Language models on their own do indeed have lots of limitations, however there is a lot of potential in coupling them with other types of expert systems.

Absolutely, I even have a dedicated section “Trying to insure combinatoriality/compositionality” in my notes on the topic https://fabien.benetou.fr/Content/SelfHostingArtificialIntelligence

Still, while keeping this in mind we also must remain mindful of what each system can actually do, not conflate with what we WANT it do yet it can not do yet, and might never will.

Image gen did not exist in any way shape or form before.

Typical trope while promoting a “new” technology. A classic example is 1972’s AARON https://en.wikipedia.org/wiki/AARON which, despite not being based on LLM (so not CLIP) nor even ML is still creating novel images. So… image generation has been existing since at least the 70s, more than half a century ago. I’m not saying it’s equivalent to the implementation since DALLE (it is not) but to somehow ignore the history of a research field is not doing it justice. I have also been modding https://old.reddit.com/r/computationalcrea/ since 9 years, so that’s before OpenAI was even founded, just to give some historical context. Also 2015 means 6 years before CLIP. Again, not to say this is the equivalent, solely that generative AI has a long history and thus setting back dates to grand moments like AlphaGo or DeepBlue (and on this topic I can recommend Rematch from Arte) … are very much arbitrary and in no way help to predict what’s yet to come, both in terms of what’s achievable but even the pace.

Anyway, I don’t know what you actually tried but here is a short list of the 58 (as of today) models I tried https://fabien.benetou.fr/Content/SelfHostingArtificialIntelligence and that’s excluding the popular ones, e.g. ChatGPT, Mistal LeChat, DALLE, etc which I also tried.

I might be making “the same mistake” but, as I hope you can see, I do keep on trying what I believe is the state of the art of a pretty much weekly basis.

Might open up a GDPR related issue there. I don’t think people using such a library assume they need connectivity nor that their data would be send to a 3rd party.

What an impressive waste of resources. It’s portrayed as THE most important race and yet what has been delivered so far?

Slightly better TTS or OCR, photography manipulation that is commercially unusable because sources can’t be traced, summarization that can introduce hallucinations, … sure all of that is interesting in terms of academic research, with potentially some use cases… but it’s not as if it didn’t exist before at nearly the same quality for a fraction of the resources.

It’s a competitions where “winners” actually don’t win much, quite a ridiculous situation to be in.

Thanks for the clarification. I checked the code you linked and noticed recognize_google and seems it’s relying on https://github.com/Uberi/speech_recognition which then seems to rely on https://github.com/Uberi/speech_recognition/blob/master/speech_recognition/recognizers/google.py so basically are they using an API, sending all the audio data to Google servers?

I used pandoc for this.

Please come back and share if it’s done better or worst and if so along which dimensions. Quite curious to better understand the differences.

converting audio files to markdown must be a pretty recent feature

Quite curious… does it actually do that and if so how? Because STT to get a plaintext file or subtitle (so with timing) has been available via e.g. Whisper quite efficiently for a while now. If this though does do more, e.g. structure (differentiating a title, list, etc) I’d like to learn how.

Worked well for me, used it for couple of months, maybe an entire year. Re-installed my OS and didn’t put it back.

PS: related, asked on Github too https://github.com/microsoft/markitdown/issues/20#issuecomment-2544630753

audio transcription tool

Thanks for the clarification but I’m a bit confused here, like audio transcription, STT, done by e.g. Whisper? If so what’s the use case? When I think of Office documents audio transcription is not something I have in mind.

soffice works as CLI, can be called from Python and has plenty of related tooling, e.g. https://pypi.org/project/unoserver/ so I agree, I’m confused at what’s actually novel and better than that or even dedicated long lasting FLOSS projects like pandoc.

FWIW if you are interested in such tooling consider also soffice and pandoc which have (as far as I can tell) similar features but have been existing for years now and are not related to Microsoft.

Did too until recently, started to switch to qpdf aqs it seems more openly maintained while doing about the same job with, arguably, clearer documentation than pdftk.

Seems like my time to move away from Authy. Any drop in alternative for iOS? Ideally I could export services and load them back, not manually adding/removing 1 by 1. Even if I can’t though, suggestion still welcomed.

Just ran a VR game for Windows just this morning, worked like a charm, didn’t tinker one minute (using Proton and SteamVR, Valve with NVIDIA, just for context).

Then you also read things like https://www.forbes.com/sites/jasonevangelho/2024/08/21/linux-scores-a-surprising-gaming-victory-against-windows-11/ on non technical websites… and can’t help but wonder if it “will” be easier or… if it’s already done.

The trick is do the opposite, namely bring vim everywhere, e.g using Tridactyl you can bring some behaviors to the browser and, in this very textarea from lemmy, if I press Ctrl+i I get gvim, when I exit it, the content is back in the textarea and I can reply. Vim everywhere.