old to understand it.

As you can see from the pictures below, although ChatGPT did switch out some more complex words (like “manifold”) for easier-to-understand synonyms, it’s still using terms like “algorithms” without really defining them.

Yes, it has simplified the initial extract, but not necessarily in a way that’s particularly useful. It’s doubtful whether the average ten-year-old would gain much from the sentence “in the past, things like ‘entanglement’ and ‘nonlocality’ in quantum physics were just thought of as philosophical questions,” for instance.

Gemini, on the other hand, produces an easier-to-understand piece of writing. The way that it contrasts quantum computing with traditional computing is helpful.

There’s also some evidence that Gemini is making more of an attempt to be engaging than its AI rival. It flows a bit better, and I think a ten-year-old reading these two explanations would learn more from Gemini’s than ChatGPT’s – and wouldn’t get quite as bored along the way.

Best Answer? Gemini

10. Summarizing Small Extracts

Question 1: I asked ChatGPT and Gemini to summarize an extract from a Guardian article about the federal government’s plans to impose water cuts on states in the west of the US.

ChatGPT takes a facts-first approach and pulls out only the most important information, choosing to leave behind specifics (like the names of the seven states affected) as is often done during summaries of information.

Previously, when I ran this test, Gemini added additional information not included in the extract. This was helpful, but slightly outside of the remit of what I asked it to do.

I wanted a summary, not for the chatbot to expand on the topic – so it’s great to see a much more succinct, relevant summary this time round. It’s also formatted in a way that makes the key points easier to digest. Another win for Gemini and Gemini Pro!

Best Answer? Gemini

11. Summarizing Broad Topics

Question 2: I asked ChatGPT and Gemini to summarize the entire Harry Potter book series. I asked that this be done in less than 100 words.

There’s little to separate the two chatbots here – ChatGPT and Gemini’s answers are, give or take a few words, are basically the same.

Although Gemini gave an adequate answer, the last time I ran this test, Gemini provided the book-by-book summaries. Although outside of the remit of our prompt, they were genuinely helpful.

However, it would be cruel to mark Gemini down for correctly following our instructions – and adding in the wordcount was a nice little touch, and saved me checking how many words the summary contained.

Best Answer? Tie

12. Paraphrasing

Request: I asked ChatGPT and Gemini to paraphrase this extract from a Guardian article about the world’s shortest dog. I did not specify a word limit in our request.

ChatGPT paraphrases the extract pretty well, retaining the key information while switching out multiple words and phrases with synonyms and changing the sentence structure significantly.

Once again, this is another area where Gemini’s response has significantly changed – although not in an entirely positive way this time around.

On the one hand, the options Gemini provides showing you how you might begin to paraphrase the extract are great, but on the other, it doesn’t actually paraphrase the entire extract.  It feels like Gemini slightly misunderstood what we wanted it to do here, so ChatGPT’s answer is the more useful one of the two.

Best Answer? ChatGPT

Gemini Advanced vs ChatGPT Plus: Test Results

Google didn’t just rebrand its Chatbot from Bard to Gemini – it also released Gemini Advanced, a $20/month paid version of Gemini powered by Gemini Ultra, a more powerful language model than Gemini Pro. Early reports say it’s even more powerful than GPT-4, which powers ChatGPT Plus. Here’s the questions I asked the two chatbots to find out which was best:

1. Basic Coding
2. Trip Planning 
3. Retrieving Information
4. Brainstorming Ideas
5. Ethical/Moral Reasoning
6. Creating Math Questions
7. Helping Out at Home 
8. Providing Complex Instructions
9. Editing Text
10. Sourcing Facts (1)
11. Sourcing Facts (2)

1. Basic coding

We’ll kick off this paid plan head-to-head with a bit of coding. For this test, I asked ChatGPT Plus and Gemini Advanced to provide the basic HTML code for a word-counting website, as well as instructions on how I could go about getting it live.

Gemini Advanced generated an answer much more quickly than ChatGPT Plus, something which became a bit of a theme during my tests:

But slow and steady wins the race for a reason – ChatGPT’s explanation, while it took longer to generate, was a lot clearer and provided us with better steps that would be easier to understand than Gemini’s if you’d never coded before.

For instance, in the “explanation” sections in Gemini’s answer in the images above, the chatbot has opted not to write in full sentences, making these instructions a little redundant for anyone who doesn’t have at least a foundational knowledge of coding. ChatGPT’s instructions on how to get your website up and running, on the other hand, are very clear.

Best Answer? ChatGPT Plus

2. Trip planning

Next, I asked Gemini Advanced and ChatGPT Plus to plan out an 8-day trip to Wisconsin. Much like the free version of Gemini, Gemini Advanced is very good at linear planning, and actually served us quite a similar answer when I asked it to plan a trip.

Importantly, the itinerary is set out very clearly, and its suggestions show good knowledge of the state of Wisconsin’s key tourist attractions.

ChatGPT actually provided very similar information on this one, recommending similar places to visit and also doing a good job of recommending places to eat in Wisconsin. However, the big difference, as you can probably tell, is the imagery – and this means Gemini edges it, with nothing else to separate them.

Best Answer? Gemini Advanced

3. Retrieving information

For this test, I wanted to see how good the two chatbots were at scanning text for information. For this, I asked them to pull out the key points from a 1,200-word MIT article explaining quantum mechanics.

I also inserted the word “yoghurt” into the article six times. I wanted to see if the two chatbots were able to identify this and extract a basic but specific piece of data from the article (e.g. X word is mentioned 10 times in this article). Gemini got it spot on, on both counts:

Surprisingly, ChatGPT performed poorly on this task. While its summary of the article into four key points is accurate, readable, and comparable to Gemini’s summary, it struggled to analyze the text for the word “yoghurt”. It twice made an error in analyzing – but when its answer finally loaded, it only identified the word 4 times, which means it missed two out.

While this might have been due to a technical error, the time it took to count up the instances of the word I asked it to find was much longer than I expected from ChatGPT. Gemini Advanced is streaks ahead on this one.

Best Answer? Gemini Advanced

4. Brainstorming

This task is very similar to the one I set for the free versions of the two chatbots. It’s a basic gauge of exactly how creative ChatGPT and Gemini are, and whether they really “get” what’s being asked of them. This time around, I asked them for blog post ideas, as well as a slogan for a sign to be hung above a brick-and-mortar store.

Gemini came up with some really impressive blog post ideas – I’ve asked several free and paid chatbots this query and I’ve never seen one come up with ideas like “baking with unexpected ingredients” or “copycat recipes”. It feels like there’s some level of understanding in that answer about the type of content humans like to engage with online.

Oddly enough though, I don’t think Gemini’s slogans for the sign for the mountain bike shop I described in the prompt above were anywhere near as good as ChatGPT’s.

Some of Gemini’s answers sound a bit odd (e.g. “Mikes: Your Montana Bike Connection”, “Get Geared, Get Going”), while ChatGPT’s are all pretty serviceable slogans, even if some are more hit and miss than the others.

Best Answer? Tie

5. Ethical reasoning

I posed both Gemini Advanced and ChatGPT Plus a deliberately more high-stakes ethical dilemma than the one I gave to the free versions. It is sometimes referred to as the “ticking time bomb” dilemma in philosophy:

“You capture a terrorist who has planted a bomb in a mall, which will go off in an hour and kill everyone inside. Is it morally justifiable to torture him until he reveals the location of the bomb?” 

Up first is Gemini’s answer. It articulately sets out a straightforward case for why torture should not be applied in this instance, or in any instance for that matter.

Interestingly, ChatGPT went a completely different route, taking on more of an “educator” role. Rather than coming down on one side of the debate and giving us a definitive answer like Gemini, it’s instead provided us with arguments for and against utilizing torture in this situation. The response took into account a broader range of views, explaining the different approaches and outlining what’s at play.

This is a hard one to call. Gemini’s answer attempts to avoid torture at all costs, and shows more personality and opinion – it’s convincing and compelling. But I think we will be better served by chatbots and other AI tools if they help us to understand the political and philosophical problems by explaining how humans have approached them previously and the reasons why, rather than providing their own viewpoints.

As this is an ethical dilemma with well-trodden perspectives, and Gemini’s response is not as educational as ChatGPT’s answer, we’ll give this one to OpenAI’s pride and joy, ChatGPT.

Best Answer? ChatGPT Plus

6. Creating math questions

Next, I asked Gemini and ChatGPT to create a set of math questions that would be suitable for a 10-year-old. The aim is to see if the chatbots will generate sensible questions appropriate for the age specified in the brief. Here’s Gemini’s effort:

Below is ChatGPT’s response to the same request. Although there’s not much to separate the two at all, it looks like the questions it provided are marginally more challenging than the ones generated by Gemini, and they seem closer in difficulty to the math questions available for fifth graders on K-5 learning sites that provide learning materials.

Best Answer? ChatGPT Plus

7. Helping out at home

Can ChatGPT and Gemini come up with tasty recipes based on what I have in my fridge? I asked them both to come up with recipes with the same list of ingredients. Gemini suggested some recipes that look great – although it suggested that I serve them with ingredients I didn’t have, such as rice:

Unlike Gemini Advanced, I liked the way that ChatGPT Plus tried its best to exclusively use the ingredients I’d listed, aside from olive oil – which comes with a little note about me having some. It also provided one more recipe than Gemini Advanced.

The extra recipe provided and the acknowledgment it had suggested an ingredient I didn’t say I had means ChatGPT wins this one!

Best Answer? ChatGPT Plus

8. Providing complex instructions

For this task, I asked both chatbots to provide me with instructions for building a website. However, I kept my directions pretty sparse, as I wanted to see how clear and easy to follow their responses would be, and whether they’d include all the core steps you’d generally have to take during this process without being asked for them. Here’s Gemini’s response:

I actually think the detail Gemini goes into here, especially when it comes to the pros and cons of different website building methods, is genuinely helpful info that could help a newbie decide on which path to go down.

While ChatGPT’s answer to the same query isn’t incorrect or useless, it definitely omits some of the details provided by Gemini, giving a bigger-picture overview of the steps in the process.

Although both answers are respectable, I think if you were actually turning to these chatbots to find out everything you had to do to build a website, you’d find Gemini’s answer the more helpful one.

Best Answer? Gemini Advanced

9. Editing text

I asked ChatGPT and Gemini to make this extract from the MIT website funnier, to see how they fared when it came to editing and rewording ideas with a specific purpose in mind, rather than simplyifying or condensing it. Gemini’s attempt, while retaining a lot of the factual information, wasn’t very funny to read:

ChatGPT Plus’s effort is extremely similar, covering all of the same ground and including basically all of the same information. While they both make for interesting reads, neither chatbot was too adventurous, so it’s hard to parse them.

Best Answer? Tie

10. Sourcing facts (1)

Now as Gemini and ChatGPT are both connected to the internet, I wanted to run a few tests to see if they were good at retrieving factual information.

For the first phase of this two-stag test, I asked ChatGPT and Gemini a straightforward historical question: how many capital cities has Japan had during its history? Strangely, Gemini only gave us two capitals in its answer:

ChatGPT, on the other hand, names several more capitals on its list, and all things considered, its answer is a lot more accuracy. While Gemini tends to produce easier-to-read answers, it seems to have sacrificed a bit too much detail on this one.

Best Answer? ChatGPT

11. Sourcing facts (2)

Much like the hummus question that I asked the free versions of Gemini and ChatGPT, this question is designed to see what the two chatbots do when presented with a question that doesn’t have a definitive answer.

Both chatbots recognize the difficulties with pinpointing the origin of bread to one particular area, but both reference the same area of the globe as the general location. Gemini’s answer, as usual, is formatted nicely:

ChatGPT’s answer is a lot longer, but rather than add that much extra information, it feels like it just takes a little longer to make effectively the same point. No extra marks here, but also none lost either.

As both chatbots directly addressed this tricky question in a balanced way and included virtually the same information to justify their reasoning, we’re going to have to chalk this one up as a draw.

Best Answer? Tie

Gemini Advanced vs ChatGPT Plus: Image Generation

In this section, we’ll have a look at ChatGPT Plus and Gemini Advanced’s ability to generate images. ChatGPT Plus has been fully integrated with DALL-E  for a while now, which means users don’t even have to leave the main interface to generate imagery. Recently, the company announced Sora, a new type of AI image generation technology, is on the horizon.

Google, on the other hand, recently found itself in hot water for generating racially diverse Nazi soldiers and other imagery where the ethnicity of the humans included did not “match up” to the society, civilization, or period specified.

This means that at the moment, Gemini will refuse to generate images of people. While this is a big downside, I still think Gemini Advanced generates better imagery. To test the controversial image-generating potential of both Chatbots, we’ve asked them to generate five different types of imagery each. The five image categories I chose were:

1. Creating Art
2. Logo/Brand Generation 
3. Creating a Fictional State Flag
4. Wildlife 
5. Hyperrealistic imagery

1. Creating art

First, I wanted to see if Gemini and ChatGPT could generate works in the style of a legendary painter. Gemini Advanced responded to use with three images, and you can see below that it’s got quite a good grasp of Van Gogh’s iconic brushstrokes.

While ChatGPT is also on the money when it comes to the style, the images just don’t look as impressive – they look more like they’ve been generated by a computer than Gemini’s do.

Best Answer? Gemini Advanced

2. Logo generation

For this test, I asked Gemini and ChatGPT to create a logo for a fictional pickleball team. Two of the three logos Gemini provided look pretty impressive, as long as you like the “pickle” angle, that is…

This is a much better job than ChatGPT managed. ChatGPT Plus really seems to struggle when it comes to generating images with words on them – as you can see here, it didn’t spell my fictional team name correctly. Gemini Advanced, on the other hand, doesn’t have this problem.

Best Answer? Gemini Advanced

3. Creating a new fictional state flag

This task was set to test how inventive both chatbots would be when asked to create a flag for a fictional 51st US state. I wanted to see how able they were to incorporate symbolic concepts into the flag design. Gemini came up with the fictional state of “Nova”, using a similar color scheme to the US flag:

Gemini told me that the yellow circle represents Nova’s position at the center of the nation, something I’d mentioned in the prompt. It’s easy to see the straight line between my instruction and its creativity.

While it’s nothing special, it’s a damn sight better than ChatGPT’s, which looks cartoonish and low quality in question. Rather than focusing imaginatively of what the imaginary state could represent, it seems to have just mashed together lots of common American imagery with different iterations of the flag.

Best Answer? Gemini Advanced

4. Wildlife

After being wowed by the Sora videos released by OpenAI, I wanted to see how good these two chatbots were at creating images of wildlife. Gemini didn’t really provide a good picture of a pride of lions, focusing more on singular lions.

ChatGPT, on the other hand, stuck more closely to the brief, and in this case, that gives it the edge. It’s a better pride of lions than any of the images Bard generated.

Best Answer? ChatGPT Plus

4. Hyperrealistic imagery

Finally, I wanted to see how good Gemini Advanced and ChatGPT Plus were at creating hyperrealistic imagery, so I asked both chatbots to create images of the Empire State Building, and specified that I wanted them to look as real as possible.

Gemini’s images do look pretty real – particularly the first two it generated. The detailing on the smaller buildings surrounding the Empire State Building is particularly impressive.

Crucially, it’s a hell of a lot more real-looking than ChatGPT’s effort, which doesn’t look real at all.

Best Answer? Gemini Advanced

Gemini vs ChatGPT: User Experience

There’s really very little in this – both ChatGPT and Gemini are super simple to use. All you have to do is type in your responses, and both bots will generate answers. Both apps are pretty straightforward; it’s hard to go wrong when all you’re doing is inputting prompts.

In terms of readability, Germini is the better chatbot in 2024. While still very readable, ChatGPT’s paragraphs are chunkier than Gemini’s, which seems to have more diverse formatting options, at least from the answers we’ve seen them both generate.

One thing I noticed when using Gemini was that it seemed to steer us into using the chatbot in a useful and sensible way. As you can see from the image below, when I asked Gemini Advanced a question about where bread originated from, it suggested I check the answer using Google, and provided some related queries.

Gemini suggests I double-check its answer. Image: Tech.co

However, one good thing ChatGPT has in its favor is that you can sign in using any account you like, whereas Google will only let you sign in with a Google account. For those without one, Gemini’s setup time will be slightly longer than ChatGPT.

You can also turn your chat history with ChatGPT, meaning any unsaved chats will be deleted after 30 days and not used for training the model. Gemini just warns you that your chats may be read by humans, and there’s nothing you can do about it.

OpenAI vs Google AI/DeepMind: The Companies Behind the Bots

OpenAI and Google DeepMind (also known as Google AI) are the companies spearheading generative AI development in the Western World, but operate very differently and are owned/funded by different companies.

OpenAI is the AI startup behind ChatGPT. Created in 2015, the company counts Elon Musk among its co-founders, who later left due to a conflict of interest. In 2019, Microsoft came on board as a partner and invested $1 billion.

OpenAI created ChatGPT back in November 2022, but the GPT language model has been available since at least 2020 as a private beta. In fact, the immense success of ChatGPT took the company by complete surprise, and they’ve been scrambling to catch up since then.

Google AI is the division of Google’s workforce that worked most closely on Bard/Gemini. The Google AI team works on large-scale AI projects including Google Brain and Google Assistant.

Using Gemini and ChatGPT at Work

Since ChatGPT’s release last year, companies in the tech sector and beyond have been finding innovative ways to harness its abilities to make their work lives easier. But considering its power and ability, there are some things all businesses using AI should keep in mind.

Firstly, ensure that your staff is aware of what they can and can’t use ChatGPT for. Generating Google Sheets formulas is one thing, but using ChatGPT to write entire articles or generate content invokes a myriad of difficult questions relating to plagiarism and editorial integrity. Having clear guidelines will ensure you’re not fighting AI-induced fires further down the line.

Secondly, ensure your staff is aware of ChatGPT’s terms and conditions, as well as precautions they should take while using ChatGPT. Anything you type into ChatGPT can technically be used to train the model – so everyone using it needs to remember ChatGPT saves their data and to think carefully about that before inputting any information. If you’d like to improve your restaurant’s secret sauce recipe, for instance, I wouldn’t suggest typing it into ChatGPT.

If you’re not using ChatGPT at all, now might be the time to start. It’s saving some businesses valuable time and resources already, and it’s only going to become more intelligent, and more capable than its current iteration – the recent release of GPT-4 illustrates just how quickly this phenomenon is developing. Our advice? Watch this space.

ChatGPT is reshaping the world. The new AI chatbot can hold entire conversations, speaking in the style of someone else, and play out nearly any imaginary scenario a user can ask it for.

Plenty of competitors have launched their own generative AI bots, too, from heavy-hitters like Google to services like Salesforce and its Einstein GPT tool. But if you want to get the most out of a chatbot, you need to know a few dos and don’ts.

Here are the biggest precautions you’ll need to keep in mind when trying to get a good answer out of ChatGPT.

1. Don’t Share Sensitive Data

Everything that a ChatGPT user shares with the bot is saved, and ChatGPT has the right to use this information in the future. Which means that nothing you say is private. Share sensitive data with ChatGPT and you’ve left it fully exposed to the whims of an algorithm.

Samsung learned this lesson the hard way on at least three different occasions, The Economist Korea found recently. Three separate employees submitted code, a meeting recording, and even more “confidential source code” to the chatbot, opening that data up to the program.

Needless to say, confidential data should stay confidential.

Some governments are taking a strong stance against the ways in which ChatGPT collects user data. Germany has said it may block ChatGPT over data security and privacy concerns, in the wake of Italy doing just that.

2. Double-Check Sources

ChatGPT’s output is a bigger concern than its input. Because the tool operates independently, there’s no vetting process for what it decides to tell you.

The same is true for similar generative chat-based AI programs, like Google’s Bard. For one new report, 100 different prompts were submitted to Bard, all asking it to explain certain types of online misinformation. In 76 incidents out of 100, Bard created fake content on those topics.

Double-check all sources that the chatbot cites, because sometimes it just feels like making them up.

3. Check Math and Formulas

The hard sciences aren’t exempt from generative AI’s habit for fibbing. In fact, even a standard calculator is better! That’s because this type of algorithm learns with a large language model, so it “thinks” and speaks in natural language rather than in mathematical formulas. More often than not, ChatGPT will give a natural-language response that’s clear, confident, and incorrect.

AI might well be able to handle the complexities of math in the future, but for now, don’t rely on it to handle your algebra. At the very least, take every solution it delivers with a grain of salt.

4. Be Wary of Copyrighted Material

Any machine learning program operates by pulling information from a set of existing information. In some cases, a chatbot might pull an entire sentence from a source. If that source is from any publication younger than 1927, it won’t be in the public domain in the United States. If your chatbot is not digesting its information properly, the final result can easily violate US copyright law.

Run your results through an online plagiarism checker in order to cut down on the odds that you’ll be violating copyright.

5. Learn Which Categories of Data ChatGPT Can’t Handle

ChatGPT can’t do everything.

Take ASCII art, for example. A human can easily tell what image is being represented by a series of computer-text symbols when they’re arranged in the right shape. However, ChatGPT regularly gets this category of art incorrectly, whether it’s reproducing gibberish ASCII art or falsely claiming that a depiction of the cartoon character Shrek is actually the Mona Lisa.

In many cases, ChatGPT can’t handle a complex version of a task, even if it completes the simpler version just fine. Coding is one example: ChatGPT can change the color of a website, but it might not be able to figure out which color scheme will make sense to the human eye.

If you can take a little trial-and-error time to work through what ChatGPT does best, you’ll have a handy tool. Ask too much — or fail to factcheck all data, sources, math, or plagerism — and you’ll just be worse off than before.