Ancestors

Written by Owen on 2025-01-31 at 18:53

Lotta folks are quite rightly dragging OpenAI for their apparent stance on intellectual property, vis a vis competitors using their site. That's valid, but I think there's a much funnier analysis hiding under it.

OpenAI is telling the world that they missed a revenue opportunity. A big one.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 18:54

What OpenAI is specifically accusing the DeepSeek team of doing, down at a nuts and bolts level, is sending prompts to their ChatGPT products, writing down the responses, and doing statistics on them to produce a model that, under training, becomes likely to give the same answers ChatGPT would.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 18:56

OpenAI gives the general public free, minimally-metered access to ChatGPT. There are rate limits, and you can pay to raise those limits, but when you provide a prompt interface that can be used by any anonymous internet user, you necessarily have to allow anyone who has access to a bunch of IP addresses to use your product for free.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 18:57

OpenAI doesn't do this as a charitable gesture. They do this to build their business. The core proposition of OpenAI as a business is that they have a product which is too capital-intensive for anyone else to run, but which is too useful for anyone else to ignore. Getting people to use the product for free - and particularly, getting people to talk about how good their product is, for free - reinforces that narrative.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Toot

Written by Owen on 2025-01-31 at 18:57

So, it is good for OpenAI as a business to be able to show increasing use of the ChatGPT interfaces, including the free ones. DeepSeek's alleged queries would make a plot of users over time go up and to the right.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Descendants

Written by Owen on 2025-01-31 at 18:59

However, that's only good for OpenAI as long as that growth in usage supports the business narrative! The second DeepSeek said "we can do what they do, but cheaper," they threatened that narrative - particularly, the capital-intensive-nobody-else-can-do-it part.

OpenAI was suddenly left with the narrative that they had, for free, helped a competitor build their product.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 19:00

In a sensible business they would have stopped to ask DeepSeek's team to pay for that access, capturing value in the process and likely also dissuading the team from using ChatGPT in the first place. But they didn't, and I'd speculate that they didn't because it didn't occur to them that the usage growth they were seeing could lead to this outcome.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 20:04

But Wait, There's More, because the thing OpenAI is accusing DeepSeek of doing also shines investor-legible light on a necessary property of LLMs as statistical engines: if you can do statistical analysis on the outputs, you can reproduce the statistics embedded in the engine.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 20:05

So the whole idea of giving everyone in the world access to your AI, secure in the premise that it's a black box that they can't copy, goes out the window. If you want to stop your users from copying it, you need things like contracts, and a budget for enforcing those contracts. Which are all incompatible with the "everyone will use this and we'll collect the rents" business proposition OpenAI is utterly reliant on for their value.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 20:06

There are tons of businesses that can capitalize on this kind of environment - this is Oracle's bread and butter, and Salesforce's, and Microsoft's. "Business-to-business" sales is all about maximizing contract revenue and minimizing contract risk.

Those markets are very, very valuable, but they're also small.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Owen on 2025-01-31 at 20:12

Sam Altman and friends now need their investors to believe that there is a future where OpenAI's products are available to a much wider customer base, in spite of the observed reality that a motivated customer can steal the product through its outputs, and that they aren't about to be pigeonholed into being a B2B vendor selling statistical engines to other businesses.

Or he needs his investors to believe that he can retool a company that is abundantly not doing that, into one that can.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by Alex P. 👹 on 2025-01-31 at 20:29

@owen

gotta admit, "make your own product that proves someone else's product is direly overhyped" is a very powerful move for a hedge fund

=> More informations about this toot | More toots from saddestrobots@jorts.horse

Written by Owen on 2025-01-31 at 20:34

@saddestrobots I'm so here for it.

=> More informations about this toot | More toots from owen@mastodon.transneptune.net

Written by mhoye on 2025-01-31 at 20:50

@owen @saddestrobots "Fund Your Operations By Short-Selling Your Competitors" is gutsy as hell.

=> More informations about this toot | More toots from mhoye@mastodon.social

Proxy Information
Original URL
gemini://mastogem.picasoft.net/thread/113924498328652751
Status Code
Success (20)
Meta
text/gemini
Capsule Response Time
449.949505 milliseconds
Gemini-to-HTML Time
3.044545 milliseconds

This content has been proxied by September (3851b).