Richard Jacobs 
 

The AI providers that we are looking in no particular order, include:

• Gemini
• Hugging Face
• Anthropic
• Cohere

What is it about Claude that stands out for you?

Lyds

I'm finding Claude to produce better, more natural results, especially Claude 3.5 Sonnet (and Opus but that's more expensive to use as API) compared to ChatGPT (although the only model that I can select in Studio seems to be a ChatGPT 3.5, wish that was upgradeable to 4o or 4.

I'd also like to see Perplexity as an option, it's already a great research tool, besides API, plus they give out free API credits if you're a Pro subscriber, so those would come in hand.

And a real dream come true would be if it was somehow possible to set instructions for the Claude, OpenAI etc API so you can give specific instructions for specific translations. Not a list of options, but a text box where you can enter instructions as you'd wish.

 Richard Jacobs 

If you don't see the choice of GPT-4, this may be a limitation of your Open AI account. GPT-4 is only available if you have a paid account.

Richard Jacobs 

Richard Jacobs said:

or even better Claude 3.5 Sonnet

As Lydia mentioned we did plan to add LLMs related to Anthropic next... but that part was delayed because it's almost impossible getting an account set up.  We have no problem with any other provider but Anthropic have the worst support I've ever come across and cannot sort out payment for me with the same card we use for every other account we work on.  So as many users here we are also not immune to the internal shenanigans of technology companies!!

So we parked Anthropic for the time being and are currently looking at Hugging Face.  This one is important because it will give us the ability to support local LLMs which are becoming better and better and smaller and smaller.  As computers get faster the performance is also pretty amazing.  Whilst my work laptop (i7, 16Gb RAM) is a little slow handling them, my personal laptop (i9, 32Gb RAM) is amazing.  Faster than using ChatGPT to chat with a local model and I have two larger LLMs installed at the same time I can play with and switch between.  So no need to connect to anything or pay for anything.  For many users a very desirable solution!

If Anthropic ever come back to me with a sensible approach to getting the account sorted we'll come back to it.  In the meantime I'm not too impressed and don't have any desire to waste any more time on trying to chase them.

Oh sorry, I'm always on the latest everything, so latest Studio (2024) version with latest update (I assume since there are no updates available)

I have a paid OpenAI account.

Sorry to hear that, I use Claude only through API and no issues setting up a paid account or adding funds, but might be different since you use it as business I guess.

Found GPT-4o, that's all done. Also have set a bunch of new prompts for me, but I also tried to set an editor prompt to go through the text at the end and check for spelling/grammar errors. But I can't get that to work. Is that something that is possible yet? The prompt causes an error when setting up the project (haven't tried changing it in an existing project).

Edit: got it working in the AI Assistant though but that's another prompt so not connected to the project settings I assume.

 Richard Jacobs 

Richard Jacobs said:

that's another prompt so not connected to the project settings I assume.

All held and globally available.  Not project specific.

 Richard Jacobs 

Yes, you can make the AI Assistant leave the translation source alone and just correct the translation.

The main point is to tell AI explicitly that it is not a translator, contrary to what the system prompt said.

I just resorted to LanguageCloud because I can't get my keyboard shortcuts for Claude to work in Trados/Parallels (I'm on a Mac) for some weird reason, but also because the OpenAI quality is just considerably less than Claude (3.5 Sonnet in this case). Claude produces much more natural translations than OpenAI ever could really. At least right now.

As I'm interested in how this all works, can you tell me why you need Anthropic to sort an account, while it basically is me paying for the API usage? I just have no clue how this all works, and I see plenty of options to use the Claude API in external apps so I'm intrigued how this would work for something like Studio, where you apparently need Anthropic's assistance.

Richard Jacobs 

Richard Jacobs said:

As I'm interested in how this all works, can you tell me why you need Anthropic to sort an account, while it basically is me paying for the API usage?

We need an account and API key to be able to develop the solution!  They don't offer free developer access.

The reason we need their assistance is twofold:

1. they ask you to contact them if you want this to be a Team account... and we would prefer that.  But they never respond.
2. if we use our Company Credit Card they fail it.  The same card we use happily everywhere else.  Definitely their problem but they won't accept that so we're stuck.

If they don't engage properly with us, we cannot fix the access problem, and to be quite frank we have better things to do right now than waste our time with a provider that makes things this difficult!

Does that answer your question?

Richard Jacobs 

Richard Jacobs said:

As I'm interested in how this all works, can you tell me why you need Anthropic to sort an account, while it basically is me paying for the API usage?

We need an account and API key to be able to develop the solution!  They don't offer free developer access.

The reason we need their assistance is twofold:

1. they ask you to contact them if you want this to be a Team account... and we would prefer that.  But they never respond.
2. if we use our Company Credit Card they fail it.  The same card we use happily everywhere else.  Definitely their problem but they won't accept that so we're stuck.

If they don't engage properly with us, we cannot fix the access problem, and to be quite frank we have better things to do right now than waste our time with a provider that makes things this difficult!

Does that answer your question?

Understood. And I know you have better things to do, haha. But I might contact them myself. It doesn't cost you any time, and well, it won't hurt to try, right?

Richard Jacobs 

Please feel free!

Back again! Would supporting Openrouter.ai be an option at all? That would bring a ton of models to Trados really. But of course, I have no clue if that's an option at all for you.

Richard Jacobs 

Richard Jacobs said:

Would supporting Openrouter.ai be an option at all?

Already on our list.

Is Openrouter support on the horizon already?

Richard Jacobs 

I doubt it.  Too many things to do and not enough resource to do them all.  Don't forget you can always do these yourself, or employ a developer to do them for you.  They all use the API.

I know but to be honest it's easier to resort to other tools where I can use the AI models I want, in some way, so I'm actively avoiding using Trados Studio right now unless I really have to. Quality of OpenAI has been shockingly bad lately, despite high quality prompting.