Irreplaceable Research Skills in an AI Era

[@rehankhalid6161]

Being a researcher is not just writing and presenting. There is alot of experimentation and variables involved which definitely requires a human with a right understanding.

[@henkroodt9393]

Thanks! One of your best and important opinion pieces.

[@DrAndyStapleton]

Thank you for the super thanks!

[@tweedrhino]

Thanks!

[@vtheb1299]

Serendipity. If you read the history of science, so much of discovery comes from serendipity. It's a wonderful thing that allows huge leaps of thought. A simple example, penicillin was discovered when people went on vacation and the petri dish was contaminated with a mold. BUT as someone pointed out, being a scientist is about seeing what's important in that dish: the fact that the bacteria were killed by the mold! And not just tossing it because ot didn't conform to the pre designed script. No AI will replace that.

[@vtheb1299]

@Bailiol read what I wrote beyond the first word. It takes a human to understand which ones of your 10 trillion simulations' outcomes are meaningful and important. AI is cool, we're not against it, just staying real about its limitations. Just as we are about our limitations.

[@nosaosawe3158]

​@@Bailiol you think ai would be able to prioritize attention in ways that human do?😂

[@AndriyG0]

I am not sure about that :) A few years ago, we thought that the Tuning test could tell us something, and machines wouldn't be able to replace people in creative professions...

[@richardauzier8979]

AI still cannot replace people in creative professions, it is not even creative. I suggest you read "AI Snake Oil" and read about what creativity is

[@hanskrakaur9830]

@@richardauzier8979 That was true when "AI Snake Oil" was being written. Now you can write it with AI, in the same tone and style as the author.

[@henkroodt9393]

@@hanskrakaur9830 and AI can spontaneously come up with that concept linking AI and snake oil? I think not.

[@katjarozantseva8069]

Trying to make masters in a new country during quarantine broke me so hard that the point about human connection made me weep

[@MarkJacksonParkES]

Ultimately, to assume limits upon AI is to assume limits not upon machine but human intelligence.

[@johnbarryyallagher1128]

asking the right and new questions that can lead to developing new and impactful areas of research

[@henkroodt9393]

Regardless of where the future takes us, this is one of your best and important videos. Good job, great arguments!

[@organisedme]

it will never collect data and again even if it summarize the paper its not the same as reading the research paper?

[@dtyn]

Yup, my dissertation idea came to me while I was half asleep 😂

[@asabaal]

I love your videos! Wish I had this while I was in grad school!

[@NafisehKanani-ht3hm]

Great , thanks

[@ilovedrake-w3p]

bro your video was shows in the israeli news congratulations

[@yana.mp4]

Not sure if i agree with Andy OR the commentors completely! I found chatgpt to be useful for giving me some basic ideas and explaining concepts but when i prompt it for something more novel, or to explain a more complex idea it seems to not perform as well. I think it may one day outperform us but it will definitely need matter experts to construct the right prompts and interpret and proof the output

[@hanskrakaur9830]

Wait a little bit.

[@mohammedelbouzidi7005]

I am really impressed by your videos. Thank you.

[@Zahidbwp]

Nice informative video.

[@GenAIBlackBelt-qs6uh]

You have my favorite accent. I can't wait to see your AI clone on a stage one day, with the same unique energy.

[@GenAIBlackBelt-qs6uh]

It will need to wear the same shirt and beard.

[@hanskrakaur9830]

Well... I have a masters in computer science. I feel the situation is very, very complex. But, just for this, let´s try to keep it simple: Gpt-01 may not be able to beat some areas of "top" researchers, but its currently beating most areas of many "bottom" researchers; not everybody is "brilliant" in academia, and a most graduate students struggle. What I see is that the amount of "fields" and "abilities" of AI is growing rapidly, and sometimes very unexpectedly. Some departments are not hiring as many people as before. The way ahead that I see is to become a really, really good prompt engineer: is the new alphabet.

[@charliepan4055]

How kind to bring this so mild. AI will beat us all over the place.

[@happinys189]

I've watched your videos for a while now, since before ChatGPT took the world by storm. What once kept me coming back to your channel over and over again in search for a greater understanding of the nitty-gritty scientific business, the meat and potatoes beyond pop-science videos/articles and what it truly meant to engage within the profession of academia, is now a morbid curiosity. AI is simultaneously miraculous in all things research related and then some, yet here, not enough. Particuraily, I find this video's highlighted "human traits" interesting. Did *you* assess the tools you've built your platform upon lacked in this specific way? Or did the tools you built your platform on generate these apparent pitfalls? A pointed question, I know. But if these "future proof-skills" are about "survival," and also given the flux state of AI and its ever changing nature, to what degree of confidence can you ensure this video will be relevant in the future? Will it be relevant in a year? Two? A month? I don't expect a response. I respect the hustle and all, but I wonder, just how human is your content? Or is it just that? Content. Nonetheless, seeing "human" in the title caught my interest. Cheers!

[@emmanuelameyaw9735]

Passion...AI can't replace that.

[@iAPX432]

Try prompting for it, you would be really surprised!

[@lorenzoleongutierrez7927]

Good points 👍

[@EssentiallyAI]

The "sixth sense" could be derived from emergent behaviors. Don't underestimate this. Think of game 2, move 37. (You know what I'm talking about.) I'd avoid saying, "Never, ever". I'd never, ever say "Never, ever"! 😁

[@gayatricasey6618]

I have a query. If the supervisors asks us are you using AI.. What should be our formal reply ?

[@hanskrakaur9830]

Copy and paste, without critical thinking, is seen as a bad practice. Using it as a tool, on the other hand... is quite different. Finding articles, errors on your statements, debate, brainstorm, code, test, review, creative solutions, data analytics, and even "critique" on your own work, is seen as a extension of your abilities. If you are a student, be sure to develop the skills you are paying for. That being said... three months ago, Gpt-4 was very bad on high-school arithmetic... try o1 on textbook physics problems now.

[@klvfu]

Truth!!!

[@kumardigvijaymishra5945]

Great vid. I certainly have thought about AI's impact on research. Also there are surely counter to each of your carefully crafted AI cons. 1. AI doesn't need human connection because it's built on thousands of millions of human researchers worldwide. 2. Sixth sense or intuition is also the cause of several expensive scientific failures. Perhaps it's good AI doesn't have human intuition. And again that depends on how we define intuition. Intuition is just deep research. There are AI models based on deep research. 3.... and so maybe AI is omniscient.

[@나는야멋쟁이-n4r]

These kinds of videos help me so much as a master’s student who dreams about becoming a researcher.

[@bladekiller2766]

Why do you think AI will never acquire the intuition that you are talking about?

[@EladIdan]

AI is the most broad and useless term in computing. Sounds like he's talking specifically about LLMs, which are not likely to become creative. But the fact that he just uses "AI will never" shows he doesn't actually understand the field, or if he does, doesn't care to spend the effort to accurately describe the problem. This seems like just a reactionary rant.

[@hanskrakaur9830]

That the is the mind bending challenge: To accept that expert_systems+neural networks+unsupervised_learning > human_intuition(most_economically_tasks) is a very hard pill to swallow for us, but very easy code to compile for it.

[@bluegamer75100O]

“hey vsauce”

[@tar-yy3ub]

Didnt find this convincing at all is thos a joke? Usually when Ihear someone say "AI can't replace the human element" it's signals they either aren't keeping up with latest developments or haven't thought deeply about how to use these tools

[@hanskrakaur9830]

True. Andy has a lot of experience in AI, and use it as a prompt engineer.... but I bet he is not in the AI development field. I had many teachers that share Andy´s ideas: it was a know "mantra" before neural networks and transformers.

[@johnbarryyallagher1128]

Identifying the unknown unknowns by making connections across what seems different disciplines that a researcher has exposed themselves in the process of actually living, lol

[@UddipanDasgupta]

I believe most of the points raised in this video would be surpassed by AI in coming years

[@BibhatsuKuiri]

deep down we all know what ai is capable off :)

[@SalomeKapata]

Hooray.

[@DefenderX]

I think the tear down criticism is damaging to humans just like any other toxic work environment. If you're "comfortable" with it, think it's correct or even good you're most likely been conditioned over many years of being broken down so many times that you're desensitized. But compared to an AI, where trust is still being built, AI will give you the same positive criticism every time (unless you tell it otherwise), and that can be perceived as disingenuous. Still, you shouldn't have a toxic environment and exclude people who need a good work environment to work. That's excluding a lot of good researchers. You can still do good research without a toxic academic environment.

[@Olaf_Schwandt]

super

[@DanielOlsson-wr6nl]

I think it could be informative sometimes. It may be interesting, but It’s more about a lot of people around. I have questions about Julius, I don’t know whom to ask. - default or chat gpt 4? - it seems to be limited to 250 queries. (Basic). I believe I will need more? Alternative?

[@hanskrakaur9830]

The 20 bucks pro version is enough for me, but my teachers are very "easy going". The limit can be managed working by modules or objectives and starting a new task: Keep a hard copy of the advancements and give it as a starting point each time. Paste Images and tables as jpg, instead of giving it as text. Use gpt-4 for easy task, writing, brainstorm, article search and fact check with internet, and 01 to fact check results, conclusions, logic, unseen fallacies, error detection, statistical analysis, and overall consistency. Give it the context, problem and tasks (stepts), and then, ask it to generate its own prompt, and see if it understood the idea.

[@charliepan4055]

This isnt going to age well....

[@iAPX432]

All wrong.