Jump to content

ChatGPT AI Tool— We all work for robots now


956 Worldwide

Recommended Posts

15 minutes ago, Elvis said:

Wow.  That was really interesting.  I have an engineering degree, so I took matrix analysis and vector calculus.  Without that knowledge and an understanding of multidimensional space and vectors, that would be total gibberish.

If you liked the math, you should check out the whole Deep Learning series, he starts out with visualizing the vector operations and a 3D model that he uses for explanations. I did linear diffeq and vector cal at Texas and I still feel like the explanations improved my mental model and understanding

Link to comment
Share on other sites

Here's a pretty easy way that I think about a multi-dimensional vector (or space).  A 2-dimensional vector is easy: X and Y axis.  3 dimensions add Z.  4 dimensions add time.  Want to add a bunch of other qualities?  add a 32-bit color space...there's a bunch.  Then any other real or non-real qualities: frequency, race, religion, quantum spin, etc.  Big multi-dimensional vectors, space, and matricies are pretty interesting.  I'm not really that smart any more, so maybe that is all bullshit.

Link to comment
Share on other sites

28 minutes ago, Elvis said:

Here's a pretty easy way that I think about a multi-dimensional vector (or space).  A 2-dimensional vector is easy: X and Y axis.  3 dimensions add Z.  4 dimensions add time.  Want to add a bunch of other qualities?  add a 32-bit color space...there's a bunch.  Then any other real or non-real qualities: frequency, race, religion, quantum spin, etc.  Big multi-dimensional vectors, space, and matricies are pretty interesting.  I'm not really that smart any more, so maybe that is all bullshit.

Bro do you even eigenvector??

The dimensionality of most RAG knowledge bases is at least 1000 parameters, and in the enterprise usecases I've helped my customers deploy they're upwards of 15,000 parameters. It's pretty crazy because at those higher dimensionality spaces you can start to categorize and encode wildly different types of information. You can entangle the lingustic meaning of "clear blue sky" with the actual color of sky blue, and the idea of warmer temperatures, and everything else that could contextually enrich that phrase.

Or brought into the business context, you can use a multimodal input to process and transform pictures of items coming off a production line for QA and create a pretty damn good description and written report that's checking visual and spatial qualities. Or you could feed a knowledge base all of your contracts originated from your various branch offices and have an O(1) lookup for the meaning within those contracts to determine what it is you've actually comitted to or are due. Sure you still have a lawyer validate it, but you can sift through SUCH A HUGE VOLUME of information incredibly quickly and accurately to only act on the things that need it.

Link to comment
Share on other sites

  • 3 weeks later...
18 minutes ago, bolverk said:

 

image.png.f719cfa504549e9c9cd6d69126995978.png

image.thumb.png.223f3ea209b45cbe12a2c018f5396e85.png

image.png.8e05284689a1acc0aa380024acbf7998.png

Well, if true.....that's a huge concern for literally every business that has an obligation to maintain the confidentiality of certain information.  So, literally the entire world of healthcare enterprises, financial services companies, legal services/law firms....you fucking name it.

AI is a virus that every platform is injecting into everything you use, whether you want it or not.  And it can't be fully controlled or even have the scope of its activities fully defined.

What could go wrong?

Link to comment
Share on other sites

16 minutes ago, Vegas64 said:

Mmk GIFs | Tenor

 

So Ed Zitron and Dahobbs are morons.

Why do you think he is wrong? Currently, the system is of questionable reliability by OpenAI's own admission. Given how training it works, it isn't clear how adaptable this process would be to changes in website design to the extent the tasks interact with a website. If the tool isn't near 100% reliable and accurate, it isn't clear what value it brings. Right now, it is essentially a Rube Goldberg  version of macros and web scraping. 

Link to comment
Share on other sites

14 minutes ago, Dahobbs said:

Why do you think he is wrong? Currently, the system is of questionable reliability by OpenAI's own admission. Given how training it works, it isn't clear how adaptable this process would be to changes in website design to the extent the tasks interact with a website. If the tool isn't near 100% reliable and accurate, it isn't clear what value it brings. Right now, it is essentially a Rube Goldberg  version of macros and web scraping. 

For one thing, Ed's fearmongering of not being functional because of a redesign of the GUI is dumb to me. Even the "intelligent" automation that exists today (the second and third generation of macros and web scraping) are resilient enough to handle UI changes and not break the bots. If even these dumb automations which have been using dynamic capabilities and computer vision for years, you think the new OpenAI digital agent/computer vision will be more fragile and break?

Secondly, $200/month is not incredibly expensive and for the use cases that this can be pointed at and the value captured can be massive. Industries with massive amount of manual or (headcount) intensive labor to punch keyboards or copy and paste and move data around spreadsheets into software systems, and especially the OSP/BPO space, will see massive business value. And even if it's not 95% accurate as Ed suggested was the break-even point, there are still massive efficiency and cost gains (full time employees, manual labor hours) that can be captured.

 

Link to comment
Share on other sites

@Dahobbs My hunch is ole Ed Citron must not be very technical and/or a practicioner of any sort of technology (or owner of any business process) to be making these asinine claims. But happy to be wrong and have him be a Sam Altman-type and know some secrets. But at the surface level, it was moronic to share his moronic opinions and pretend they were insightful or intelligently provocative. They were embarrassing.

ETA: Sorry if you are Ed, if Ed is you.

Edited by Vegas64
Link to comment
Share on other sites

4 minutes ago, Vegas64 said:

@Dahobbs My hunch is ole Ed Citron must not be very technical and/or a practicioner of any sort of technology (or owner of any business process) to be making these asinine claims. But happy to be wrong and have him be a Sam Altman-type and know some secrets. But at the surface level, it was moronic to share his moronic opinions and pretend they were insightful or intelligently provocative. They were embarrassing.

ETA: Sorry if you are Ed, if Ed is you.

Ed Z has been a tech writer for more than a decade and actually has a pretty good handle on this stuff insofar as a writer would. 

But more to the point: openAI is hemmoraging money like crazy and loses more money the more that people use their ChatGPT subscriptions. Its a tremendous amount of energy and compute capacity to train these models, and the multi-modal agentic design only amplifies the number of token used to respond to a query.

As a technical specialist in this field, I can tell you that their offering is more like a prototype than an enterprise-ready solution. That's not to say all of this tech is bullshit, but holy hell there's a lot of snake oil being bought and sold

  • Hook 'Em 4
  • Like 1
Link to comment
Share on other sites

3 minutes ago, Vegas64 said:

For one thing, Ed's fearmongering of not being functional because of a redesign of the GUI is dumb to me. Even the "intelligent" automation that exists today (the second and third generation of macros and web scraping) are resilient enough to handle UI changes and not break the bots. If even these dumb automations which have been using dynamic capabilities and computer vision for years, you think the new OpenAI digital agent/computer vision will be more fragile and break?

Secondly, $200/month is not incredibly expensive and for the use cases that this can be pointed at and the value captured can be massive. Industries with massive amount of manual or (headcount) intensive labor to punch keyboards or copy and paste and move data around spreadsheets into software systems, and especially the OSP/BPO space, will see massive business value. And even if it's not 95% accurate as Ed suggested was the break-even point, there are still massive efficiency and cost gains (full time employees, manual labor hours) that can be captured.

 

Simply because current tools are somewhat adaptable doesn't imply this tool must be more so. There is actually zero guarantee of that. You're expecting of level of generalization that LLMs simply don't exhibit. If the website is within its training set or very similar to a website within its training set, it should work. But, website design changes fairly regularly. Amd many changes designed to promote usability actually confuse human users and result in lessened usability until the person becomes accustomed to it. That is simple for a human to do, but much more complicated for an LLM. Any a change that confuses a person is likely to also confuse an LLM like this. I also question the idea of using this product for web interaction generally. The applications where it might make sense are better served by dedicated APIs because those are more direct interactions and programs interfacing with them are less susceptible to errors. 

As to this being a solution to data entry, I think that is questionable as well. Many such positions have been replaced by automation already where it makes sense. And that automation is much more reliable than anything demonstrated by OpenAI. Your belief that lower accuracy is rates would still be beneficial is also questionable. Garbage in/ garbage out is common saying in the industry for a reason. High volume of a garbage in isn't inherently useful just because there is a lot of it. Usually it just means it will cost more to fix. 

Link to comment
Share on other sites

16 minutes ago, Vegas64 said:

My hunch is ole Ed Citron must not be very technical and/or a practicioner of any sort of technology (or owner of any business process) to be making this asinine claims. But happy to be wrong and have him be a Sam Altman-type and know some secrets. But at the surface level, it was moronic to share his moronic opinions and pretend they were insightful or intelligently provocative. They were embarrassing.

What experience do you have with tech? Your criticism of a fairly obvious set of concerns would have me question your acumen. And keep in mind the $200 a month costs you're citing don't give unlimited access. Each request requires a significant amount of energy to process. The type of volume tasks you're thinking this will apply to will be many orders of magnitude more expensive. Again, I think this solution is significantly less efficient than current automations. It only makes sense to use it in situations where current tools don't work well, but it isn't clear it would actually be good at that. 

Link to comment
Share on other sites

On 12/17/2024 at 8:54 AM, Captainant said:

I think this is missing a big opportunity in linguistic data science. Parents are thick, complex documents that are hard to parse through. If the USPTO were to do a feature extraction (which uses genAI mechanisms and models) on its corpus of patents that would greatly simplify the process of checking for prior examples when processing new applications. Transformers are still being explored and learned about, but they already have great use as a tool for processing large and disparate sets of data - like patents.

Retrieval augmented generation and feature searching through documents is going to be very transformative. A customer of mine is exploring a similar technique for getting their arms around their huge fleet of contracts where every region has their own way of doing things and nobody in their central governance wants to take the time to read hundreds of thousands of contracts. It's allowed them to get hard, verifiable data out of their paper and start using it at scale to inform future decisions making

They're absolutely NOT using it to originate contracts though. That would be lunacy. 

Patent searching is where any benefit will be seen.  The PTO has its own search databases and there are AI-driven search tools in the private sector.  I'm not really sure how good or useful they are. They are quite expensive.  Once AI searching has become more mainstream and proven, they'll probably adopt it for their search tools.

I'm sure the PTO is concerned with examiners using "open AI" to formulate rejections and other patent legal correspondence.  And that runs a huge risk of generating gibberish and the equivalent of seven-fingered humans and would be a disaster.

And security of using commercial or free AI is a big concern.

The difficulty with patents is that a "patentee can be his own lexicographer."  The text descriptions of the same or similar structures or features varies tremendously across the spectrum.  One of the best tools a patent examiner has is the ability to scan drawings once a search result is obtained/narrowed by text-based searching.  I'm not sure how well adapted AI is to that.

Edited by TwiceHorn
  • Hook 'Em 1
Link to comment
Share on other sites

8 minutes ago, TwiceHorn said:

Patent searching is where any benefit will be seen.  The PTO has its own search databases and there are AI-driven search tools in the private sector.  I'm not really sure how good or useful they are. They are quite expensive.  Once AI searching has become more mainstream and proven, they'll probably adopt it for their search tools.

I'd bet my bottom dollar those quite expensive systems are just up-charging an off the shelf system or reference architecture. Hell my desktop PC with a 3080 can index 100,000 pages of documents in but tens of minutes, and what I've helped my customers implement is on the order of a few cents of one-time compute costs per ingested document. But they're also building it themselves (with some help 😉) instead of buying professional services from Oliver Weiman or Deloitte, which is saving them a fat bundle

16 minutes ago, TwiceHorn said:

I'm sure the PTO is concerned with examiners using "open AI" to formulate rejections and other patent legal correspondence.  And that runs a huge risk of generating gibberish and the equivalent of seven-fingered humans and would be a disaster.

I mean, we're already running with chatGPT-authored executive orders. May as well smoke em if ya got em at this point

  • Haha 3
  • Rage+1 1
Link to comment
Share on other sites

1 hour ago, Vegas64 said:

@Dahobbs My hunch is ole Ed Citron must not be very technical and/or a practicioner of any sort of technology (or owner of any business process) to be making these asinine claims. But happy to be wrong and have him be a Sam Altman-type and know some secrets. But at the surface level, it was moronic to share his moronic opinions and pretend they were insightful or intelligently provocative. They were embarrassing.

ETA: Sorry if you are Ed, if Ed is you.

stupid-forest-gump.gif

Link to comment
Share on other sites

38 minutes ago, thunderlounge said:

I‘Ll be happy when they finally stop calling these glorified algorithms “AI”. 

Man, that was the prevailing theme from CES this year and it sucked. Ed Zitron actually reported live from the floor for 5 days this year and discussed that and many other topics

It's crazy how many products and services are claiming to be "AI" simply because they put it behind a chatbot powered by an LLM. But someone has gotta squeeze those snake for their oil, dagnabbit!

It's frustrating because it muddies the water and sucks the oxygen and trust away from ACTUAL use cases for the technology. Transformer architectures are pretty neat, and yet still offer loads of innovations in fusing together disparate data sources to synthesize a more rich understanding. We're still in the metaphorical vacuum tube era of that computer science. Things tend to accelerate fast, but the bottlenecks they're increasingly running up against are power and water. 

Gonna get reeeeeeaaaaaalllllll interesting.

  • Hook 'Em 2
Link to comment
Share on other sites

12 hours ago, Captainant said:

I'd bet my bottom dollar those quite expensive systems are just up-charging an off the shelf system or reference architecture. Hell my desktop PC with a 3080 can index 100,000 pages of documents in but tens of minutes, and what I've helped my customers implement is on the order of a few cents of one-time compute costs per ingested document. But they're also building it themselves (with some help 😉) instead of buying professional services from Oliver Weiman or Deloitte, which is saving them a fat bundle

I mean, we're already running with chatGPT-authored executive orders. May as well smoke em if ya got em at this point

I suspect that to be true as well.  However, it would seem to be readily achievable to train a system on just US patent documents and along with the existing classification system and bibliographic information achieve a pretty useful database and query system.

It seems, though, that the current AI offerings more deal with finding related patents and families of patents and collating their prior art references than with actually locating a given piece of relevant prior art.

An example of the lexicographer problem:  We have a ton of Asian-originated US patents, which means they're translated from a pictographic language, which never goes well.  I was reviewing a Chinese-originated patent that used the term "swallowtail" when referring to a "dovetail" structure.  A translation issue that is an excellent example of the wild variation in language of patent specifications describing the same or similar structures.

One major difference between the US and European Patent offices is that EPO divides responsibility for searching and examination, while US examiners do both.  There's not much overlap in those skill sets.  Under time pressure, examiners generate a fair amount of gibberish without AI and poor searches doesn't help that much.  The worry at this point would be that pushing through the gibberish to get to the meat of the matter would extend patent pendency.

Link to comment
Share on other sites

Evidently there’s some new AI app called Deep Seek. I don’t know shit about this, but supposedly it was built with less capital and is open source? Stock market futures down, presumably on the whole AI bubble getting blown? And btw it’s from China. So there’s that geopolitical angle. Chamath gives his breakdown. 
 

https://x.com/chamath/status/1883579259769462819?s=46

 

Link to comment
Share on other sites

Ok so sounds like my ability to buy tickets to a hot concert are going to get even shittier?  My first thought was all the scalper asshats will run the $200 ChatGPT tier to broaden their ability to gobble up more tickets by obfuscating they’re essentially script kiddies.  

Fuck I miss the days of seeing Mumford at Stubb’s for $35, parking across RR in that weird little garage thing for $8, and not having to be on like 4 ‘presale’ lists just to buy a shitty $90 seat at Moody.  

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.



×
×
  • Create New...