I really like this part: "AI is a tool. It is not always the right tool. Consider carefully whether, given its weaknesses, it is right for the purpose to which you are planning to apply it."
On the video front, have you tried Descript @Ethan? I've delved into this recently and find it's AI features very helpful in my daily flow.
Ross, I'd love more on this too. I find Descript so helpful. What I loved most was it's interactive onboarding explainer video. I've never seen a company do that before and I was elated someone thought of it
This is brilliant. Thank you for the cogent utility! As you note: “Documentation-by-rumor is a weird choice for organizations claiming to be concerned about proper use of their technologies, but here we are.” I hope Altman, et al remember to send you a check.
Thanks Ethan for a most informative summary of the current state of play (& links to learn more).
The image at the end reminded me of Vannevar Bush's Memex, envisioned in 1945. Who knew it would take so long to come this far, and now we're here, the blistering pace of change makes it a challenge to keep up at times! https://history-computer.com/memex-guide/
I'm a writer and have been playing around with Writer.ai & Jasper and other tools. I write about it a lot and that's prompting (see what I did there?) a lot of my colleagues in the creative field to reach out to discuss this. Of course they are tracking with these changes and having various reactions as you can imagine.
A lot are centered around scraping. Even if Firefly is using licensed work, I've heard that those artists didn't agree to AI. It's not stated in their contracts. If you don't follow Jon Lam, he might be worth a look. He's very vocal.
I know this is a huge topic and not one you may want to enter but it would be nice (for me anyway) to hear your perspective.
As of today, I'm unsubscribing to several AI related newsletters b/c of their fear-mongering headlines (often paired with stupid tone-deaf emojis) b/c tech brahs and wanna-bes seem to forget about the anxiety this new technology is causing.
But not yours. So thanks again for the measured, thoughtful work you do. :)
If I may recommend another product - aimed at coders:
Fine transforms dev tasks into tailor-made specifications and uses AI agents to implement them. With it, devs can focus on solving problems and shipping higher quality software instead of solving bugs and manual work. It's free!
This was great post. I've just started to use ChatGPT to help with writing and w.o.w. do I feel guilty about it. But I would be lying if it didn't make my writing cleaner and clearer and more what I want it to be.
I wish it made sense for me to employ a real person to do this. But often times, I'm writing or re-writing last minute, and have a (self-appointed) deadline to hit. And I'm not making any money from this endeavor ... (yet - I'm hoping I can make that change). Having a pocket editor and someone to lob ideas and get something bad has been a huge help.
But yeah, still feel very conflicted about using it.
For companies wishing to take advantage of the above but looking to protect their data, do you have any opinions on what's available or up and coming? e.g. Emad at Stability AI has been talking about develping bespoke versions for countries, vertical markets, companies with data security, but do you know of other such initiatives?
"If you are going to use AI as a search engine, probably don’t do that."
Absolutely. I have, however, had good results with the AI-assisted search engine Perplexity.ai. Ask a questions, get back a brief summary of answers, with (here's the important part) links to references on the web. Thus you can check Perplexity's work.
I find it very useful for resolving questions of how to do X in software package Y, or how to resolve hardware or software errors.
I think that's right, that Code Interpreter will completely change the way data is studied. I mean, instead of becoming good at manipulating data in a spreadsheet, you can just talk so "someone" about your data. "Draw some charts for me that would be useful to my business" is such a ridiculous magic spell if you're a business owner, or if you're at all in the realm of data analysis.
I always worry about clarification, and asking the right questions. Sure, CI will give you a response that's correct *enough* but does it actually answer the question you have? I'm not a data scientist, and maybe this is what the quote about changing how data scientists are trained is alluding too. But that still feels like a big gap to me having worked in organizations that only *kind of* knew what they wanted--despite how self-assured they were in asking for it :D
Andy, we have a lot of "learning curve" pain ahead of us!
I am confident that data scientists are the first group to push back on this (beyond confident, in fact-- I have plenty of first hand info!), but I'm equally confident that their concerns will be addressed, one by one, until it makes absolutely no sense to process data any other way.
We have a really, really big data problem right now. We have much, much more data than we can ever process without leveraging AI. We have a choice: never, ever process that data, or start leveraging AI to figure out what we can about it.
It's going to be ugly in the interim, probably at least the next decade, but we're going to move past this, and far past it, I think.
I don't disagree there - AI is absolutely the future and I'm already expecting historians in the future to take about the AI revolutions the same way they do about the industrial revolution. We had a lot of the same concerns then about the industrial revolution as we do about the AI revolution, and we turned out OK there :-)
I think there's a bit of a difference in the scope and rapidity of impact that we didn't see in the industrial revolution that will present us with new challenges as a species. I know you didn't type it out as "Learning curve of pain" but that's how I first read it, and I kinda liked hah -- I think you are right we have that not just with data scientists, but everyone.
I think a lot of disagreements about AI supremacy and the like center around misunderstandings about time lines. EG, "do you think AI will replace most human jobs?" can be a tricky question.
Eventually? Of course, probably all or almost all.
Within six months? Nah.
AGI is going to be a thing. I don't know if it'll be a thing in the next five years, or the next fifty. I'd guess probably sooner than the next 500 years, but that really feels like speculation, since we've never done this before. Not exactly this, anyway.
Thanks for this Ethan! Where you talk about the use of AI with code, maybe point to GitHub copilot? To me the implementation of this software inside the IDE seems a lot more elegant and it will likely catch up in power soon (as far as I know GitHub is owned by Microsoft as well, could hence benefit from OpenAi advancements)
I've been running parallel tests on Bard & Bing chat for image context recognition and surprisingly am finding better results on Bard. Have you experienced this? I'm wondering if GPT4's abilities have declined even for image recognition as few articles are suggesting.
I really like this part: "AI is a tool. It is not always the right tool. Consider carefully whether, given its weaknesses, it is right for the purpose to which you are planning to apply it."
On the video front, have you tried Descript @Ethan? I've delved into this recently and find it's AI features very helpful in my daily flow.
The key sentence is the one you quote!
Ross, I'd love more on this too. I find Descript so helpful. What I loved most was it's interactive onboarding explainer video. I've never seen a company do that before and I was elated someone thought of it
Same here. I continue to be impressed by its features and the ease for me to produce video. It must half halved the production time on many projects.
I wish I had more work to do with it. I go months between projects and forget everything! Ha. But it's easy to find a tutorial and get back at it.
This is brilliant. Thank you for the cogent utility! As you note: “Documentation-by-rumor is a weird choice for organizations claiming to be concerned about proper use of their technologies, but here we are.” I hope Altman, et al remember to send you a check.
Thanks Ethan for a most informative summary of the current state of play (& links to learn more).
The image at the end reminded me of Vannevar Bush's Memex, envisioned in 1945. Who knew it would take so long to come this far, and now we're here, the blistering pace of change makes it a challenge to keep up at times! https://history-computer.com/memex-guide/
Ethan, I'm a huge fan. Thank you for what you do.
I'm a writer and have been playing around with Writer.ai & Jasper and other tools. I write about it a lot and that's prompting (see what I did there?) a lot of my colleagues in the creative field to reach out to discuss this. Of course they are tracking with these changes and having various reactions as you can imagine.
A lot are centered around scraping. Even if Firefly is using licensed work, I've heard that those artists didn't agree to AI. It's not stated in their contracts. If you don't follow Jon Lam, he might be worth a look. He's very vocal.
I know this is a huge topic and not one you may want to enter but it would be nice (for me anyway) to hear your perspective.
As of today, I'm unsubscribing to several AI related newsletters b/c of their fear-mongering headlines (often paired with stupid tone-deaf emojis) b/c tech brahs and wanna-bes seem to forget about the anxiety this new technology is causing.
But not yours. So thanks again for the measured, thoughtful work you do. :)
If I may recommend another product - aimed at coders:
Fine transforms dev tasks into tailor-made specifications and uses AI agents to implement them. With it, devs can focus on solving problems and shipping higher quality software instead of solving bugs and manual work. It's free!
https://www.fine.dev/
This was great post. I've just started to use ChatGPT to help with writing and w.o.w. do I feel guilty about it. But I would be lying if it didn't make my writing cleaner and clearer and more what I want it to be.
I wish it made sense for me to employ a real person to do this. But often times, I'm writing or re-writing last minute, and have a (self-appointed) deadline to hit. And I'm not making any money from this endeavor ... (yet - I'm hoping I can make that change). Having a pocket editor and someone to lob ideas and get something bad has been a huge help.
But yeah, still feel very conflicted about using it.
Thanks Ethan,
For companies wishing to take advantage of the above but looking to protect their data, do you have any opinions on what's available or up and coming? e.g. Emad at Stability AI has been talking about develping bespoke versions for countries, vertical markets, companies with data security, but do you know of other such initiatives?
"If you are going to use AI as a search engine, probably don’t do that."
Absolutely. I have, however, had good results with the AI-assisted search engine Perplexity.ai. Ask a questions, get back a brief summary of answers, with (here's the important part) links to references on the web. Thus you can check Perplexity's work.
I find it very useful for resolving questions of how to do X in software package Y, or how to resolve hardware or software errors.
https://perplexity.ai
I think that's right, that Code Interpreter will completely change the way data is studied. I mean, instead of becoming good at manipulating data in a spreadsheet, you can just talk so "someone" about your data. "Draw some charts for me that would be useful to my business" is such a ridiculous magic spell if you're a business owner, or if you're at all in the realm of data analysis.
I always worry about clarification, and asking the right questions. Sure, CI will give you a response that's correct *enough* but does it actually answer the question you have? I'm not a data scientist, and maybe this is what the quote about changing how data scientists are trained is alluding too. But that still feels like a big gap to me having worked in organizations that only *kind of* knew what they wanted--despite how self-assured they were in asking for it :D
Andy, we have a lot of "learning curve" pain ahead of us!
I am confident that data scientists are the first group to push back on this (beyond confident, in fact-- I have plenty of first hand info!), but I'm equally confident that their concerns will be addressed, one by one, until it makes absolutely no sense to process data any other way.
We have a really, really big data problem right now. We have much, much more data than we can ever process without leveraging AI. We have a choice: never, ever process that data, or start leveraging AI to figure out what we can about it.
It's going to be ugly in the interim, probably at least the next decade, but we're going to move past this, and far past it, I think.
I don't disagree there - AI is absolutely the future and I'm already expecting historians in the future to take about the AI revolutions the same way they do about the industrial revolution. We had a lot of the same concerns then about the industrial revolution as we do about the AI revolution, and we turned out OK there :-)
I think there's a bit of a difference in the scope and rapidity of impact that we didn't see in the industrial revolution that will present us with new challenges as a species. I know you didn't type it out as "Learning curve of pain" but that's how I first read it, and I kinda liked hah -- I think you are right we have that not just with data scientists, but everyone.
We're on the same page for sure.
I think a lot of disagreements about AI supremacy and the like center around misunderstandings about time lines. EG, "do you think AI will replace most human jobs?" can be a tricky question.
Eventually? Of course, probably all or almost all.
Within six months? Nah.
AGI is going to be a thing. I don't know if it'll be a thing in the next five years, or the next fifty. I'd guess probably sooner than the next 500 years, but that really feels like speculation, since we've never done this before. Not exactly this, anyway.
I appreciate the dialogue!
"Some things to worry about: Deep fakes are a huge concern, and these systems need to be used ethically."
Aren't deepfakes an entirely different thing compared to AI generated videos?
I'd like to share a tool I built, which makes you use ChatGPT, Bing, Bard, Claude in one place: https://chathub.gg
Thanks for this Ethan! Where you talk about the use of AI with code, maybe point to GitHub copilot? To me the implementation of this software inside the IDE seems a lot more elegant and it will likely catch up in power soon (as far as I know GitHub is owned by Microsoft as well, could hence benefit from OpenAi advancements)
Thank you, Ethan! This is super helpful! One piece of information from a German guy: Claude is only available in the USA and UK.
Comprehensive.
Can you give me some real-world examples of the current state of AI adoption in US schools and universities?
I've been running parallel tests on Bard & Bing chat for image context recognition and surprisingly am finding better results on Bard. Have you experienced this? I'm wondering if GPT4's abilities have declined even for image recognition as few articles are suggesting.