Study: GPT-4 outperforms Data Analysts
ฝัง
- เผยแพร่เมื่อ 9 มิ.ย. 2024
- 📈 My SQL Course 👉 • SQL for Data Analytics...
Courses for Data Nerds
==================================
📜 Google Data Analytics Certificate 👉🏼 lukeb.co/GoogleCert
🐍 Python for Everybody 👉🏼 lukeb.co/PythonForEverybody
💿 SQL for Data Science 👉🏼 lukeb.co/SQLdataScience
🧾 Excel Skills for Business 👉🏼 lukeb.co/ExcelBusinessAnalyst
📈 PowerBI for Data Viz 👉🏼 lukeb.co/PowerBI
📊 Tableau for Data Viz 👉🏼 lukeb.co/Tableau_UCDavis
🏴☠️ Data Science: Foundations using R 👉🏼 lukeb.co/RforDataScienceJH
➕ Coursera Plus Subscription (7-day free trial) 👉🏼 lukeb.co/CourseraPlus
👨🏼🏫 All courses 👉🏼 kit.co/lukebarousse/data-anal...
Books for Data Nerds
==================================
📚 Books I’ve read 👉🏼 kit.co/lukebarousse/book-reco...
📗 Data Analyst Must Read 👉🏼 geni.us/StorytellingWithData
Tech for Data Nerds
==================================
⚙️ Tech I use 👉🏼 kit.co/lukebarousse/computer-...
🪟Windows Virtual Machine for Mac (Parallels) 👉🏼 lukeb.co/ParallelsFreeTrial
Social Media / Contact Me
======================
📫Newsletter: www.lukebarousse.com/
👨🏼💼 Linkedin: / luke-b
🅧 X/Twitter: / lukebarousse
🌄 Instagram: / lukebarousse
⏰ TikTok: / lukebarousse
As an Amazon, Coursera, and Parallels Affiliate Programs member, I earn a commission from qualifying purchases on the links above. It costs you nothing but helps me with content creation.
#datanerd #dataanalyst #datascience - วิทยาศาสตร์และเทคโนโลยี
"The purpose of this work is not to replace the data analyst role or to create anxiety. Instead, we would like to explore the potential of GPT-4 to aid human data analysts in more efficient working."
Well, the corporations don't give a shit what you would like. If it's cheaper and more efficient to replace analysts, they will.
The truth of this always is that the companies are trying to make the work of one person more efficient so they can fire the second person and save money.
@@w0rp obvious to anybody who is an actual analyst
How are companies going to make money when people have no jobs?
@@AccessKelly data analysts won't have data analyst jobs, they will have other jobs
and if they don't, they're only a part of the economy
but overall I agree with your argument, I also thought about this and I think the massive displacement will cause a deflationary spiral
To put it plain and simple: Query and plot is not what a good Analyst is doing. The majority of the work is figuring out needs and blindspots, building an analysis plan or even start a data collection project, than building pipelines to get all of that data into one place. After that, figuring out what questions should be asked and than at some point serialising all of that into a product.
AI in that context is „just another“ query language.
Yes, I agree, but I'd also add on not just gathering or analyzing the data, but also cleaning the data that I find takes the most time. Maybe my work just has a lot of data quality issues (a lot of it is manual from a variety of sources that don't all track or report things in the same manner or consistently), but I would never trust ChatGPT to identify and attempt to fix issues in a dataset or to check a messy join to see if it's "good enough" or could be made better. ChatGPT generally needs to be told exactly what to do, but no one's going to know what the problem is until they open it up and start exploring the data. You need a human being to identify problems and prescribe solutions and then maybe ChatGPT can help speed up data cleaning or revalidation by automating some steps. It's still (and I think for a very long time) going to be at best an assistant at certain steps in the process.
My dreams of living in a sunny coastal town working from home for 4 hours during the day and sipping fruity cocktails by night while still living a decent life are slowly but surely withering away
Nah, just work 40 hours once and create a script that analyses data with an LLM and then work 4 minutes a day.
But don't forget to write a script that searches the internet for people doing the same thing and automatically solicits hit men on the dark web because if everyone does this it won't be viable.
instead of sorting numbers maybe you should try building something
@@jgonsalk lmao
I've always felt that if something was easy and offered a shit ton of rewards then it would just attract more and more people until supply catches up with demand and the benefits are gone. I think any easy high-paying job is temporary, because your boss definitely doesn't LIKE paying you a lot- he's probably spending every waking second looking for either someone to do it for a cent less than you, or some(thing) that can do it for free. Now he has the latter.
@@jgonsalk This would work only if capitalism didn't rely on endless growth. Endless growth means that instead of you working less for more pay, you will instead do more work for the same pay
Time to start watching some farming videos
😂
They have enough water for farming where you live?
I´ve just enrolled to welding classes
@@NutsAboutData That's going to be some hard 12 hour days.
AI-embodied robots are coming for labor jobs too
I trained an LLM on your SQL course, saved myself 4 hours
😂😂😂😂
Dangit Dave! You weren’t supposed to tell people this!
May I know how to do that? @@
What is LLM
Hahaha
thanks, just finished my Data Analyst course....hope the machine will not replace me soon )
Don't worry, the boss needs someone to fire when the AI is wrong, and the AI is too smart to be fired. AI verification engineer required :)
My two cents is the bottleneck for replacing DAs with AI is the stakeholder communication and the speed of which companies will adopt this technology. Would the stakeholders who are communicating with AI feel confident enough in themselves that they queried the AI correctly to retrieve valid results. I love AI. It helps me out at my job everyday but I don’t see it replacing DAs soon. Keep studying data analytics everyone and use AI to learn more about it! It’s a great field to join! Hope you all break in.
The big difference is speed as much as accuracy, a machine never forgets, does not marry and have kids, do not get sick days, neither respects working hours, always available, no attitude problems, will never unionize.... works on weekends, are dirty cheap to use, you can use more than one to contrast results, you can get a standardized industry model and will never quit. If you are as accurate as that machine, there you go, you have a chance.
@@carlosr7804 It also has no clue when it's wrong. If somehow it accidentally came up with -4 days bookings, it would have no clue that it doesn't make sense. Even the intern is going to know something is wrong and go figure it out. Also, how do you get it to self-correct? That's not something these things can do. It's one shot, answer, I don't care if I'm right, I'm going to say it confidently and fool all of you.
Fascinating stuff! One thing not mentioned is that in order to make the prompts work you need to be a.... data analyst. So, at best, ChatGPT is going to make data analysts more productive. Give these tasks to a non-data analyst, and they wouldn't be able to create this stuff in GPT.
(Yet: I guess that could change some time)
100%
how about 1 data analyst for prompts doing the work of 10, instead of 10 analysts 🤯
@@elOtorongo96aint no one is working their ass off that much. we are not robots.
@@elOtorongo96 Also you need to evaluate these reports. No one is trusting AI blindly. Normal human cant properly evaluate 10 reports a day.
also One big issue with the study is that the data was OPEN SOURCE. it’s not evaluated for proprietary internal company data.
I seen how GPT 4 performance in data that is not publicly disclosed or one that is not trained yet and its laughable 😅
Exactly. The stakeholders aren’t going to know what to ask lol
As data analyst ive never been paid to code or to just make graphs and say what they mean, im paid mostly to understand business and use the data to deliver better products, if i cant understand which information are relevant to the business, theres no work done, its something that llm cant do.
Of course you can do a list of questions to be answered but the model are unable to understand which one is important to be presented and that is a issue that i cant see being solved quickly specially because those needs are super dynamic.
But
People who are the kind of analyst who just plot graphs with same simple evaluations in mind, they should be aware
what do u mean by they should be aware
I have no idea what a data analysts day looks like but i gasped "that's it??" when i saw what chat gpt did, sure it's impressive kinda but our undergraduate physics lab data analysis was the hardest thing in the world compared to this
Bruh, 31 yo and just started a new bachelor related to data science. Still in my 1st year and now i see this?! I'm dead. Like, what else am i supposed to do now? Im a guy, so no onlyfans option. Lol
Well, you can still do ML, NLP, data engineering and business intelligence and AI so it's not that bad.
😂
You are in Data Science not Data Analytics, right?
Plenty of guys make money on only fans, lots of gays pay money for that
start selling the shovel
Man, from what the study shows, I think Data Analysts have a lot to be happy about here. This essentially means, that aspects of data analysis that could not be automated before, can be automated now. The study shows that GPT-4, if given some very specific instructions in natural language, can provide some very valuable outcomes. But this does not replace the need for human data analysts, because it's the humans who can take the fuzzy questions (e.g. hey Luke, why the heck is our sales low this year?) and figure out the best way to approach the problem. So, in this example, Luke can find out what metrics he should be looking at to find the answer to the question.
As a result, in the end, you still need your human senior data analyst like Luke, he is the one who can bring order to chaos. Once he's done that, he can apply some unprecedented levels of automation thanks to GPT-4. So, it is in the hands of your experienced analysts that LLM's like GPT-4 are such powerful tools.
Agree. This was a very carefully staged experiment on a clean dataset with a specific set of questions. Datasets are never that organized and ready to go.
Which means the company can fire all of Luke's coworkers and make him do all the work for less pay, since most of his skill set is now obsolete. He's just cleaning inputs and prompt engineering. In a few years, he can be replaced by someone even cheaper with a skill set focused on those areas rather than data science in general. BTW, If the paper authors didn't intend this to threaten data analyst jobs, why did they include a cost comparison?
@@adrianmizen5070 It's not as simple as that. Prompt engineering is easy when you have instructions as clear as school homework. Any time you have a fuzzy question - the AI is as lost as a five year old in the woods. But yes, you're right, Luke's 'co-workers' who are doing the more "mundane" parts of the jobs, are likely to be replaced by AI.
@@saminyead1233this sounds like cope tbh, when companies want to cheapen out they will
How long did it take the researchers to figure out what prompts to write and how to write them. Did they know the data in the datasets beforehand that allowed them to tune their prompts?
Bruh they want us all to be fast food workers and janitors
😂😂 true. Bartender , body guards, cab driver etc etc
Fast food workers will be replaced by androids (it's far future but still fact). Nvidia, Microsoft, Amazon, Google investing trillions in robotics and AI 😅
Skill trades is the only way forward.
@@mndtr0 Actually, there are almost no cashiers in the europe for Mcdonalds
@@sidharthmandal9957Cab driver will be replaced by autonomous driver
ChatGPT summary :
The video transcript discusses a research paper evaluating whether GPT-4, a large language model, can effectively serve as a data analyst. The study compares the model's performance against human data analysts in terms of speed and cost-effectiveness. GPT-4 was found to be faster and significantly cheaper than human analysts, particularly senior ones. The study focuses on three major job scopes of a data analyst: data collection, data visualization, and analysis. GPT-4's performance was evaluated based on its ability to execute Python code to extract data, create visualizations, and generate insights. The model's accuracy and validity were assessed against human analysts, with varying results across different tasks. While GPT-4 outperformed intern and junior data analysts, its performance was comparable to senior analysts in some cases. However, the study acknowledges limitations in the specificity of the questions posed to GPT-4 and the small number of human analysts evaluated. Further research is needed to determine the extent to which GPT-4 can replace human data analysts. Overall, the study aims to explore how AI models like GPT-4 can aid human data analysts in improving efficiency.
Wow, you know how to copy and paste! I'm so impressed. Let me know when you have a real brain.
Using 'Study' in the title is a smart choice, avoiding using people's worries just to get views. Nowadays, it's hard to find creators who care more about making good content than getting clicks. You deserve a subscribe!
a high quality content and a great video as always, thanks a lot Luke 💙
you're my role model in data analytics. 🙏
How could that non-sensical paper even pass peer-review in such a high impact journal?
Given the tiny sample size and the fact that only standardized benchmarks have been used, neither the title, the abstract nor the conclusions make any sense:
It is quite obvious that even GPT 3 would outperform most humans on benchmarks, which are probably part of its training data given the timestamps in the github repository they cite.
Also the term „correctness“ seems ambiguous in the context of data analysis/science (and many other fields) where your answer most of the time can only be an approximation to the „correct/exact answer“ if that one singular correct answer even exists. How messy was the dataset, how was it cleaned, was there some complicated imputation necessary, what metrics were used and what assumptions/approximations were necessary to do the statistics?
If any of those questions arise in your analysis, your answer cannot be „correct“ but only an approximation to that. But they seem to have treated correctness as a boolean, which either implies that those data were trivial to analyze or that they actually have no clue about basic science.
The fact that they compared the costs of humans to a machine makes it quite clear to me that they had a strong bias (maybe even OpenAI sponsored/backed?) as the comparison of cost/hour is quite meaningless in that context. If that metric was meaningful without further considerations no business would even consider switching to the Cloud.
I guess, if the paper was published without that AI hype, it would have been rejected from each and every reputable journal. I mean the paper reads as if it was written by some bachelor students who are tasked to gain some scientific experience by writing some report about an internship and it was published in a journal with a significant impact factor of > 9.
I am really more astonished by this than by their „findings“.
Thank you for fighting the good fight out there. We are a minority. Pretty much every AI paper published now writes the conclusion section first to sound sensational, and then back fills in how to get it there. Sadly, it's such an interesting field to study, but something bad is going to happen or leak and all the money is going to dry up after the scandal. Aka, the bubble is going to burst and it's going to hurt all the good people doing real work.
Of course companies will give tons of confidential data to ChatGPT to perform the analysis.
Is that what you really think will protect jobs? Lots of companies use apps like chrome and do confidential, sensitive work on such apps(saying this from my own experience).
These companies will easily find way around concerns for privacy and confidentiality.
I mean seriously every company use 3rd party tool for different types of work. Just because it is AI it's gonna change much.
@@okplld I did't say anything about protecting job. However, AI applications have opened a new horizon to data scientist: "AI Engineering". Companies will need AI Engineeris who can build their own custom AI application for their specific purposes. Even Microsoft Copilot will replace basic exploratory analysis and the only input that it will need is a spreadsheet.
Staying only in the area of data analysis is dangerous.
😮@@okplld
Well for one, gpt enterprise exists, which is completely private, but if it didnt i bet there would still be companies willing to do it to save money
Probably business will run own finetuned models on protected servers.
I would like to see how GPT works when a non data analyst asks the questions and gives the promts. The main part of any tech worker is to interpert what the non tech guy wants
Great video Luke! I am encouraging my team to use GPT4 to accelerate workflow. One thing I'd be interested to see you cover is how well these tools perform with raw qualitative data. My early tests have shown some promising results, but my concern is they are hallucinations that just sound thematically plausible.
Hello Luke 😀 hope your all well. Hmm is there a way to check the version on dataset you put up on kaggle on the datajobs via the Kaggle API ?. I just want to make sure that I don't start downloading a dataset if I already got it 😀
As a senior data analyst this is terrifying. Admittedly most of my time as a senior is not spent analysing data but mostly on architecture, requirements gathering, and project management but you only get to that point by being able to analyse data well.
😮 whats your advice to someone just graduating with a data analytics degrew
It's going to be the same story as with SQL. SQL was built for people that have no computer experience so they can deal with data without help of programmers. It didn't happen though.
It's going to be the same for AIs - there will be 'AI programmers' that will write correct prompts.
Nuh. SQL is hard for non technical people. Natural language is way easier
@@AcctistaZFrom my experience working with neural networks in order to get what you want from them you should at least be able to write technical tasks and formulate exactly what it should do.
It's not natural language per se.
unlike SQL, prompt engineering a completely different skill/talent from programming. If you're writing something in Python, SQL, assembly, whatever, you know exactly what the program will do in response to what you write. Whereas, there is no ChatGPT language spec, you have to get to know its tendencies and experiment a lot to see how it reacts to prompts.
@@derpycyclistjr.1943I say it’s still many times easier than mastering sql.
They have a limited sample and control group. The data is also already prepared for GPT4
LLMs are extremely good at doing simple, isolated tasks that are clearly defined. Definitions are in the form of clear parameters set out by user prompts.
That being said, it's only for short repetitive tasks, and also at the risk of not providing enough context, prone to errors due to that or other factors such as unreliable external data, hallucinations, misunderstanding. Thus, they should always be a tool and require human reviewers for the outputs. For example, image recognition has been developed for X-ray scans for diagnosis, but human doctors still have to review them as they are prediction models and not absolute. The same thing goes for data analysis, outputs requires SMEs to review them.
Does this affect the junior workforce of DA? My guess is that it will likely it will, depending on the company. Analysis work is not finite. Having more efficiency does not mean there will be a lack of work. For example, SaaS products such as Tableau and PowerBI are simple to use for non-technical users, but it simply changes the way data professionals work. LLMs will be likely to be part of the workflow of many professions, including DA, but not replace them. Smaller companies that have a tight budget might use LLMs instead of hiring DAs as the expense really affects them, and they just have to deal with the downsides. But larger companies are aware and want good reliable intelligence as they want to have a competitive edge over other companies.
The tech field has always been constantly changing at a very fast rate, with many technologies such as blockchain, cloud, distributed computing, SaaS, IaaS, PaaS, etc. And AI is going to be the next one to rapidly shift the industry. Professionals will have to adapt and embrace new technologies or be left behind. That's how it has always been.
Tl;dr: introduction of LLMs will decrease demand for smaller companies without budget, but bigger companies are aware of the downsides and need human SMEs to review the data. Demand and supply rise and fall due to the hype. Gradually, people will understand the limitations of AI and what it does well and not.
For anyone reading all that, thank you. I appreciate any counterarguments and criticism! :)
Teşekkürler . Yapay zeka alanında çalışmak için MacBook Pro m3 pro almayı düşünüyorum . Pyhton programı kullanacağım . Bu laptop u önerir misiniz ? Başka bir öneriniz varsa yazar mısınız ? Çok teşekkürler .
Hey Luke please give a video for top course to become a advanced data analyst
Who created the framework you show at 0:45? Humans.
How long did it take? Probably hours or days.
How flexible is it considering the need to get data from other tables/files from other data sources (perhaps with columns with nonsense names, compromised data etc)? Probably very little.
What if you go beyond "atomic" requests but require long-range analysis that leads to the generation of a presentation?
I think they have taken a very specific setup here comparable to that of a machine on an assembly line, yet we see that men are still employed to do labor work.
Do you have a newsletter sir?
Please add a link to the research paper or add the full title of the paper (and journal name, author,...)
arxiv.org/abs/2305.15038
Automation tools don’t/can’t know the weird stuff that needs to be captured when working with real, somewhat messy data
Statistically speaking their sample of Data analyst is too small cannot conclude on the performance result, however given many small to mid business struggle also with budget to recruit they may be appealed by this, however the quality of analysis repose on domain knowledge. Also comparing annual salary against the instance is not really enough, when performing these tasks give the number of tokens needed and monetize them. Finally a data analyst year salary will not be funneled solely toward a single analysis.
I think GPT4 do great job collecting data and bring some initial insight that said, it is a matter of expertise, accuracy and cost. People tend to believe AI is cheap but it is not.
Also side not: before adding AI solutions to your business, a need of a strong data governance strategy to be implementing because it will scale and scale fast. Question: how about your data quality...?
Prompt engineering is key here. Basically look for a career to become an AI manager :)
does it tho or does it just look that way to people who don't know data analytics?
I loved your videos and wanted to get into Data science, but im glad i never got into data analytics now.
I was actually going into this field I am glad I left it early.
Me 2😢
Lol
@@userkm2what you doin now
Hey Luke,
Could you please make a new video about which operating system is most preferred by data analysts and data scientists? I've seen your old videos where you stated that Windows was the winner, but I believe times change. I want to hear your updated perspective, especially considering the new laptops available in 2024.
can you Lucke mention the link of the paper
I think that in cost they didnt take into account the time that took for they to develop the notebook, and the database used like the schemas and all that information because some neef to build it in order for the ai to function. Just using the cost off tha appi calls is just one small part.
There will always be value in people who can ask the right questions.
😂😂😂😂😂😂
Can you provide i link to this reserach paper?
arxiv.org/pdf/2305.15038.pdf
@@LukeBarousse Thanks
Can you look at pyramide analytics ? @@LukeBarousse
Yea but does it know what analysis to run?
I learnt R, data visualisation, wrangling etc etc
Just the basics, because teh quality a data analyst produces is depending on the difficulty, availablity of the sources.. ofc skill is skill.
But its pretty much enough to use gpt4-o to write reports, and plots (given that you know the theme and what you want to find/express)
Experience and skill is important in the research phase yes, but the weight may be reduced... A lot? Idk
Dey terk errr jeerrrrbs! 😢
They tuk ur jurb!
great video:)
Hi, so i am done with my studies, and my tableau license that i got through my college is expiring soon. PowerBI is not available on Mac natively, I do use my windows machine from time to time, is there any other software that you can suggest for me please
Thank you :)
Im sure powerBI has a web version which to my understanding has very similar features. Would that work? Either that or run dual boot on your pc. Or learn to use Python/R instead of SaaS products?
Tableau web?
looker is free!
The thing for me is what happens when it gets hacked etc
There are so many things to do
On the other hand better to understand and learn it now then later
We’re the data analysts allowed to use chatgpt in the study ?
The longer I work as a data professional the more I realize that being able to write code is a very little part of it.
Unfortunately, looking at the comment section, lots of people fail to notice this, which makes me worry about this industry.
6:57 to 7:00 sums up what will happen in the near future. I am glad I left this field before I became depressed.
Does this mean, if ChatGPT requires companies to pay more money for queries, then we can have our jobs back? It makes more sense for them to raise the prices in the future so the price of hiring a human person will be a cost savings.
Non-technical staff will ask the wrong questions and misunderstand the results.. I only see this being used to fend off random asks you get from stakeholders. It doesn’t account for databases with inconvenient formats or historical inaccuracy (which is often in the table overview written by a human but not as obvious from a relational diagram).
Stakeholder: I fired Larry because I can task GPT to do data analysis. Now GPT what amount of sales do I expect in the next quarter for the new shop in district 11.
GPT: In order to find sales expected values you have to *proceeds to write 10 paragraphs of procedures*.
Stakeholder: No no no, you do it.
GPT: I don't have access to the files.
Stakeholder *rings old data analyst to ask which folder the files are in*
*No response*
Hey Luke, cool video! I gotta say, when I see these GPT-4 analyses, it feels like they're just spitting out what the chart says.
You know, the real magic comes from understanding the bigger picture and weaving a story around the data. That's where the human touch makes all the difference (maybe GenAI can do better in near future).
The thing is, you need to follow the line of reasoning regarding AI use to it's logical conclusion. Nevermind replacing data analysts, the real money savings will be in getting rid of the decision makers that use the analyses.
I always think that Data Analyst job is one of the jobs that will be impacted by AI soon. Data analysis task should be in the hand of the decision maker. So the analysis and the result will be more personalized, human error would be minimized. AI should help decision makers to provide the information and insights they needed. AI-assisted tool should be able to serve most of data analysis requirements. Gen AI won't replace data analyst job, but it will surely change the job desc and reduce the number of analysts needed. I don't think the same applies to data engineering job though.
Issue regarding privacy and security won't be a problem, as option to use on-premise LLM model exists, so it can cater companies with stricter data security requirements.
don't worry gpt5 is just right around the corner😭
Ignorant data analyst and software engineers will find flaws in gpt5 and say don't worry your jobs are safe. And when it fix the flaws and come up with gpt6 these people will say the same. It is sad they are putting people in dangerous situation where they are jobless.
Absolutely right @@userkm2
@@userkm2 software engineers are very safe, AI code is most of the time garbage and it just messes the application depth and integration, roles like analysts, accountants, lawyers which are non-engineering jobs are in real trouble.
@@orange6562engineers will be gone way before lawyers, actually lawyers will be one of the last to go, but whatever keep telling urself engineers are safe until u get kicked to the curb one day while being told we got a software to replace your mechanical work.
@@orange6562 Software Engineers are less safe than analysts, accountants and lawyers, keep gaslighting yourself
GPT will always have issues analyzing larger data sets and the understanding of the industry. Worse case scenario, I can see a demand for data analysts going down but an increase in business analysts.
Can explain why plz!!
I just started learning data analysis yesterday 😅. Now a machine will eventually take it away.
By the time most people learn about this study, the results will get outdated. The speed of progression for the generative AI outpaces the speed of information.
We can't keep up. 😢
The difference would seem to be in knowing what questions to ask. AI will outperform humans when given specific questions, so you will need to find yourself in the position of being the one to determine the specific questions. Otherwise your job will be at risk
Carry on folks..DAs arent gonna be replaced any time soon LoL...do ya thing .
"Spend more time touching grass", you say? That is very... utopian of you. ^^ Wouldn't bosses everywhere just pile up more work on us cause "Hey, you can do it X times faster now that you have ChatGPT to help you!" ?
Interesting sample size 😂.
I've been ordering in too much
Is there a data engineering version of this 🤣
I’m actually curious about this, as I haven’t found any. There needs to be more research on other data science disciplines as well; not just analytics
The end is here
*not. As a MS student in computer science, believe me I've asked ChatGPT to help on my homework. It sucks at even basic logic. It's great at doing generic tasks that it was trained on some blog post to do.
@@PorkBoy69No
Yes is bad for now ,but they are going to improve and then being cheaper we are gone
@@Zane_Zaminsky Didn't ask you
@@gicoss4170 Yes future iterations might be a lot better.
Currently, if you ask it to do anything vaguely innovative, it falls flat on its face. What seems like magic to many users is the same information you could've googled for. I've asked it questions before where it used the exact peculiar (non-technical) wording in some obscure reference I had open in another tab. I would have been impressed if I didn't happen to be reading the exact source it was trained to regurgitate back to me.
Current version is glorified information retrieval with some glimmers of emergent properties. GPT architecture is in fact a form of probabilistic information retrieval. The only intelligence can occur via emergent properties.
It has limited problem solving ability or creativity; if scaling increases those abilities rather than just the information available to retrieve than THAT would be a threat. We will see.
If current version or peers to it is a threat to you as an engineer, either A: the efficiency improvements given to workers allows them to use some modest fraction less workers or B: you are terrible. To be clear, I am concerned about this making labor worth less in general and am not just trying to say "git gud".
I'm data analyst. Hope this doesn't end me. Especially since I'm disabled.
Don't worry as they sell AI courses they advertise like that ,more so If you could learn Analysis by self learning using AI for it would be piece of cake for you .:).
I don't mean to be rude towards data analysts but doesn't your job was kinda repetitive in nature? Most of what I learnt in the master was using Pandas to generate charts and preprocess data with a few commands.
@lukeBarousse. I'm taking courses in Business Analytics and Data Science at a University. Should I continue taking these courses in your expert opinion? Seems rather ummm not so great at this point.
I'm still very optimistic about the field of Data Analytics; I think it'll involve the use of AI tools (like LLM agents) more in the future but will still ultimately need a human-in-the-loop to make sure things are going right.
@@LukeBarousse will there be a new roadmap? I've been studying for the last year to become a data analyst, but have been working full time just as an Aldi worker so I have no domain knowledge nor a degree. Extremely discouraging and daunting for someone like me to hear that GPT-4 is on PAR with the SERNIOR ANALYST and above the intern :/ . Feels extremely hopeless whenever I have my study sessions now because it feels like a waste of time. Very lost.
@@SomeStatus For over a year now, I've been hearing about AI replacing data analyst jobs, and it seems there's only been an increase in data analyst jobs.Don’t feel hopeless
"People need to understand how data works, are we allowing it to dictate what we do, how we do it, and why we do it? It's essential for individuals to verify all entered data in the final analysis. Personally, I can't fully trust AI."
in order to understand handle the information provided by gpt as data analyst, you have to be an actual data analyst, otherwise other people like me won't be able to handle the prompts and determine if the information is accurate.
The stakeholders talking to ChatGpt: Do Report.
There is a reason that both Stack Overflow and Stack Exchange don’t allow AI generated answers on their sites, lol.
Time to get into plumbing
Seriously that will secure your future.
@@userkm2 It won't, when most white-collar workers will flock to blue-collar jobs, it will drastically bring the pay down, by increasing competition. The right answers to this AI disruption is taxation of profits made by AI to allow Universal Basic Income (UBI).
@@aminebarnat6528 The only way to "secure your future" (not really) is to "make it" ASAP by investing into whatever is related to AI. Tech companies, GPU manufacturers etc. You have to get on board and stay there. The stock will keep rising. It's the last chance to leave the upcoming slave class. But even this might not be enough. The economy is gonna explode and if you're poor when that happens, you're gonna be poor forever. The ladder is being pulled.
We need a video on *Pyramid Analytics*
Bruh i just graduated. Im cooked
I'm facing a problem with gpt4 analyst.
Every time I ask it to do the analysis, it's failing to execute fully. Gets stuck in the middle saying error 😢
Any solution??
chatGPT makes loads of mistakes when it comes to computational applied mathematics!
Still not worried, GPT3.5 I provided a CSV file and asked basic questions such as can you calculate the mean, mode and median on a single column and it was abysmal. Even on a small data set.
Also this study sounds useless...
GPT 3.5 does not come with a code interpreter, which is used to convert natural language into code and convert the outputs back to natural language.
This is why GPT 3.5 model does not provide good outputs for arithmetic tasks since it cannot offload to like Python or SQL for calculations.
Just briefly looking at the study, the sample size is very small and not reliable, but LLMs are very capable of doing simple repetitive tasks such as data transformations, visualization, and even simple analysis. The "domain knowledge" by inferring data from the Internet is extremely disingenuous as it implies domain knowledge is the same as someone doing Google searches, which is hilarious, in my opinion.
The title of the article says "Is GPT-4 a Good Data Analyst?" (let's assume it's a topic, a general borderline covering several brief explanation around "Good Data Analyst") yet it sneak an unrelated cost-benefit analysis because of cheeky, unscientific assumption about "AI is considered as a good data analyst if we (as academics and researchers) are able to cut the corners" 😂
If the author(s) enrolls as an undergraduate in my uni. with this unrevised off-the topic part in hand, the board of professors wouldn't accept his/her research proposal in the compulsory bachelor thesis writing in our university 😮
Please sir.... I've been a data analyst for the past 14 years. I'm finally at 300K a year and I just want to purchase a 2 more multi-families so I can retire. Please don't make any more videos like this 😭😭
Hahaha!
I better stick to my company where a lot of specific industry knowledge is needed to actually be useful as a DA. If a put a chart together and all I tell people is "Our sales are higher on Wintern this year" I will be told "No shit Sherlock!"
If people don't have jobs as a result of AI taking over almost all jobs, people will not have money. If people don't have money, they will not spend on anything. Then what data will the companies analyse if nobody is buying their products?
AI improves productivity while data analysts can become more specific and creative
Plot Twist: The whole experiment was designed a written by ChatGPT 4
Bro i just enrolled myself into GOOGLE DATA ANALYTICS Cert with coursera based off on last video....
Luke should have asked me to wait 😭
Do it. It's pretty basic. After that, do the advanced one.
Do the Ibm
Just do it. I've been waiting for computing to end for 40 plus years now.
Yes, give your data to open ai… what a good idea
Real
Hahaha! GPT has not seen the back end of my company! Good fucking luck! So analysts will end up spending their entire workday explaining to GPT what each field in SQL actually means. I'm sure it performs well with cleaned data with field names like date_of_referral but it'll be completely lost when the fields are named cmod_temp_range_txt
I'd like to add that I'm not trying to crap on gpt, I love gpt and use it daily at work. I just don't see it taking over my position anytime soon. Eventually but not now.
What happens when people who specialize in prompt engineering and data cleaning (as opposed to general data science) start coming out of college? Totally different, and more common, skill set and talents compared to what is needed for general data science, and they will be much much cheaper.
YOOOMAN Data Analyst
AI is only at the dial-up stage, but it won't take 20 yrs for the majority of people to have access to broadband.
What I'm unclear about, is the level of experience the human, using ChatGPT in this study, had in the DA profession? Then in the real world outside of this study, If the user is inexperienced, how would they if the results are bias, hallucinations, or spot on?
They forgot data scientists
Idk...Ai scares me like anyone else. The idea that all the desirable jobs will be replaced by Ai leaving us with only back breaking hard labor jobs makes me wanna vommit.
Only thing that kinda brings me peace of mind is knowing that there should eventually be diminishing returns. By this i mean, the cost required to invest in the tech is too high where it doesnt make financial sense.
Example. We still have human cashiers. We still have truckers. We still have bank tellers. We still have tax people. We still have travel agents. We still have normal vacuums. List goes on and on.
There are many examples of where there are already viable robot alternatives....yet they're not all mainstream. Humans will be bias towards prefering humans. Sometimes i dont want the drama of people, but i think sometimes humans are just better at certain things.....i mean heck companies wont even invest in WFH half the time because they believe face-to-face is "needed".
Idk i think the idea that all of our problems are suddenly over and Ai is going to do all the innovation now doesnt seem right.
I think what we're really going to see if job requirements get dramatically higher AND entry level jobs basically disappear.
Posing the right question properly is the most important role of the analyst. These questions were already posed and rather stupid to be honest. Who gets paid 90k per year to be a mindless bar chart maker?
I’ve been fairly disappointed with GPT 4 as a data analyst. It doesn’t have domain knowledge and it is honestly not very creative in coming up with explanations. But I guess that’s why I’m senior.
That's great.
Where's the damn government?
its over bro
Mass unemployment and mass starvation in 10 years.
Intern is using GPT anyway 😂
So instead of having team of 10 people for the task soon team of 2 will do the job a little bit better. What to do with 8 people who were removed? Well we have a ww3 going on right now. Enlist in military is a good idea.
Domain knowledge is the last moat between you and AI taking over your job. I've never been so glad to be part of a secretive cabal that intentionally obfuscates and hoards knowledge. Our guild will survive
First!
🤣🙌
Second!
@@SeattleDataGuy Everybody is here now 🤣
@@LukeBarousse we are!
hehehe