chatGPT

[JonA]

The response reads like it was written by ChatGPT.

Prompt: explain why chatgpt would be a poor choice to replace patent lawyers

[dw]

Wait so was I right? The language is very similar to your post.

[JonA]

Wait so was I right? The language is very similar to your post.

Yeah. I'm trying to decide if I should be offended or flattered. Am I as smart as an AI? Possibly? Is that good or bad? I dunno.

[FredM]

Does it pass the turing test ?

Rofl. Please.

It's probably the greatest leap forward in tech and biggest disruptor to how knowledge workers get work done since Google -- and it has the potential to make Google itself completely irrelevant.

I asked it to write a python script that scrapes a provided url and prints the value of an html node that has a given name. This was 1/4 of a script I spent 30 min writing last week for work -- that particular section probably took me 10-20 min of coding + googling. The code it output not only worked -- it was commented, more concise and easier to follow than mine. It included helpful context for people who've never used python like how I'd probably need install BeautifulSoup with pip and how I might make small changes to fit my particular problem.

I also asked it why my MySQL restore was taking so long. This was, embarrassingly, 40 hours of work last week for me. It gave me some options about generating the backup file in an easier to restore way. I then informed it I didn't have access to the backup process so it gave me some more options to explore. The first option was actually the solution I dug for last week (again 40 hours completed in 2 min with the AI) except, once again, better. In all my stackoverflow digging I never found the command line option that accomplishes the same thing I had done with altering the actual backup file with regexes.

The irony is that knowledge workers have been excitedly exclaiming AI would be replacing "blue collar" jobs for years now, but the first fairly functional "AI" is actually coming for OUR jerbs. Not in the way youtube personalities are claiming -- this doesn't literally replace people. But absolutely, this is going to enable Principal engineers to be more productive than a principal engineer + a few offshore or jr. engineers, e.g. This is going to obsolete most "knowledge worker" roles that require someone above them to decompose the problem. Universities just got even more useless than they've been over the last decade.

I also find it entertaining how non-technical people think this is awesome because it's Skynet. This is awesome in the same way the iphone was awesome. Or Google was awesome. Take Python code writing. The data it's training off of absolutely does not include comments and well named variables. Python programmers are generally engineers or scientists with no concept of writing easy to maintain code. And yet the AI doesn't copy what it sees. Why? Because they had trained software engineers sit there and ask it questions and score the answers for a year. They did this across many domains. It might be one of the more advanced ML implementations available to the public but what makes it great is how well thought out it is as a complete product.

As much as I'm looking forward to being 4x as productive as I was last year, I'm most looking forward to the end of Google. I thought MSFT was only donating $1B for the licensing for GitHub copilot (also effing incredible) and Azure services, but I think they were actually taking out a competitor with that investment lol.

[51M0N]

I work in government and I can already see how this could have massive productivity benifits.

We spend a lot of resources answering correspondence from the stakeholders and the general public. Every one of those letters requires someone to open it, read it, figure out what the issues are, check if we already have words to respond to the issues (which has a search cost, and corporate knowledge retention is always a problem) and draft something new if we don't.

An AI that can scan all the letters ever written is going to know that we wrote a response to a similar letter 3 years ago etc.

That means that the job becomes checking the AI generated letter to make sure it is correct and maybe making small adjustments as needed (which would feed back in to the AI), which is already done during the clearance process. Then the main job becomes dealing only with new issues.

[dw]

Does it pass the turing test ?

Rofl. Please.

It's probably the greatest leap forward in tech and biggest disruptor to how knowledge workers get work done since Google -- and it has the potential to make Google itself completely irrelevant.

I asked it to write a python script that scrapes a provided url and prints the value of an html node that has a given name. This was 1/4 of a script I spent 30 min writing last week for work -- that particular section probably took me 10-20 min of coding + googling. The code it output not only worked -- it was commented, more concise and easier to follow than mine. It included helpful context for people who've never used python like how I'd probably need install BeautifulSoup with pip and how I might make small changes to fit my particular problem.

I also asked it why my MySQL restore was taking so long. This was, embarrassingly, 40 hours of work last week for me. It gave me some options about generating the backup file in an easier to restore way. I then informed it I didn't have access to the backup process so it gave me some more options to explore. The first option was actually the solution I dug for last week (again 40 hours completed in 2 min with the AI) except, once again, better. In all my stackoverflow digging I never found the command line option that accomplishes the same thing I had done with altering the actual backup file with regexes.

The irony is that knowledge workers have been excitedly exclaiming AI would be replacing "blue collar" jobs for years now, but the first fairly functional "AI" is actually coming for OUR jerbs. Not in the way youtube personalities are claiming -- this doesn't literally replace people. But absolutely, this is going to enable Principal engineers to be more productive than a principal engineer + a few offshore or jr. engineers, e.g. This is going to obsolete most "knowledge worker" roles that require someone above them to decompose the problem. Universities just got even more useless than they've been over the last decade.

I also find it entertaining how non-technical people think this is awesome because it's Skynet. This is awesome in the same way the iphone was awesome. Or Google was awesome. Take Python code writing. The data it's training off of absolutely does not include comments and well named variables. Python programmers are generally engineers or scientists with no concept of writing easy to maintain code. And yet the AI doesn't copy what it sees. Why? Because they had trained software engineers sit there and ask it questions and score the answers for a year. They did this across many domains. It might be one of the more advanced ML implementations available to the public but what makes it great is how well thought out it is as a complete product.

As much as I'm looking forward to being 4x as productive as I was last year, I'm most looking forward to the end of Google. I thought MSFT was only donating $1B for the licensing for GitHub copilot (also effing incredible) and Azure services, but I think they were actually taking out a competitor with that investment lol.

Some very interesting points here, particularly the bolded. That is the first time I've seen this prediction, but it seems very plausible as soon as you consider it.

[Culican]

The irony is that knowledge workers have been excitedly exclaiming AI would be replacing "blue collar" jobs for years now, but the first fairly functional "AI" is actually coming for OUR jerbs. Not in the way youtube personalities are claiming -- this doesn't literally replace people. But absolutely, this is going to enable Principal engineers to be more productive than a principal engineer + a few offshore or jr. engineers, e.g. This is going to obsolete most "knowledge worker" roles that require someone above them to decompose the problem. Universities just got even more useless than they've been over the last decade.

[51M0N]

Kind of makes sense that AI will have a bigger impact on knowledge workers over physical workers. Computers already crush humans at games like chess, but I don't think there are any physical games (that require a degree of hand eye coordination) that computers win out at?

Might be approaching it in powerlifting, but even then I think getting a robot to walk onto the platform and get in position to lift the bar at an unknown venue is going to be a significant challenge.

[Manveer]

I asked it to write a little bash utility script yesterday. It did a great job, and much faster than I would've done it. Only a 12-line script, but still. Super cool. You just have to know what to ask it to do / what changes to ask it to make for your application.

[asdf]

All you guys gave OpenAI your phone number?

[Manveer]

All you guys gave OpenAI your phone number?

Yeah. Why not? I have to give it to every online store, reservation site, utility provider, streaming service…

You could create a new number with Google Voice (or similar service) if you want.

[FredM]

All you guys gave OpenAI your phone number?

lol. What do you think they're doing with it?

Thanks to needing to give my phone number out to basically every ecommerce site, I get about 10 spam calls a day. It's always obvious which ones sell my info because they do so immediately and I notice the uptick in spam calls within that day or the next.

ChatGPT doesn't seem to have sold my phone number to advertisers which is the only thing they could really do with it to "harm" me.

[asdf]

Yeah. Why not? I have to give it to every online store, reservation site, utility provider, streaming service…

You could create a new number with Google Voice (or similar service) if you want.

OpenAI doesn't accept Google Voice numbers. I tried. That gave me additional pause.

It's true that I give my number to all sorts of companies. But with them I'm not creating a history of chat interactions.

[asdf]

lol. What do you think they're doing with it?

They're an AI company. They're doing as much with it as they possibly can. Which is likely way more than I could ever imagine.

[FredM]

lol. What do you think they're doing with it?

They're an AI company. They're doing as much with it as they possibly can. Which is likely way more than I could ever imagine.

They're using it to verify you're a human because it costs them, probably, millions of dollars a day in cloud costs to deliver the experience they're delivering today at scale.

But sure. Definitely don't ask an AI how to plan an insurrection. I'm ok with Skynet knowing I don't know SQL as well as other people think I do.

[asdf]

I'm ok with Skynet knowing I don't know SQL as well as other people think I do.

Fair point. Really, I was just curious if there was a workaround.

[omaniphil]

All you guys gave OpenAI your phone number?

I just used my google account to login. Is that worse? Maybe. I don't care.

[5hout]

The irony is that knowledge workers have been excitedly exclaiming AI would be replacing "blue collar" jobs for years now, but the first fairly functional "AI" is actually coming for OUR jerbs. [/B]Not in the way youtube personalities are claiming -- this doesn't literally replace people. But absolutely, this is going to enable Principal engineers to be more productive than a principal engineer + a few offshore or jr. engineers, e.g. This is going to obsolete most "knowledge worker" roles that require someone above them to decompose the problem. Universities just got even more useless than they've been over the last decade.

Some very interesting points here, particularly the bolded. That is the first time I've seen this prediction, but it seems very plausible as soon as you consider it.

One of my main sources of income is working for a company doing document review for litigation. Company gets sued, ordered to produce all emails related to X topic that aren't privileged. Everyone involved in the direct review must be a licensed attorney. One of our most consistent warnings we get the first pass team is "We must be better than Assisted Review [i.e. ML applications]. Doc review projects are getting smaller and smaller as more work is being done automatically. Be a value add, because clients are getting pickier and pickier about why we are suggesting teams of people over Assisted Review."

It's brutal. Projects that 8 years ago would have 1M+ emails/docs (hereinafter docs) to review are now 100k docs. Better technology/greater court comfort with technology means less docs are selected for review. Back in the day a search time might have been something like "all emails with John Q Douchebags first, middle or last name in them and at least one of topic 1, 2, 3, 4, 5, 6 or 7 somewhere in document or the attachments. Now it's stuff like (Jon(with all endings) OR John OR (all alternative names for John) OR (same for middle name) OR (Same for last name)) within 50 spaces of one of (1, 2, 3 or 4) or within 30 spaces of (5, 6 or 7). Even with this decrease often the client wants to use Assisted Review so you only look at ~30k docs. The ML tech used (so far) is fucking terrible, counterproductive and irresponsibly shit and is already acting as an incredible force multiplier.

More and more firms have in-house teams that can handle their large projects or reach out to only for our Assisted Review chucklefucks to help or want a small team of actually good professionals to help them. It's a fucking massacre. I started in 2014 (ish) part time, and then made the jump to more-or-less full time (with moonlighting) a few years before Covid. These are pure knowledge economy jobs that are disappearing.

If your job doesn't involve being an actual value add, beyond doing annoying to automate tasks being done by someone with a college degree for historical reasons you need to upskill or prep your butthole. Good luck shitjournalists writing up routine stories. Already a lot of routine sports news is near-automated (from what I've read some of it is AI produced, and a custom chatGPT implementation could replace probably 80% of game report writing, especially if you live feed it twitter/social media data during the game.

@omaniphil already hit on how this will wreck patent agent/attorney work. Way too much of a patent application, once you've got claims nailed down, is completely routine. Same as above, a custom implementation. Little bit of extra training and you can probably fire a pile of attorneys and replace a few of them with a patent illustrator. Keep the best attorneys/agents around to help with claims/dealing with PTO, but the people sitting around with the drawings and claims in front of them mechanically writing descriptions of each pictured element and how it relates to the other pictured elements? GTFO.

I also think this is going to help patent prosecution a lot. Feed chatGPT all office actions and replies to office actions. Then improve the +/- ML coding by tracking the OA/Replies through the Patent Trial and Appeal Board. Use this to suggest/create form paragraphs for known to be winning responses to specific rejections and specific examiners (already you can get specific examiner data, which has been very helpful on occasion). Patent Examiners often are robotic/treat it like a game of Magic: the Gathering where they'll play a certain rejection to invite a specific response. Figuring out what that examiner wants is time consuming, but since it basically requires nothing but boring reading and regurgitation with minor changes... chatGPT hiho.

All that said, a lot of chatGPT answers look good until you really read them and I'd tune down the overall excitement from "this is big" to "this is the pebble showing people to upskill or get fucked in 5-10 years".

[FredM]

All that said, a lot of chatGPT answers look good until you really read them and I'd tune down the overall excitement from "this is big" to "this is the pebble showing people to upskill or get fucked in 5-10 years".

Good rant/agreed with most of it.

Strongly disagree with this. ChatGPT, in it's current state, is a bigger jump from solving problems without the internet to solving problems with the internet. I've been using it daily to automate, so far, literally weeks of work away (in just 3 days). So have a bunch of other people I talk to. Not just in software dev. My CEO just used it to write a blog post on an important business trend we're seeing in our industry. I asked it to write up the user stories (and then asked it to estimate them) for a very widely defined software problem and it did a better job than 80% of software architects would have because it included tickets with reasonable estimates for the gotchas most of them would have missed in this instance. I'm not sure "Product Manager" is a realistic job anymore with this thing AS IS.

It's not 5-10 years away lol. It's now. Well -- it's a year or two away because that's how long this will take to really be adopted at the scale required.

This is absolutely at a level most Data Scientists and myself (ML engineer) didn't think was possible for another 5 years. 10 years??? They might actually complete their mission in 10 years at this rate (an actual AI). Even if I don't actually think that's possible to the level non-technical people seem to think -- I'm not sure most people will be able to tell the difference when Open AI releases the product they're actually iterating toward next decade.

[omaniphil]

@omaniphil already hit on how this will wreck patent agent/attorney work. Way too much of a patent application, once you've got claims nailed down, is completely routine. Same as above, a custom implementation. Little bit of extra training and you can probably fire a pile of attorneys and replace a few of them with a patent illustrator. Keep the best attorneys/agents around to help with claims/dealing with PTO, but the people sitting around with the drawings and claims in front of them mechanically writing descriptions of each pictured element and how it relates to the other pictured elements? GTFO.

I was inhouse, but left to go back to a firm a year ago to just do Prep/Pros due to the better lifestyle and money. This week I've started to regret that decision due to ChatGPT. Time to go learn how to be a welder or plumber.