Using 'ChatGPT' et al. as references...

[Hishi Ryo]

PraNAms!

As a side note.:  How do you reference GPT models accurately?

I frequently see this 'vague' GPT model referencing as it seems to be quite difficult for some to do. Since scriptural references should be accurate, it would be great to do the same with 'GPT' models!

For example: "Chatgpt Translation" as a reference is futile!

What specific ‘type’ of model is that in the first place? Is it the old OpenAI's GPT-4 or 4o, 4o-mini, o3-mini etc. Sonnet 3.5 etc.?

Regarding Sanskrit-English translations.:

The (latest) OpenAI's GPT-4o (2024-11-20)  is ‘okay’ (mostly, not always!). Antrophic's Claude 3.5 (2024-10-22) Sonnet performs better with e.g. Sanskrit based on hundreds of translations!

For reasoning and other tasks (as well as translations), OpenAI's o1 and definitely OpenAI's ‘o1-pro’ model execute far better. Not forgetting the recently released DeepSeek R1 family. With AI translations) becoming more common, it would be highly appreciated to see more specifics! Thanks in advance.

Model specifics:

OpenAI: https://platform.openai.com/docs/models

Anthropic: https://docs.anthropic.com/en/docs/about-claude/models

Google: https://ai.google/get-started/our-models/

DeepSeek: https://api-docs.deepseek.com/api/list-models/

Also helpful:

Ollama: https://ollama.com/search

Hugginface: https://huggingface.co/models

Om Namo Namah!

https://lingosolve.ai | https://ollama.com/hishiryo

[dwa...@advaita.org.uk]

Sounds like you are the expert! It would certainly be fantastically useful for someone to do a systematic comparison. E.g. formulate a series of questions, each requiring specific quotations from PT and Śaṅkara (at least) and score the resultant answers. I have also been told that Perplexity is much better at finding accurate sources but haven’t had the opportunity to investigate yet. https://www.perplexity.ai

 

I have been using the free ChatGPT, which presumably does not use the latest model. It did admit at some point that it does not have access to Śaṅkara Sanskrit source for example and it can obviously only use commentaries and translations that are available somewhere online.

 

I don’t know anything about the other models that you mention. Perhaps you could summarize the differences? (I.e. get one of the models to provide a summary!)

 

Best wishes,

Dennis

--
You received this message because you are subscribed to the Google Groups "advaitin" group.
To unsubscribe from this group and stop receiving emails from it, send an email to advaitin+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/advaitin/CAHXTp9itE37h_x6xkZz27jcKRnswqOS%2B%2B_LqasjuRAzHzHO32g%40mail.gmail.com.

[dwa...@advaita.org.uk]

P.S. I have just used ChatGPT to find the source of Subbu-ji’s ‘neha nAnAsti kinchana’ quote in Br. And KaTha Up. I specified that I wanted the sources ‘checked and validated’ and it got them both correct!!

To view this discussion visit https://groups.google.com/d/msgid/advaitin/004501db75b2%246ab38520%24401a8f60%24%40advaita.org.uk.