Join Lee and Gwilym in the last of the Pubcast special episodes as they chat with Ed Bray from DeepRender about how AI is reshaping the patent world. Ed explains his professional journey from solicitor to inhouse patent attorney at a tech startup and how his experiences helped shape his current role. There is also a discussion on the current limitations of AI and and the broader challenges within the patent system.
Join Lee and Gwilym in the last of the Pubcast special episodes as they chat with Ed Bray from DeepRender about how AI is reshaping the patent world. Ed explains his professional journey from solicitor to inhouse patent attorney at a tech startup and how his experiences helped shape his current role. There is also a discussion on the current limitations of AI and and the broader challenges within the patent system.
It's our last one today, is it? It's been a great day, hasn't it I?
Speaker 2:think if we do less than four podcasts in a day now, we get under-stimulated. Yeah, but they are quite wearying in time, Ian, what did you just drop? Honestly the audience.
Speaker 1:Yeah, we should stop having an audience shouldn't we?
Speaker 2:No, I love it. They've been asking great questions. I hope have you got your questions lined up for this one, everybody, before it starts so, um, yeah, you've enjoyed today.
Speaker 1:It's been. That was good, aren't they? Yes, but I'm conscious that we might not put them out in the order they've been recorded, so I've said this is the last one, but this could be the first one when it goes out.
Speaker 2:It could be the first one. The football shirt one is worth if it's before the football shirt one. There's a really good football shirt one coming and it's after. Wasn't that a good football shirt one?
Speaker 1:You're almost a professional at this, aren't you?
Speaker 2:Lee Davis and Gwilym Roberts are the two IPs in a pod and you are listening to a podcast on intellectual property brought to you by the Chartered Institute of Patent Determination.
Speaker 1:So shall we get the guest? Yes, so it's Ed. It's Ed Bright. Yeah from, I've got to go in now Deep, brenda Deep. Brenda. So I got quite excited, ed, when I was looking at the line-up today, because I thought, wow, we've got concrete on at number two podcast and then we're going to be talking plastering. But it's not plastering, it is not. I'm afraid.
Speaker 3:No, it is slightly more. Well. I mean, plastering is probably very technical, but this is sort of machine, computer technical. Yeah, yeah, yeah.
Speaker 2:It's not floating stuff on a wall.
Speaker 1:Correct. So tell us about you kind of part of your history of you as a patent attorney, and then about kind of where you are and what you do.
Speaker 3:Yes, so I have an interesting background in that I did a physics degree like most people, but I first trained as a solicitor, and so that was the two years of law school and then two years of training contract. But I found that that was kind of too far away from the technology. So then I jumped ship to the patent attorney side and was at Marks and Clark for about eight years just over eight years and then went in-house at a little small startup about a year ago called Deep Render, and at the same time I also on the side build patent attorney tools, legal tech tools and stuff.
Speaker 1:So it's yeah, I wear my hats. We're talking about all of it, aren't we? Can I go with a couple of questions first, gwilym, is that?
Speaker 3:okay, it's it Go on.
Speaker 1:From what you've just said there. So first is about the solicitor training and then moving across to kind of patent attorney training.
Speaker 3:Did you find that? That made you kind some of these things? But the day job is very different to just understanding what's going on and so to that extent it didn't help. But there are other things, like do handling assignments, recordals, all that type of stuff. Where you get all of that stuff? So it probably helped a little bit but the jobs are very different, so it probably helped a little bit, but the jobs are very different.
Speaker 1:And so second question it's quite an interesting kind of career move to go sort of private practice to small niche in-house yes Techie, start up in a kind of area that's going to grow massively.
Speaker 3:How did you find that move? Yeah, so, to be entirely honest, I was bored with private practice, which is why I was looking for interesting and slightly unusual jobs.
Speaker 2:I find that inexplicable, yeah.
Speaker 3:And so then this job came up amusingly. So they actually put the job up on the SIPA job page and, just by total coincidence, I happened to look at it and they were worried that they weren't going to find anyone for like three to six months, and literally that same day, um. So I applied and then I sort of interviewed and they had it all sorted in like a few weeks. So, wow, um, yeah, but it's been great. Um, and maybe not unusually for a startup, but unusually for private practice, I am among the oldest employees and I'm like 35, so the vast majority of us are kind of in the 20s, 30s, um.
Speaker 3:I'd like to say basically everyone is a massive geek, massively nerdy. It's all maths, phds, um, and there's a lot of crazy, cool stuff going on I've also pictures on a website tell a story.
Speaker 1:I think you know yes and um. So that's quite an interesting move in it to go to go from kind of private practice to in-house what's the how, how different have you two questions? Yeah, how. How distinct or different have you found the, the two roles, and do you find yourself being able to exercise your geeky side more in the in-house role?
Speaker 3:uh. So second question yes, absolutely Like it is. So law firms are everyone's really nice, it's really fantastic, but there is an element of you asked to work in a law firm.
Speaker 3:When you then jump to some startup environments, it feels very much like a startup environment and it feels a lot more like bubbly and excitement and all that kind of stuff which just like you know, as much excitement as there is in in law firms to a sense is there's a lot more of it in in startups, um, and I guess the other part is from the just the day job.
Speaker 3:Like in private practice. You kind of you are trying to give the gold standard of service to your client and so there's all these other considerations that constantly come in in the startup side. You really focus on the commercial side of things and you know if your goal is to do something and you can get it done, um, in a way that might you might not do in private practice, then you do that because that is the best for the business, that's the best commercial strategy to make and it's this kind of you're focusing a lot more on the end goal of what patents are for, rather than trying to create a perfect. You know to do the drafting and prosecution perfectly because you were trying to do gold standard practice.
Speaker 1:Yeah, it makes absolute sense. I'm going to keep going, if that's alright. Yeah, so I will let you in in a moment, obviously. So I've now lost my train of thought. That's awful, isn't it? Were you the first in-house?
Speaker 3:No, so there's actually two of us, which is quite unusual for a small startup. So there's only I think we're now headcount of 35, 40, and two of us are patent attorneys, and this is really because the space we're in which is data compression, video compression, is very IP heavy. It's a kind of standard, essential patented type space there has been a lot of.
Speaker 3:There is a lot of litigation in the space generally essential patentee type space. There has been a lot of. There is a lot of litigation in the space generally, um, and so the the founders were very keen to make sure the ip strategy was done correctly from the beginning, um, and so, very unusually as well, we have a very large portfolio for a company this small um, essentially yeah, we don't know the space, but it does seem quite early to have two patent attorneys.
Speaker 3:Yes, yeah, and I suppose that the interesting part of all of this is with video compression is the technology really up until now has been based on stuff from the 80s, the 90s, and although you can say we've made improvements, incremental improvements, the underlying concepts haven't really changed. And then in I guess it was 2016, there was some very early research where people were like actually you can use just autoencoders and neural networks to do the same activity and do end-to-end compression without any of the stuff from the old days. It turned out that it's a lot harder than that early research suggested, but the promise is that if it does work, then you are getting significant performance both in terms of compression quality and in much smaller bit rate sizes over anything that traditional compression can do, and so that's really the goal of the startup is to focus on that, and because it is such a massively valuable industry, we kind of wanted to get in first to have all the foundational patterns in the space.
Speaker 1:I'm going to let my geeky colleague come in at the moment, but just so I make sure I understand what's going on here, you can speak in a moment. So by compression it's taken a big file format and making it smaller for ease of communicating it around the world. So I'll go with that. I need to know what I'm talking about, all yours actually I've got loads of questions.
Speaker 2:I think I'll stop. Last thing you said, just to get out of the way. So MP4 is obviously a major coding paradigm or whatever the word is. I don't know that much about it, but what I've read about it it's a complete dog's breakfast of about 26 different compression technologies. Yeah, and it gets more complicated every time.
Speaker 3:So if you plot sort of on the graph of every year when they make subtle changes to the standard or whatever, the complexity goes up and the amount of compression performance starts to level off. So we are getting to the point where we're plateauing and we have maximized what traditional technology can do, and so everyone is very keen to find the next step of continuing that trend. That was my deep question.
Speaker 2:I'm going to take a few steps back now. I've got quite a few things to talk about. First of all, um so, physics degree to solicitor yes, not the most obvious direction. I would have said correct, yes, what drove that?
Speaker 3:yeah, so, um, again, this is quite a long time ago. Um, so in uni I had a housemate who was a lawyer and I hang out with a bunch of law students. Um, I was not that keen on going into academia. Um, I mean, at the time I was working on some quantum computing stuff and everything was lasers in a dark room and it wasn't so fun. Um, and because I live with lawyers, they were like oh yeah, you can still do a law conversion course, do a training contract, all the rest of it. So I went along to some of these career sort of days and and um, did a what they called summer vacation scheme back in the day and turned out it was pretty fun. Um, and so I thought this sounds like a fun career and I just went into it focusing on like a patent specialist firm. So bristos does a lot of patent work. Um, yeah, and so, yes, that's really how I ended up there.
Speaker 2:Um, next question yeah, so this is the patent attorney, not the most obvious career track also correct.
Speaker 3:Yeah, yeah. And so then when I got there and after I sort of really started doing the work and when you do your training contract you jump through different seats and you do different areas of law I sort of quickly realized that even in a firm that does a lot of patent work, the work, the day job, the grunt work you are doing is really not technical. Like it might be helpful if you can understand the expert reports or whatever else, but you're not actually really in the nitty-gritty of the technical stuff and I really kind of missed that. Um, and then I was like, well, what other jobs could there be? Um, my dad actually suggested if he looked at being a patent attorney and I was like I know what? Actually that sounds pretty good. So I applied for some, uh, some roles and then and got the one in oxford at marston club, but you're stuck with the patent attorney, obviously, but gone in-house you.
Speaker 2:Are you inventing at all as part of your code? Yeah, so so I, I, um yes, actually is the answer.
Speaker 3:So, and and and formally so, we actually filed the patent on some of our early work and we got notice of allowance like a few months ago, um. So technically, yes, I am also an inventor, um, but this is more like in. Again, this is the separate hat I wear, where I I build stuff, I script stuff, um to see really if there's any development. It's nothing to do with data compression, um, but whether it's elements in ai and just general kind of um progress in various fields. Statistics data can be used in a useful way in the patent space, um, and that is what I in my side gig and sort of start to monetize that as well.
Speaker 2:In terms of you painted an interesting initial picture of the life in a startup. It does sound like a kind of an 80s tribute movie to the HP early days, is it yeah?
Speaker 3:So I mean, like the real fun thing that I feel from working in a startup is everyone genuinely feels and it might just be in the space we're in that we are building something really really cool and the technical challenges are like really really hard, but when we solve something, everyone is genuinely like no one else in the world has done this. This is a world first and it's that kind of feeling of excitement that you get, which, even if you're working with inventors every day in private practice, you you're not part of that, whereas if you're going to be in-house, you are genuinely part of that. Um, but then separately as well. Um, you know, because I can code and I do kinds of fun stuff and I'm on the sort of the business ops team in in the startup. Um, I also write scripts and data analytics stuff just for the business side of things. Um.
Speaker 3:So a fun thing uh, I recently did was, um, we're sort of starting our well. Mostly at the end of our raising round for a series A, the founder of D-Render didn't really know which investors to approach first, which VCs to go for. So I built a scraper to scrape vast numbers of VC and then fed that into an LLM and kind of got them to rank each other and do like 1v1s AB testing and then found a list of like top which is most likely to invest in us and we start at the top and work down, and this time around with our fundraising, it's been much easier compared to last time.
Speaker 3:So there's a lot of stuff and you prefaced that with this is a fun thing to do it genuinely is so, and this is the thing where I think the choice to go from sister back into patent attorney and then to startup for me has been great, because I can go back to creating stuff and building stuff that sounds?
Speaker 2:it sounds fascinating actually there's a lot of use of ai, which is kind of gaming the system, isn't it? There's a huge amount of data out there that people are not using wisely and I bang them on on the back. This occasionally, like all the scientific creations that's been in the last 150 years, is in the patent system and people do not use that properly. That's interesting. I've picked up two kind of moments where you kind of gamed the system.
Speaker 2:So one is just there you worked out where the VCs were that might have the right kind of skill set, interest and so on, and went straight for them. It sounds like that was a win. The other one I'm quite interested in, because it's self-interested, is the fact that your business has gained the process and decided that a lot of IP and two patent attorneys very early on is the right way forward. Does that decision come from kind of experience, from intuition? Why have you gone for that?
Speaker 3:It is unusual, yeah, I mean so I think the founders are reasonably well advised by their original kind of seed investors pretty well advised, and I think the the problem with startups in the codex space has always been that it is a really difficult space to get into because the existing paradigm is so entrenched and the, for example, that there's different verticals within this space.
Speaker 3:One of them is video on demand, things like youtube and so on, and they, they are massive and and if you can get them as an end user, you're sorted for life. But the problem is incorporating something into that is a very slow sales cycle. It might be five years, 10 years, and so the whole industry moves slowly. And within this space, all the incumbents have vast past portfolios, and so I think that, against that backdrop, when the original investors decided to invest in the company, they knew that backdrop and so they said, well, how can we still maximize our chance of success? And so that was a very early decision that they made, which, again, I think is correct and seems to be working pretty well right now.
Speaker 2:I'm interested in any background on the founders because they sound like quite well yeah, so.
Speaker 3:Imperial graduates, both super duper smart. They originally were, I think, doing a sort of class assignment on image processing and it was taking them forever to transfer the data sets from whatever um sort of cloud-based storage to their local machines to actually do the exercise, um. And they were wondering I wonder if they could do this with a different way. And they did some early work and it turned out they could and that kind of then just spun off from Imperial into the startup that it is now.
Speaker 2:Proper Facebook backstory kind of vibe to that.
Speaker 3:Yeah, and it's interesting as well. So the startup was part of Intel. Have this startup program called Intel Ignite? The deep end has won the award for it. What is it? Last while ago, and when Pat Gelner, who's the CEO of Intel, introduced the startup on the stage, he basically said these guys are the biggest nerds you have ever seen. And it's like okay thanks, thanks, I guess, but yeah, so it's pretty cool to be part of that.
Speaker 2:I want to ask a bit about the funding side as well. So, because that's always part of what obviously keeps you going, are you monetizing it or are you kind of still Codec?
Speaker 3:product. Yeah, so we're still basically pre-revenue. We have a working, basically working product. But again, the problem within the Codec space is people tend to want to have a reasonably well-working product, both encode and decode side, before they'll even start looking at you. So it is still pre-revenue, but we actually have a working product which no one thought was possible like a year or two ago.
Speaker 2:And so you're looking to break into kind of conventional YouTube market. Are you looking for the next generation of users for this?
Speaker 3:Well, so interestingly, originally there was always going to be some question about do existing devices have the right kind of hardware to run this kind of software? Luckily there has been this massive wave of interest in AI generally across every platform and every space. You know there's Windows advertising their AI PC or whatever it is, and all these things run on something called an NPU, which is a neural processing unit, and the general idea is our codec works really really well on NPUs, so as they propagate. The best thing about all of this is with compression. Existing technology needs a special ASIC, so a special little chip that is dedicated just to do the compression. With our kind of codec. You don't even need anything specialist, you just need NPU, which every device will have in the next five years or so.
Speaker 1:Even I understood that. Even I understood that.
Speaker 2:Beautiful explanation, because NVIDIA was it today has just been announced as bigger than Apple, haven't they.
Speaker 3:Yeah, the stock price has gone through, although I think that might be a bubble, but, yeah, no financial advice, no financial advice.
Speaker 2:Another declaration there as well. What? About your um your coding background. That's quite interesting, yeah, so it's kind of interesting.
Speaker 3:So I don't have no formal training. Um, I did a bit of coding at uni, um, I've always kind of just messed around on side projects. I guess the the original stuff was back when I was a child. I've got two brothers and we were having LAN gaming parties and trying to get the various computers to talk to each other. Never worked, and so we'd spend six hours trying to get the networking stuff set up and then it'd be the end of the day we'd play like 10 minutes of a sort of co-op game and our parents would call us down for supper.
Speaker 3:So you know, it kind of started that sort of stuff, um, and then over the years I've just just for fun, built stuff, um, and again it's, it's just, yeah, no formal training. But when you play with these things, um, you pick up tricks and sort of just learn as you do it, which again I think is you know there are plenty of people, the traditional ways to go computer science or something else and then you kind of get formal training, whatever, there are plenty of really good programmers out there who just do it, and then you learn by doing it.
Speaker 2:So can I talk about the patent tools? Can I carry?
Speaker 1:on yeah, I was going to go there. No, no, no, you don't mate. No, you do it. No, you do it. You know what you're talking about. You do it. You know what you're talking about.
Speaker 2:It's funny.
Speaker 1:Lee pointed out that when we were in the States, AI came up almost on much every podcast and today especially the AI one you weirdly didn't talk about it then.
Speaker 2:Of all the podcasts, we've talked about it a few times today. Talk about it all the time, some educated, some just it's going to cure all ills. You mentioned that you had a side gig to make AI support for the patent area. What sort of products are you looking at there? You had a side gig to make AI support for the patent area. What sort of products are you looking at there?
Speaker 3:Yeah, so kind of interesting. So the obvious use case is what we kind of started out doing as well. It's just a drafting tool for just like creating this draft spec, whatever. But then on, the more interesting use cases I actually think are on the prosecution side and it is a harder problem to solve. But that is really where I think the time saver is and and this has kind of been validated by, as we've monitored, like kind of usage metrics we find that actually, even if you give patent attorneys tools to basically draft the patent, they only really use it to create a skeleton, so like paraphrase their claims. You know they'll still draft the claims manually. The key part of the invention that actually really matters, they'll still draft the claims manually. The key part of the invention that actually really matters, they'll still draft manually. And so the actual time saving isn't very high. And then you look at that use case tools to just do the claim paraphrasing and the basic stuff have existed since the early 2000s and they haven't been widely adopted. So the drafting stuff I think is a little bit overhyped. We have a drafting tool, but yeah, it exists On the prosecution.
Speaker 3:Prosecution side it is a more interesting problem because, um, really, how you? What you do is you're a trainee, you go away, you look through the prior art, you find a normal feature, do your argument, boom done. But the going away, going through the prior art, that takes time. You're charging for time.
Speaker 3:Um, and that type of task, like finding features, needles and haystack, that type of thing can be done pretty well with ai, but within the scope of that, there are serious limitations with all existing tools and how good they can do that. Um, and you know we're no exception. Everyone's tool suffers currently from the same problem and I'm going to go into technical details if you like, because it is actually fascinating and it helps to kind of explain, yeah, problems of what's going on. So, um, every single tool that alleges that they can find something in a large corpus of text pretty much now uses a concept called rag, where you take a chunk of text, you cut it up into lots of paragraphs or sentences and you turn these things into an embedding or like a vector, basically, a series of numbers and you kind of you could imagine this thing as like a big space of vectors, like a big vector space of stuff pointing lots of directions.
Speaker 3:And in this big space I'm trying to find something. So I take the thing I'm trying to find like a word or a sentence, and I cut, put that into a, into a vector as well, and in vector I put it into the vector space and I say, well, what's nearest to it and semantically that's likely to be the most relevant. And so we say, right, I'm going to take the top five most relevant things and then put those into the prompt of a large language model and then ask the language model is this feature in these five chunks of text? And the language model will confidently answer yes or no and so on. So you basically do a search for something, feed that to a language model and get it to talk about the answer.
Speaker 3:But the problem with this kind of approach is that you are entirely reliant on your vector search or any other type of search being accurate. If you don't get the correct chunks of text, you'll get the wrong answer. And so everyone is focusing on the large language model aspect of things. But really the hard part is in the search and no one has solved this. You can go to the think from last week where stanford looked at all the existing legal research products like lexus, nexus and some other ones, and they found some hallucinations in like 30, 40, 50 of the answers. So hallucinations are when a language model confidently answers something but it's incorrect and it's based on something sort of made up made up, yeah, um and so there's this kind of.
Speaker 3:There's a lot of hype and there's a lot of buzz about it, and the products do not ours included, do not live up to the promise of we're all going to be replaced by AI. But underneath all of the hype, there is genuinely, I think, something super useful, and it's just no one has quite found really how and why it's going to be useful yet so it's about.
Speaker 3:When you say it's about the search, do you mean about the search criteria or about no, so so the the accuracy of the documents, of the chunks of text you get back, all right. So the the part of the problem is why you even need to do this to begin with is that most language models have a limited context space, and even language models that are advertised to be able to take in an entire novel and tell you who did what on what page. When you actually look at granular detail, they suck. That's a technical term.
Speaker 1:Yeah, exactly.
Speaker 3:And so the problem is that, well, if you can't put the entire book into something when you're searching for something, you have to put parts of it in.
Speaker 3:And so which parts do you put in to get the answer? And you just have to hope that the parts you've chosen through your search contain the answer, and if they don't, you're going to get the wrong answer. Yeah, but then? So something we're trying now, which is again, it's it's crazy and who knows if it's going to work is we've basically adopted the assumption that language models won't be able to do very well for long contexts. Um, so we're just going to say, right, we're not even going to bother trying to do long contexts, but can we brute force this by saying, rather than picking a few chunks of text of my novel, I'm just going to go through every single chunk of text and check it with a large language model, and so you generate. It's a huge amount of GPU, time and things, but so far our results are really accurate. Um, compared to this, like rat-based approach, um, and something we've just started doing is claim charting the entire lte um standard.
Speaker 2:So, basically, um again, this is, this is kind of like.
Speaker 3:It's kind of super technical and maybe irrelevant, but tell it, in telecom space there's like these standards, um, one of the problems people have is how do you know if your pattern is actually essential to the standard yeah, um, and what people up until now have either been doing is a semantic based approach with all these vectors and if it's similar, you say yeah, that's probably standard, essential, but no one's actually checked and the only way to check is to really to do a claim chart and it's really expensive to get experts to do this. It takes forever and no one's probably probably done more than like I don't know, like say five. They say like a few thousand out of the human gene though.
Speaker 3:Yeah, yeah, so no one, no one's done more than like 100, like the total amount of self-clared patterns right now is, like I don't know, the database, 140 000 or something, and so you'd have to do, uh, if you just take claim one, brute force map 140 000 like individual claim ones against these like 200, 300 page standards. So there's it's, it's, it's like a super computing exercise and we've started doing it, um, and we'll see if that goes anywhere.
Speaker 2:I've actually, you see, you actually kind of preempted my next question, which is it? Going back to the vast amounts of patent data that's out there? What else is out there that we could be doing? And that's that's clearly a clear-cut one is getting to get telecoms and work out who's patent covers who in simple terms. Another one for me is just a bugbear of mine actually is the whole FTO thing I'm trying to work out. If you've got a product, a complex product, and someone says, right, does this infringe stuff? Is that the kind of thing you think there's a space for?
Speaker 3:then how do you sort those out? That's really too many to brute force check everything and maybe for a super valuable patent you might try, but there is no solution to that problem, basically. And you'll see a lot of people saying, oh, we found the solution. We can find anything in any corpus of text instantly and basically all they've done is taken semantic search and thrown that into a language model and that's okay. But there's plenty of studies out there now that say that doesn't actually work.
Speaker 2:That's disappointing because that's the one thing I actually want.
Speaker 3:I know, right, and you think also, if someone would have discovered this before at the patent offices, right, because this is something the patent offices would love to have, a ground truth. Is this thing novel, yes or no? Yeah, but that doesn't exist. It is odd.
Speaker 2:I mean we are besieged by AI products and unfortunately they're mostly kind of samey.
Speaker 1:Yes, With all due respect to all the real ones everywhere.
Speaker 2:you know there's a few really important things we still don't quite have.
Speaker 3:I think part of the problem is it's technically you can get 70% of the way there very easily now, but how do you go from that 70% to 80% to 90%, 95% to 100%? Technologically we aren't there yet. So, despite all the hype, that isn't there yet.
Speaker 2:I don't want to lose you. I don't want you to disappear into AI.
Speaker 1:That's a fair point.
Speaker 2:I have one more question, back to life in the startup and the commercial versus I don't want to lose you, I don't want you to disappear into AI. That's a fair point. That is a fair point. I have one more question. Go for it and it's back to life in the startup and the commercial versus legal slash, academic advice. So how do you reconcile?
Speaker 3:that as a physicist.
Speaker 1:Yeah, yeah.
Speaker 2:We like perfection, we like ones and zeros and everything, and suddenly you've gone from the partner's only private practice point but you get as close to 100% as you can. So presumably startup world, where actually you mentioned 70%.
Speaker 3:I don't know where you're, yeah, so I mean so one example, say, is we've got a US examiner who is not like, who's really difficult to get around at the moment, and the question is do we want to keep this thing pending? Do we want to keep paying our US attorney, their fees et cetera, yes or no? There probably is a way around. We could probably make some arguments et cetera, but commercially it's just we're just going to drop it. What's the point? We've got better things to do and it's the kind of thing where we can make that decision quickly, easily he's not going to bother spending lots of time on this. Where we can make that decision quickly, easily, he's not going to bother spending lots of time on this. Whereas if you get that kind of office action in private practice, you might sit there going ooh-ing and ah-ing for a while. What's the best strategic decision? I don't know. Whereas when we're in charge, we're just like, yeah, it's fine, drop it, it's sort of much quicker. Bit of a random one for me Do it.
Speaker 3:So in everything you say, ed, you seem to be trying to invent or create yourself out of a job. Yeah, that's correct. That's correct. Did you gain that bit, did you?
Speaker 1:but also him and all of all of our members. Yeah, do you feel a bit like an oppenheimer? Yeah, so are you, are you? A destroyer destroying of industries yeah.
Speaker 3:So as much as I would love to get myself out of a job and automate myself away, I don't think that is coming in our lifetimes. And actually I suppose more interestingly is I think there are other threats to the patent profession, more so than AI. Is the bigger threat not doing it? Well, no, I think the biggest threat is, for example, the general commoditization and weakening of the patent system.
Speaker 3:That type of stuff, yeah, is more of a problem for patent attorneys generally, because if you find that you're as a business, your patents are less valuable, do you spend the I don't know sort of 500 000 over x years building a reasonable portfolio, or do you just spend that instead of buying two or three more researchers and keeping things a trade secret? Yeah, and the, the needle or the, where that sort of falls, I think is moving more and more towards the get a few patents but actually they're so hard to enforce, like, say, in the us, the, the was it 90 percent invalidation rate, p-tab um, the most of the big tech companies basically just ignoring the existence of patents? Um, as smEs and startups, it's really hard to say that a patent is actually valuable when it comes to enforcing it, and that, I think, is a bigger threat to our industry than AI coming on and taking our jobs. Can I be philosophical then?
Speaker 2:Because the point of the patent system is to simulate innovation. I think.
Speaker 3:That's absolutely right.
Speaker 2:If it's being weakened, then there's two options. Either people are driven towards trade secrets, which diminishes innovation for different reasons, because it just stops the flow of information, or we stop innovating because there's no point. You can't protect it anyway. Do you think the patent system should be weakening?
Speaker 3:I mean, you're a startup, you've got to deal with a bunch of big tech patents, so I think, as it weakens now and as a result, probably, costs are also generally being driven down, and this is for various reasons. Like the cost of doing a draft now is probably approximately equivalent to five years ago, maybe a little bit before I hadn't really kept up with inflation. Like those kind of factors, um, like they will end up causing more of a problem in the longer term, I think. So what was the question again?
Speaker 2:Do you think this potential weakening of the patent system is a good? Thing, as a startup, you know your innovators are good or bad for you guys.
Speaker 3:So, the point I was going to get to was, as patents weaken, they get cheaper, which is great for us if we're using it to attract investment. So our earlier guest talking about having a patent that attracts investment is useful. That system is great, probably for five to 10 years until the investors catch on that. Actually, patents aren't so great.
Speaker 3:So, it's a great system currently for us because they're getting cheaper, we can file more and we look great to investors, so we can get investments and that works, because then, ultimately, the purpose of the patent system is to encourage innovation. How do you do that? Make sure the money flows to the right people doing the right investment, which you can do with attracting investors. But then that may suddenly stop when investors are like yeah, we don't care about that, because the paper it's written on is not worth it.
Speaker 2:And so it's sort of a yeah.
Speaker 3:I don't know where it's going to go.
Speaker 2:Well, no, I think the answer is don't erode the patent system too much Correct.
Speaker 3:Yeah, it's done well for hundreds of years.
Speaker 2:Again unsurprisingly on two IPs in the pub we like IP.
Speaker 1:Sort of in the name, is it? Should we ask the audience if they've got any questions? It would be unfair not to ask them anything about vector space anyone?
Speaker 2:um, um, yeah, it just reminded me of a. Again, not not to advertise anything, but it reminded me of an excellent series of stuff, just Silicon Valley, and I think it's it's the live version of Silicon Valley yeah.
Speaker 3:So for those who don't know, Silicon Valley is a show about Silicon Valley and it's like a compression startup. Oh, right and it's an AI startup and it's literally like we even have some people at work who look quite closely to the characters. I think you're just missing the beginning of God.
Speaker 1:Not such a big flurry of questions, are you? It's the vector space thing have you got any more.
Speaker 2:I'm done. That's a fantastic setup. I was going to ask who Brenda was. Brenda, you named your firm Deep Brenda.
Speaker 3:Deep Renda, oh right.
Speaker 2:Sorry.
Speaker 1:Deep Renda. No wonder we didn't get the plastering reference at the start.
Speaker 2:You're on fire today.
Speaker 1:You're on fire today, aren't you Never knowingly sleeping on the job? Are you sat there thinking? I forgot to ask this to any of our guests. Are you sat there thinking? I forgot to ask this to any of our guests. Are you sat there thinking oh, they didn't ask me that one.
Speaker 3:No, not really.
Speaker 1:In which case we move swiftly to the closer. Have you got a closer? No, it's your job. I don't want to do that.
Speaker 2:You always do one at a time.
Speaker 1:I thought you went to every other one. No, no, I've done one today. You always do one like that. I thought you went through every other one. No, no. So I can only really think about regurgitating stuff that we've done previously when it comes to AI. So we've done in the past kind of what would you get for AI to usefully take out of your life? I can't remember what your answer was last time what would you get for AI to usefully put into your life?
Speaker 2:AI to usefully put into your life. Ai to usefully put into my life. I think it would have to be some way of helping me play guitar better. I want to kind of play what I do and then put it through.
Speaker 1:AI and make it sound like it's in my head. You've got your tuner device, haven't you?
Speaker 2:Because I remember you talked about that, yeah that's really good for just giving you ease and your A's and everything, but I would like to just play what I think sounds amazing and then to go through AI to come up with something fantastic on the other side.
Speaker 1:That's a really good answer. Now I'm going to ask an expert. So, autotune right, oh yeah.
Speaker 3:Yeah, what would I add? So coming up with story ideas. So I've got two kids and they just want me to tell stories over and over again and for like after five hours you kind of run out of ideas. So, um, probably that I have. Actually I have actually used various tools for this of course you have.
Speaker 3:Yeah, and then also like illustrated it as well yeah so basically you chat gpt for like coming up with story ideas, write a story and then get it to generate pictures and then I just show it to the kids and they're like look once upon a time and like I love telling stories. It's just the problem is it's after five hours the muse can go yeah.
Speaker 1:Lee. So a lot of people don't really kind of appreciate this, but I hate, I hate networking. I'm not, I force, I force myself to do networking.
Speaker 2:It's transactional.
Speaker 1:Yeah, yeah yeah, and I always feel a bit kind of out of my comfort zone, kind of lost in space and stuff like that. So anything that could stop me from having to do it, something that could like network kind of artificially for me, I'll take that. How do you do that in the room with people in it? I don't know. This is my question. I don't have to respond to.
Speaker 3:Have you seen the latest ChatGPT video where they have a little thing on their phone talking? Everyone can just have their little AIs talking to each other.
Speaker 1:You just put them all on a phone, they network and you go and have a drink. That's so relaxing. Go to the bar. Don't talk work, let the AI talk work. Thank you for coming on yes. I would like to say that was absolutely fascinating, and I'm sure it was. I got lost at times, but I understand it more than I did when I did my research earlier. It's now it's what made sense to me. Good thanks for being with me all day today. It's been a long day, hasn't it?
Speaker 1:it's been brilliant, we've got there four great podcasts four a really really good podcast if you've enjoyed this podcast as much as I have, then obviously what you need to do is leave us a little review on the podcasting platform of your choice so that other people can find us. That's what we need. We need Hank O'Tooney, ai, decent podcast finder. That's what we need, or just.
Speaker 2:AI can just do the podcast. See you on the next one. See you soon, bye.