AI – ChatGPT – The next big thing?

The following employs a mix of abstracts from a variety of articles just written about the emerging and exciting new AI ‘technology disruption’ which offers the prospect of revolutionising productivity by enhancing and quickening most of the processes currently undertaken either at work or play and so increasing outputs and improving outcomes

It may well have a bigger impact on the way we do things than the internet or iphone

It’s a game-changer – what the new Knowledge world has been needing for years – perhaps the ultimate productivity device – the support that cuts the waste of time or time-consuming work from most processes 

The following extracts are from a mix of articles by Erica Pandey, Dan Primack and Ina Fried – plus Mark Sellman and James Marriott – plus Sabrina Ortiz – plus Mélanne Ghahraman – plus

Axios on facebook
Axios on twitter
Axios on linkedin
Axios on email
1. What is ChatGPT:
  • ChatGPT is a free (for now) site that lets users pose questions and give directions to a BOT (aka Robot) that can answer with conversation, term papers, sonnets, recipes — almost anything, in almost any style you specify
  • GPT definition = Generative Pre-trained TransformerNOT General Purpose Technology as used to describe electricity or the internet!
  • Who behind it?

    • ChatGPT was launched in November 2022, having been developed by a US company, OpenAI 
    • Most of the technologies it uses were invented at Google, Meta and Deepmind
    • Microsoft is now investing $10bn in OpenAI to incorporate ChatGPT into its search engine Bing
    • Google and Meta are working on similar technologies
    • Anyone can use ChatGPT for free via a simple web interface, although there’s talk of OpenAI planning to launch a paid version
  • What does it do?

    • Most software is specifically coded to do certain tasks – if the programmer didn’t think of it, the software doesn’t do it
    • Generative AI programs like ChatGPT, though, can create unique content in response to user prompts
    • Its power comes from being able to write sentences because it can accurately predict the next word to write, like auto-complete but on a huge scale
    • Users are able to ask a wide range of questions and it returns an answer almost immediately
    • ChatGPT is a system designed to generate content based on the specific prompts that are fed to the chatbot
    • The chatbot uses large language models to get a better understanding of user questions and respond accordingly – whether it’s writing emails, code or summaries for YouTube videos, ChatGPT can automatically create text that responds to the requested topic
    • The main goal of this initiative is to increase performance and productivity between businesses and employees.
  • Other GPT systems?

    • Lensa AI = Digitises portraits
    • Fliki = Turns text into video with voices
    • Cedille = A writing assistant
    • Onloop = For performance management
    • Dall-E = Another OpenAI project which creates realistic images from a text description and allows customers to change the picture on the front of greetings cards, say
    • Stable Diffusion = Forces journalists to reckon with ways automation could improve their jobs, or displace them.

 

2. AI Pluses – What it can do:

  • ‘Moravec’s paradox’ = Things people find easy, AI finds difficult – and things AI finds easy, people find difficult
  • A lot of the tasks it is capable of doing are tedious and time-consuming for humans viz:
    • Writing an introduction or some code is a small proportion of the total
    • Automating the processing mortgage applications
    • Helping overworked nurses in the UK during the height of the Covid pandemic.
    • Passing MBA final exams at Wharton School of Business – “It did an amazing job answering basic questions on case studies but less well on maths and advanced analysis” – “Deserving a mark of B or B minus”
    • Passing Bar exams – albeit it earned only a C+ from the University of Minnesota
    • Building apps
    • Co-hosting podcasts
    • Passing all three parts of the U.S. Medical Licensing Examination, although just barely, as part of a research experiment
    • Generating emails for you based on text prompts.
    • Writing code and creating simple websites.
    • Revolutionising the education system – plus it will allow pupils to cheat by writing their essays, bringing an end to homework (about time too!)
    • Automating menial news-gathering tasks, like aggregating data, but there’s a growing concern that an over-dependence on it could weaken journalistic standards if newsrooms aren’t careful – Newsrooms have been using automation for years to strengthen their journalism and free up time for strategic functions that require more human judgment, like investigative reporting.
  • You can also tell the chatbot you’re a gluten-free lover of Italian food and it’ll spit out a meal plan and grocery list for you in seconds
  • And it will even help us fall in love by generating personalised messages or poems for Valentine cards, say

3. AI Minuses – What it can’t do:

  • The current version of ChatGPT has some severe limitations, as even its creators acknowledge:
    • It’s prone to mistakes and generating wrong output – aka hallucinations so it’s unreliable for now
    • It will require supporting mechanisms, tests, and other quality checks to use it at a scale where quality and accuracy are important
    • It can’t distinguish fact from fiction – for sure, humans have trouble with this too but they understand what those categories are – as a result, it confidently asserts obvious inaccuracies, like “it takes 9 women 1 month to make a baby.”
  • If we want to assess its reliability, it can’t tell us where its information comes from.
  • Its information is outdated – ChatGPT’s knowledge of the world ends sometime in 2021, though this is probably one of the easier problems to fix.
  • It can answer a lot of questions, but it doesn’t actually “know” anything — which means it has no yardstick for assessing accuracy, and it stumbles over matters of common sense as well as paradoxes and ambiguities.
  • When used by news site CNET to write financial articles about corporate earnings, they were filled with errors
  • However, now that the technology has become so advanced and accessible, it’s become harder for newsrooms to draw the line between leveraging AI and over-relying on it.
  • For many news brands, using AI to write stories would not be worth the risk to reader trust – “The ability to audit how the AI did what it did is very thin, and that’s always going to be a concern for the news industry,”
  • Educators are also worried – New York City schools were quick to ban the technology from their networks, an approach that many see as wrong-headed, akin to trying to prevent students from using calculators or Wikipedia – a better approach could be to embrace the technology and encourage students to identify where it falls short — a set of skills that will be critical for a generation of knowledge workers that will be toiling alongside such technologies.
  • Biassed? – If its data is determined solely by its originators then its wisdom offered will be much like the limited first few pages of Google – biassed by their ‘search’ algorithms about what’s known and out there (and who pays most to appear on first pages) so keen competition is vital to ensure AI gets better and better – like the internet which is under nobody’s control

 

4. Who against ChatGPT?

    • Yann LeCun, Meta’s chief AI scientist – “A flashy demo – not a particularly interesting scientific advance – it just regurgitates text that it’s been trained on and has no understanding of the real world so lacks basic intelligence – most of human intelligence is not text-related – however, it is well engineered”
    • Gary Marcus, professor of neural science at New York University:
      • “I don’t think it can solve the truth problem – it’s not actually analysing its dataset saying ‘this thing I am saying, is it consistent with what I know?’ And this lack of validation step  in my mind is fatal for making it a serious, full-service search engine”
      • “AI cannot be creative because it lacks internal understanding, is merely a ‘king of pastiche’, blindly pasting together influences from human artists (but that is what original human artists were doing)
    • Education expert: “ChatGPT sometimes produces superficially plausible essays that fall apart under closer scrutiny” – but plenty of humans write essays like that too

 

5. What next?

  • Instead of replacing human workers, AI will be used to assist and enhance their abilities – by taking on repetitive and mundane tasks, AI will allow humans to focus on more complex and creative endeavours
  • AI can already generate new ideas and insights that humans may not have been able to come up with on their own
  • But the more these models grow, the more they will require assistance – creating a need for new jobs and more studies – n.b. ‘Just because you have a tool box doesn’t mean you know how to fix a broken appliance’
  • And before any major transformation in our everyday lives, the technology has to be reliable and proven to make no errors – and we are not at that point – yet
  • OpenAI and other AI companies will keep pushing to improve accuracy, reduce bias and eliminate other problems – as of now, no one knows whether the technology’s drawbacks can be overcome — or whether ChatGPT and its successors might ever become truly dependable.
  • In the meantime, the applications will get much more specific:
    • The real power, many say, will come as businesses combine such AI tools with their own data.
    • A company like BMW would be able to merge real images of its logo and cars with the output of a tool like Dall-E to create commercials at a fraction of the cost.
    • Or studios like Disney could apply generative technology to develop an endless array of sequels, spin-offs or games built around their existing universe of characters
    • Today’s AI couldn’t complete such projects, but it can provide endless shortcuts.
  • And the hype is going to get crazy. The excitement around ChatGPT and generative AI is real, but the tech industry loves to go overboard. With the tech economy in retreat, every struggling company and fledgling startup is eager to hop on a bandwagon – ChatGPT itself can be used to generate more hype. That creates a greater need for smart journalists who can separate fact from fiction — assuming the technology hasn’t already put them out of work.

6. Conclusions:

  • The technologies underlying ChatGPT will ‘dramatically’ automate many of the mundane writing and coding tasks – but there’s a lot of creative and unique writing and coding that involves imagination, synthesis and other complexities that won’t be easily solved shortly by these AI tools
  • Thus, humans will be required to produce the correct output
  • AI tools are good at delivering words or code but less good at knowing whether it is any good, or even accurate
  • Without human direction from a person well-versed on the subject at hand, these tools are much less useful
  • Hence business leaders are currently trying  to figure out if tools like ChatGPT could enable them to cut costs, offer new products or get a leg up on competition.

 

 

Leave a Reply

Your e-mail address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.