Pages: [1]   Go Down

Author Topic: Poor old OpenAI  (Read 5370 times)

KLaban

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2752
    • Keith Laban Photography
Poor old OpenAI
« on: January 31, 2025, 10:34:37 am »

Poor old OpenAI, did DeepSeek rip you off then? Boohoo.

Perhaps you'll now know what it feels like for creatives?

But not holding breath...

Chris Kern

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2204
    • Chris Kern's Eponymous Website
Re: Poor old OpenAI
« Reply #1 on: January 31, 2025, 12:08:01 pm »

Poor old OpenAI, did DeepSeek rip you off then? Boohoo.

There's plenty of evidence of Chinese companies stealing or reverse-engineering technology intellectual property from other countries, but that doesn't appear to be true in this case.  The DeepSeek software was developed from and is available as an open source model, and consequently its code is publicly available for other research groups to work with.

Fareed Zakaria (CNN, Washington Post) published an interesting column this week about the DeepSeek approach:

Quote
. . . over time, open artificial-intelligence systems are likely to outperform closed systems. . . . Many have pointed out that DeepSeek used Meta’s open-source Llama model to train. It also used Qwen, a family of AI models, also open-source, put out by the Chinese technology giant Alibaba. While DeepSeek is currently the best, China’s big technology companies have been releasing a number of AI models, mostly open-source, that are getting better and better. If the history of technology is any guide, the ability to see the innards of these models and understand their reasoning should lead to greater and faster technological innovation than using closed models that others cannot use for collaboration.

The entire column is available here, although I can't guarantee the link will work from all locations.

Rob C

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 25125
Re: Poor old OpenAI
« Reply #2 on: January 31, 2025, 05:26:15 pm »

Every day the value of being human recedes further into a sentimental past.

I’m really surprised how so few seem to have understood that about the digital revolution. Yes, on a superficial level it has brought us new toys once unimagined, but just as truly it has brought about a widening of class distinctions and the inescapable truth that for those of our tribe not gifted with technologically inclined minds, we are being driven inexorably into a second-class citizenship far different to the old one, where such social divisions were largely defined/dependent on class and educational opportunity. The new second-class citizenship is more corrosive yet, and its resulting exclusions perhaps deeper, if only because those newly affected are bright enough to understand that, for them, it is a matter of losing a standing that they once had, whereas for the older sub-class there was no something they once had - unless, of course, you want to include the opportunity of holding down lowly jobs nobody else wanted. I wonder who, in Trump’s brave new America, will take on the jobs that the poorer of the immigrants were so happy to accept? I suppose they will later be presented as ingrates, who quit America because of its new political colour… and millions will buy into that belief, expulsions forgotten.

Am I alone in seeing that increasing automation in production has led to huge losses of factory jobs, jobs perhaps not glamorous, but skilled, nonetheless, and with skills comes personal pride and satisfaction; if people lose that, end up serving fries for a pittance, technically they are employed, but hey, there is employment and then there is employment that brings satisfaction and self-esteem. As all current indications seem to agree, the rich are getting richer and the poor poorer. The problem there lies in the fact that employment figures are seen as good: yes, more people may have jobs than ever before, but the wealth figures surely indicate that full employment does certainly not signify people are better off than ever before.

Redcrown

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 544
Re: Poor old OpenAI
« Reply #3 on: February 01, 2025, 12:18:50 am »

I'm an old techie who retired in 2000, and I struggle to keep up with new stuff. In that attempt I just finished a book called "AI Snake Oil" by Arvind Narayanan. I highly recommend it. It's a good tutorial for the non-tech audience. But as the title suggests, it goes into debunking a good deal of the hype about AI taking over the world. It's filled with real-world examples of AI gone wrong. Best example: several AI companies have spent billions making systems to read resumes (CV) and watch job interviews to then predict the best candidate for a job. Turns out their success is not much better than random chance. Ditto for AI systems that analyze college applications to pick students that will finish a degree with reasonable grades.

The book was published in 2024, so it's current. I got it as an Ebook from my public library.
Logged

KLaban

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2752
    • Keith Laban Photography
Re: Poor old OpenAI
« Reply #4 on: February 01, 2025, 05:09:58 am »

I'm no Luddite. I've already benefited enormously from the use of AI in medical research and development and probably wouldn't be here typing had that not been the case.

Hell, as a creative I've also reaped the benefits of AI learning and do so on a daily basis. That said, there has to be an understanding that this very thing that is benefiting creatives could see the decimation of the industry. The AI industry has to respect creative copyright.

Hopefully I'm preaching to the converted.

Rhossydd

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3457
    • http://www.paulholman.com
Re: Poor old OpenAI
« Reply #5 on: February 01, 2025, 05:17:21 am »

It's filled with real-world examples of AI gone wrong.
I've tried ChatGPT quite a lot recently and I've found it's not to be trusted.

A historical research question gave a hopelessly wrong answer.

I've also tried using it to write Arduino computer code. Occasionally good helpful results, but more often there are errors. Bizarrely when prompted with the error it immediately sees it and repairs it (but sometimes introduces new errors).

I'm sure that AI in specialised researched tasks has the possibility to be of great benefit, but it I get the impression that it needs good oversight to ensure there aren't catastrophic errors.
Logged

JoeKitchen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5080
Re: Poor old OpenAI
« Reply #6 on: February 03, 2025, 09:32:07 am »

I'm no Luddite. I've already benefited enormously from the use of AI in medical research and development and probably wouldn't be here typing had that not been the case.

Hell, as a creative I've also reaped the benefits of AI learning and do so on a daily basis. That said, there has to be an understanding that this very thing that is benefiting creatives could see the decimation of the industry. The AI industry has to respect creative copyright.

Hopefully I'm preaching to the converted.

It's a double edged sword and until it makes it through the courts, who knows what will happen. 

My wife and I use the AI tools in Photoshop all the time and it saves me a lot of time removing things like exit signs and the like.  However, it can be used for nefarious purposes. 

I cant say the brand, but about an year and half ago my wife photographed sandwiches for packaging with a major US client who uses an unique honorific in the brand name.  About a month ago, she was in a local supermarket and saw AI versions of those exact images being used on the store brand sandwiches.  (Patties were switched out but everything else was extremely similar only looking like an AI rendering of the remaining parts of her images.)  We did a Google search on the store brand sandwiches, and, without using that honorific in the query, the search results had it in the metadata.  So it appears the store is ripping off that brand's packaging. 

My wife spoke to an attorney and there are so many issues, including even who is at fault.  Is it the store or the AI company?  But the long and short of it is, until a case makes it to the Supreme Court, there are no precedents to base any lawsuit on, and the cost to take a case that far is outside the realm of a small business like ours. 
Logged
"Photography is one percent inspiration and ninety-nine percent

Rob C

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 25125
Re: Poor old OpenAI
« Reply #7 on: February 03, 2025, 10:34:20 am »

Joe, I expect that many musicians would have issues with the likes of the online music providers, too. Record sales no longer seem to be where it’s at, and I gather that the only people making money are the ones such as The Rolling Stones, who can tour when fancy takes them. For the rest, do it for love, not to get rich. The big players who sell you cheap, perpetual listening of anything you may want to hear, are the new power.

I don’t know how big a part AI plays in their game, but without prompting, they sure seem able to pull up all sorts of stuff similar to what one has been listening to over a period.

Nutshell: artists are for the easy screwing.

Robert Roaldi

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5513
    • Robert's Photos
Re: Poor old OpenAI
« Reply #8 on: February 03, 2025, 12:09:15 pm »

Joe, I expect that many musicians would have issues with the likes of the online music providers, too. Record sales no longer seem to be where it’s at, and I gather that the only people making money are the ones such as The Rolling Stones, who can tour when fancy takes them. For the rest, do it for love, not to get rich. The big players who sell you cheap, perpetual listening of anything you may want to hear, are the new power.

I don’t know how big a part AI plays in their game, but without prompting, they sure seem able to pull up all sorts of stuff similar to what one has been listening to over a period.

Nutshell: artists are for the easy screwing.

None of this is an accident, it's not the result of the markets seeking efficiencies. Somebody with power wanted it this way. This podcast interview with Cory Doctorow from the Canadaland public affairs podcast would probably interest you. It's about how the current monopoly in the arts (mostly music industry) evolved.  https://www.canadaland.com/podcast/bonus-cory-doctorow-knows-why-monopolies-are-killing-art/. As a bonus, he is an interesting speaker so the hour goes by quickly.

My guess is that AI will be a continuation of this.
Logged
--
Robert

Rob C

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 25125
Re: Poor old OpenAI
« Reply #9 on: February 05, 2025, 05:07:16 pm »

Thank you for the link, Robert. I had no idea how little the proportion that The Beatles were able to gather from their work and art. Considering how rich they appeared to be, it’s mind bending to imagine the wealth that the record company must have mined from them. Better be in sport: in today’s local Mallorcan newspaper it says that local boy made good, tennis pro Rafa Nadal, has just sold 45% of his sports company for 85 million euros. No wonder he enjoys nice yachts: probably has to buy them just to save tax!

To have such problems!

Robert Roaldi

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5513
    • Robert's Photos
Re: Poor old OpenAI
« Reply #10 on: February 08, 2025, 09:15:23 am »

Thank you for the link, Robert. I had no idea how little the proportion that The Beatles were able to gather from their work and art. Considering how rich they appeared to be, it’s mind bending to imagine the wealth that the record company must have mined from them. Better be in sport: in today’s local Mallorcan newspaper it says that local boy made good, tennis pro Rafa Nadal, has just sold 45% of his sports company for 85 million euros. No wonder he enjoys nice yachts: probably has to buy them just to save tax!

To have such problems!

This is off-topic for this thread but I wanted to point a series of novels by Cory Doctorow, https://us.macmillan.com/series/themartinhenchnovels, about the fictitious forensic accountant Martin Hench. Hench is a retired forensic accountant who takes on private contracts now and then. You might like the character. He lives in a former rock band's tour bus and has to get up at night to pee. Along the way, Hench points out all the various ways that the powerful distort and take advantage of the modern financial environment to enrich themselves. All too plausible.
Logged
--
Robert

Robert Roaldi

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5513
    • Robert's Photos
Re: Poor old OpenAI
« Reply #11 on: February 19, 2025, 02:19:46 pm »

I did a little digi-scanning of old negatives recently, both B&W and colour. What the world really needs is AI software that fixes dust stops on scans. Now that would make life better.
Logged
--
Robert

Chris Kern

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2204
    • Chris Kern's Eponymous Website
Re: Poor old OpenAI
« Reply #12 on: February 20, 2025, 09:28:46 am »

I did a little digi-scanning of old negatives recently, both B&W and colour. What the world really needs is AI software that fixes dust [spots] on scans.

I've been told by someone knowledgeable in the imaging industry that removal of surface imperfections from scans and prints is an active area of research.  Apparently, it's a difficult problem for machine-learning.

Adobe's Noise Reduction feature works quite well for me, but I suspect that noise from digital sensors is sufficiently uniform to make it detectable by a neural network.  Dust, scratches, bubbles, and chemical marks on negatives occur so randomly that creating an effective training set of scanned images is probably much more of a challenge than creating one for sensor noise.  The tools I've tried that purport to do it have a tendency to introduce so many smoothing artifacts that the result appears artificial—i.e., no longer photographic.

Robert Roaldi

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5513
    • Robert's Photos
Re: Poor old OpenAI
« Reply #13 on: February 20, 2025, 09:38:56 am »

I've been told by someone knowledgeable in the imaging industry that removal of surface imperfections from scans and prints is an active area of research.  Apparently, it's a difficult problem for machine-learning.

Adobe's Noise Reduction feature works quite well for me, but I suspect that noise from digital sensors is sufficiently uniform to make it detectable by a neural network.  Dust, scratches, bubbles, and chemical marks on negatives occur so randomly that creating an effective training set of scanned images is probably much more of a challenge than creating one for sensor noise.  The tools I've tried that purport to do it have a tendency to introduce so many smoothing artifacts that the result appears artificial—i.e., no longer photographic.

Thanks. I'm running out of old negatives that I care about, the ones with the most dust. When I shoot and scan a roll of film these days, a rare thing, I make sure to only remove the negatives from the sleeve once, keeps dirt to a minimum.
Logged
--
Robert
Pages: [1]   Go Up