ChatGPT as alternative to Google or Wikipedia

Because a number of articles this week suggest Google views ChatGPT as a competitive threat and hopes to respond by adding chat capability to Google Search.

It makes sense for OpenAI in turn to add search capability to their chat software.

That said - the ability to incorporate search technology into ChatGPT makes immense sense at addressing concerns about the accuracy of its output. If a user has the option to request links to the source of ChatGPT’s response, then that could immensely increase the use cases for ChatGPT while simultaneously reducing concerns about inaccurate responses in situations where accuracy is essential.

Plus so many more potential uses - like "Show me the flight tomorrow PIT-ORD with the shortest elapsed travel time " or “Show me all restaurants in Podunk IL which serve Grilled Chicken Caesar salad” etc etc.

Doesn’t Google already do a pretty good job with questions like this? It did when I submitted your examples. (I substituted Peoria for Podunk.) Plus it gives links to its sources. How can a service like ChatGPT keep up when it seems limited to writing plausible text but makes things up and cannot show its work?

ChatGPT would be a huge benefit because it does an excellent job at letting the user to specify the nature and format of the desired output. So if the info were accurate it would be essentially the world’s most easily used web scraper.

That said - I do not know if ChatGPT itself is an accurate source to answer what are essentially business questions for OpenAI. But with that disclaimer, it does seem to (unfortunately) agree with your conclusion that this is not likely to occur:

But even not Google is giving you „an answer“ onto your question in normal cases!
It offers you hundreds, often millions, of possible answers, and you have to search the results for the right/fitting/best outcome!

Exactly. Sources matter, how you came to the conclusion matters and output formats are worthless if the data contained therein are BS.

3 Likes

Totally agree

And that is why I have to believe that the ability to report sources is high on the list for ChatGPT.

Without that ability its real-world utility - except for amusement and curiosity - drops immensely.

With that ability it becomes a life-changing tool for both business and educational applications.

I believe we humans have a peculiar bias, for we tend to underestimate the evolution of these models. The GPT-x evolution is nothing but formidable (the breakthrough being the “Attention is all you need” paper from Google team) and the main issues pointed in this thread that make ChatGPT a confident bullshitter will be solved in future releases.

I did not see this coming but it is here to stay. As humans we need to reflect on second order derivatives of this technology, many white collared jobs are going to get heavily impacted (writing code is only one of them) and I am not that confident our societies are prepared for the wave that is coming. Of course new jobs will be created, as with any other disruptive technology that delivers real value, but it will take some time: right now, ChatGPT satisfies more work demand than it creates.

Also of note: the aforementioned evolution is not eternally exponential unless some foundational breakthrough happens. These things are autentic data guzzlers and the volumes of data for training grow beyond the exponential. Some sources mention that as soon as 2026 there will be no more available textual data to train the successors of ChatGPT.

2 Likes

Wow. I used the same prompt and ChatGPT gave me five different links. However, as you noticed, every single link was for some obscure study having absolutely nothing to do with vegetables.

I then mentioned the error, and it apologized and produced five different links, each having the same problem where they had nothing to do with the title given. Nothing to do with vegetables. I asked it why, and it finally apologized, gave me a summary of its understanding of the benefits of vegatables, and no further links.

For this particular use case, which is the one most people seem excited about, ChatGPT is currently wildly off. Can you imagine if Google search produced links that didn’t match their descriiptions? It would be out of business in 30 days.

For comparison, I went to PubMed and searched for “are vegetables healthy” and it gave me 1,145 results. On the first page, each result had the word vegetable both in the title and in the body of the article/paper. If ChatGPT can’t even replicate that simple task, it is far from complete. Perhaps the next generation will show an improvement…

One more thing I have noticed. ChatGPT is very wordy, and it writes like someone who learned English in the last year. It writes very ‘flowery’ and it does not sound natural. It’s like it trained on writing produced by ‘english as a second language’ content farms. I don’t think any novelists or technical writers have any reason to fear for their careers…

If you had a relative in your family, a cousin, let’s say, who had earned himself a reputation as a confident bullshitter, would anyone ever trust him again? Why would they?

I hope it gets better. Just for fun I’m posting the experiment I ran in my last newsletter. I wanted to see how good it was if asked good questions around Scrum (the focus of my work). I will let you judge the results:

I think many of my readers have tried ChatGPT or in the past few weeks. I thought it might be fun to see how good a job it does.

Describe the purpose of Daily Scrum

The Daily Scrum is a short meeting held every day by a Scrum team to synchronize activities and create a plan for the next 24 hours. The purpose of the Daily Scrum is to inspect progress toward the Sprint Goal and to identify and remove any impediments that are blocking the team from meeting that goal. The Daily Scrum also helps the team become more efficient and productive by ensuring that everyone is on the same page and has a clear understanding of the tasks they need to complete. Additionally, the Daily Scrum helps to foster collaboration and communication among team members.

On this one it does an excellent job. It is missing all of the mechanics - but I didn’t ask for that. It misses the idea that the event was intended to eliminate other meetings from the team member’s work day.

The tool should take a bow, it does better than most humans so far. Letter Grade: A

On to the ScrumMaster:

Describe the role of the ScrumMaster

The ScrumMaster is responsible for facilitating the Scrum process, ensuring the team is following the Scrum framework, and removing any impediments the team may encounter. The ScrumMaster also serves as a coach and a mentor to the team, helping them to become more effective and efficient. The ScrumMaster also helps the team to identify and prioritize tasks, as well as manage and monitor progress. The ScrumMaster is ultimately responsible for ensuring that the team is successful and that the Scrum process is adopted and followed.

This one has some more issues:

  • “removing any impediments the team may encounter” - True - however it doesn’t cover the oft missed point, much of that comes from coaching the team in how to remove their own impediments
  • “The ScrumMaster also helps the team to identify and prioritize tasks, as well as manage and monitor progress.” - The SM has no role identifying tasks, prioritizing them. Nor do they manage and monitor progress.

Outright missing (quotes from the ScrumGuide): “Coaching the team members in self-management and cross-functionality”, “Helping the Scrum Team focus on creating high-value Increments that meet the Definition of Done” and “Ensuring that all Scrum events take place and are positive, productive, and kept within the timebox.”

So far on ScrumMaster they might get a C- for turning the ScrumMaster into a manager and missing so many elements from Scrum the ScrumGuide

Finally this answer completely misses their service to the Product Owner and the Organization. So I asked it a supplementary question:

Describe the role of the ScrumMaster in coaching the Product Owner

The ScrumMaster serves an essential role in coaching the Product Owner. The ScrumMaster works closely with the Product Owner to ensure that the product vision is clear, the product backlog is well-prioritized, and that user stories are well defined. The ScrumMaster also helps the Product Owner to understand the value of each story, and to ensure that all stakeholders are working together to move the product forward. The ScrumMaster also works to ensure that the Product Owner is taking an active role in the product development process, and that they are making decisions that are in line with the product vision. The ScrumMaster is an important source of guidance and support for the Product Owner, and helps them become a better leader and product owner.

Scrum doesn’t require User Stories, but let’s ignore that detail.

Issues:

  • “to ensure that all stakeholders are working together to move the product forward.” - Scrum doesn’t require the stakeholders to work together, I’ve encountered more than few situations where the stakeholders represent competing factions in an organization.
  • “… works to ensure that the Product Owner is taking an active role in the product development process, and that they are making decisions that are in line with the product vision.” Yes the PO needs to do their job, no the ScrumMaster isn’t their baby sitter.
  • “…an important source of guidance and support for the Product Owner, and helps them become a better leader and product owner” - More of the same, my late father would have called this drivel

Outright missing (quotes from the ScrumGuide): “Helping the Scrum Team understand the need for clear and concise Product Backlog items” and “Helping establish empirical product planning for a complex environment”

Ok I’m giving it a D - here, it’s barely passing.

So when it comes to explaining basics of Scrum our machine learning tool is currently getting a C

Last thing I thought I might try ask a real coaching question from a recent Lean Coffee:

What functions/tasks should a BA perform on a cross-skilled scrum development team?

  1. Facilitate communication between the Agile cross-skilled development team and stakeholders.
  2. Develop user stories and use cases to ensure customer needs are met.
  3. Analyze customer requirements and create a product vision.
  4. Participate in sprint planning, daily stand-ups, and retrospectives.
  5. Prioritize tasks and manage the product backlog.
  6. Identify and remove roadblocks for the team.
  7. Help break down complex tasks into smaller, achievable goals.
  8. Monitor progress and report to stakeholders.
  9. Develop and maintain a product roadmap.
  10. Make sure the product meets customer goals and expectations.

Wow. At best they turned our BA into a warmed over product owner. This answer is so sad, I will just grade it an F and go off to make coffee. Do you want to discuss a better answer without any AI? Come to our next LeanCoffee.

Apparently the AI needs some more real world work experience before we set it loose on unsuspecting victims. To be fair, I suspect the errors are example of what is said in the training data. The old adage Garbage In, Garbage Out predates machine learning by over 60 years.

The funny thing, my father complained AI was promised his entire career and never got it. I can tell him we’re still waiting.

[1] GIGO Garbage in, garbage out - Wikipedia - “The first known use is in a 1957 syndicated newspaper article about US Army mathematicians and their work with early computers, in which an Army Specialist named William D. Mellin explained that computers cannot think for themselves, and that “sloppily programmed” inputs inevitably lead to incorrect outputs.” - Article clipped from The Times - Newspapers.com™

1 Like

Would take some time, of course, but I would not say I’ve never trust him again. People -and software!- can change.

I would not say this is GIGO by any means. You are a SME on Scrum, that’s for sure, and are entitled to think that ChatGPT is spitting out garbage on your subject domain. If you happend to meet a stranger in a party and started saying what ChatGPT is saying, would you think they are saying garbage or would you be thinking “Gee, this person knows about Scrum, I can have a discussion with them”.?

This is the kind of depth you can expect to get from ChatGPT from almost any topic, I find it is a massive breakthrough.

1 Like

@pantulis The comment comes from the fact the training data was likely of low quality. It is successfully regurgitated randomly bad information on the subject of Scrum. My real point is that these tools are only going to be as valuable as the training data.

Clearly it is unusable in its current beta release. This release is focused on natural language interaction for both questions and answers; by that measure its performance is phenomenal.

The huge remaining question is whether its ability to produce accurate links to sources of information will be included in a future production version. If yes, then it would be extremely useful as a product. If no, then clearly it would be better for ChatGT to simply refuse to provide references at all than for it to provide outright fabricated references.

For eight decades, we have waxed and waned in our enthusiasm for Artificial Intelligence. I’m with this guy …

The Promise of Artificial Intelligence

In this provocative book, Brian Cantwell Smith argues that artificial intelligence is nowhere near developing systems that are genuinely intelligent. Second wave AI, machine learning, even visions of third-wave AI: none will lead to human-level intelligence and judgment, which have been honed over millennia. Recent advances in AI may be of epochal significance, but human intelligence is of a different order than even the most powerful calculative ability enabled by new computational capacities. Smith calls this AI ability “reckoning,” and argues that it does not lead to full human judgment—dispassionate, deliberative thought grounded in ethical commitment and responsible action.

… Smith worries that, impressed by AI’s reckoning prowess, we will shift our expectations of human intelligence.

Don’t let your enthusiasm for the promise of Artificial Intelligence blind you to its ultimate limitations.

2 Likes

The potential is for training to improve the output quality beyond the quality of the original training data. First, human-scored training, then adversarial training, and finally a self-adversarial training (at our current level of knowledge. This has been the progression in simpler domains, e.g., playing the board game Go.

That is exactly why my interest and “excitement” would be in a ChatGPT that can efficiently collect items from the web based on my instructions.

Interpreting the information is my role; I agree that AI to interpret that info and also write about it is pie in the sky.

But I think having AI collect the info to make my job of writing easier is quite realistic.

FWIW This was my father’s comment. He was one of the earliest Computer Scientists - earning a PhD in the late 1960’s.

He frequently grumbled that Marvin Minksy promised him AI from the 70’s on. As I joked in my newsletter - Dad we’re still waiting. The best we have today is things that regurgitate text.

3 Likes

But the sole “breakthrout” is, that the software is able to talk to you.
The Problem is, that the time it is talking to you is a completely waste of time, because all that is said just is hot air…
And that is the Main Problem within the current discussions about ChatGPT, the most people do not understand, that the only thing this “AI” is doing, is talking, but without ANY kind of knowledge about the Topic it is talking about.

2 Likes

This AI has to be specially trained for the specific science area like an Scientist itself, to be of any help.
Otherwise it is spilling you with more or less useful information, and wasting your time while you dig thru that, hopping finding the useful Needle in the Haystack of Informations.