>While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.
What do you mean, as OpenAI employee, you cannot tell that his work has entered the training data ?
But also correct me if I'm wrong, if the two mathematician were really close to finish this problem, and their conversation were used by OpenAI,
shouldn't the Agent have succeeded way faster/efficiently instead of using "4.9 million messages and used about 300 billion output tokens."
I’m not a mathematician so take this with a grain of salt. Apparently terry tao commented that the approach used for the Euler paper can “probably” be used for solving NS but it’s still technically challenging and can probably be done with an LLM with a lot of compute. To me the crux of the issue is whether the insight were stolen so that the problem becomes something that is in the domain of LLMs. This is much different than LLMs coming up with the insight. OpenAI wants everyone to think the LLM came up with the insight and solved the thing by itself even though they have perhaps an army of researchers.
Having the chat logs enter the training data and having them have a meaningful influence on the ultimate result the model produces are very different things.
The text for all the Goosebumps books are certainly in the training data and to some small amount influenced the solve. But their contribution was so vanishingly small it would seem absurd to say R L Stein should have recourse for contibuting to the solve.
The equivalent would be taking a (fully offline) LLM and asking it about the ending of one specific Goosebumps book, and it revealing the twist. And although that specific book was (probably) only once in the training data, a high parameter LLM can usually "remember" the twist.
The only way it would be able to tell you the ending is if it was somehow given more importance in pretraining, loaded into context, or represented in multiple sets of training samples. I have a blog that I make very LLM friendly and usually load posts up into context when I’m working on something relevant. I’ve also opted to improve models for everyone. Despite this, the model can’t recognize my site or any of my posts when I ask it to recall without internet usage (I also turn memory off btw).
That's not correct for a SOTA model with trillions of parameters. Those have immense amounts of knowledge trained into their parameters. Try it. It "remembers" the ending of random books, Goosebumps and otherwise.
That's the entire point of why knowledge cutoff is so important (if you use them offline).
I find it entirely believable that the Navier-Stokes conversation was auto-flagged as high value training data, and burned-in the models knowlegdge base.
"shouldn't the Agent have succeeded way faster/efficiently instead of using "4.9 million messages and used about 300 billion output tokens."" No. Have you seen LLM output? The fact that it managed to converge at all in the problem space is amazing, an LLM is a 'book', the context window is the page number, the page has a token, that's it, that this huge book of weights already contains a method to search a problem space at all and actually go down a 'good' path eventually is remarkable, but it's going to be incredibly inefficient at doing so, a book doesn't have any kind of real-time model building like a brain.
One reason would be support for far more file formats. Currently 2077 conversions are supported in How to Convert (many more to come), while 412 are supported in FileConverter. How to Convert also runs on Mac, Windows and Linux, while FileConverter is only on Windows.
It looks like a great project, but I can imagine it is hard for the maintainer to have time to add these features without any financial compensation. I'm not in a good financial place to make How to Convert open source yet, so am hoping users paying the current $7.5 will help support it's development.
It's a one time fee that gives you the project forever.
Sounds good, also I see that FileConverter needs "Microsoft Office installed and activated"
and you seem to use LibreOffice, +1 really.
On another note it would be great if you could change how you display those 2077 conversions in your website, I find it really awkward to look at them by scrolling and if my mouse get out of the popup it disappear.
If you can do something "similar" to FileConverter where you can see all of them (at least a lot) on one screen, it helps with doing CTRL + F and searching your file format. :)
Also I totally understand not making opensource but marketing your apps as privacy focused but it's closed source is hard to give you trust.
I'm not an expert in analyzing apps, I guess people could use wireshark and see that nothing is coming out of your apps.
The privacy aspect of the app is there are no servers converting the file for you. Everything happens in the browser (in the demo) or using local tools. The demo in the browser's privacy can be confirmed with for example Chrome Developer Tools. For the app, people can confirm no sending of files with tools like wireshark or little snitch, and if people have ideas on how I can improve that transparency, then I'd invite it.
Wouldn't it be a good idea that you make a normal website to convert file format like any others, but the online service serve as an ad for your offline conversion app ?
I'm actually partially implementing that with the demo at the top being able to use WASM tech to run the conversion software in the browser. That way, you can still use a website and convert entirely on your own machine.
I think kids will have a hard time learning and being smart with AI chewing everything for them.
I've read a few stories about parents questioning the over-use of AI from their child, adding to that I've seen my fair share of adult who cannot do anything without asking ChatGPT first.
For example, SDL 3 was officially released last week, which I think is significant. It was posted on HN, but it didn't get enough votes to be visible to me.
While it's clearly content marketing aimed at the hype of DeepSeek, it only mentions Lago in a single sentence.
It's just an article that is aimed to get you to hear about Lago, star their GitHub repository and eventually talk about the "open source" billing tool you heard about called Lago.
(I put Open Source in quotes because I think it's open source version is just Freeware with most features being Call To Action to a paid version. Fair disclosure I have https://github.com/billabear/billabear which is a competitor)
Eh, it's not disguised as anything, it's content marketing. Lago the API billing solution is not suggesting their product as a better alternative to ChatGPT.
To be fair, their ability to target HN is very good. I'm always impressed with their marketing. They put a good title that resonates with HN while the original title is something else for SEO.
2000-2005 is the sweet spot. Cars had immobilizers but not keyless entry. Basic electronics: airbags, electric windows, radio. Engines were naturally aspirated with port injection.
What do you mean, as OpenAI employee, you cannot tell that his work has entered the training data ?
But also correct me if I'm wrong, if the two mathematician were really close to finish this problem, and their conversation were used by OpenAI, shouldn't the Agent have succeeded way faster/efficiently instead of using "4.9 million messages and used about 300 billion output tokens."