PRIVATE ProteusSFX-AI tests

[Eduard Nemeth]

Is he now playing against himself on Lichess? Doesn't he find any other opponents?

[IbaiBuR]

If you wanted to make something like that work (the AI modules I mean) , you'd have to pay for the OpenAI API (It is not expensive, I pay less than a euro each month) and fine tune a model that fits your module requirements, that probably would be a way.

I think it is not that difficult but it needs a decent amount of training data.

Regards,

Ibai

PS: By the way, why did you create this 2 new Lichess BOT accounts:
- Alemor64
- No-Stockfish-Bots

How do you know who these bots belong to? Another 2 bots from ProteusSF , he already has 8 bots, now 10!

It is easy

Alemor64 was his previous human account, but now he upgraded it to BOT account now and created this new human account: Microsmeta.

If you look carefully at the games of No-Stockfish-Bots you will notice that @Microsmeta is always spectating, it uses the exact same CPU as ProteusSF BOTs (arm 8c/8t, basically M1) (but I think it now uses a different one) , you can check with !cpu command. Besides, the BOT plays very similar book lines as his other BOTs.

[AlexChess]

If you wanted to make something like that work (the AI modules I mean) , you'd have to pay for the OpenAI API (It is not expensive, I pay less than a euro each month) and fine tune a model that fits your module requirements, that probably would be a way.

I think it is not that difficult but it needs a decent amount of training data.

Regards,

Ibai

PS: By the way, why did you create this 2 new Lichess BOT accounts:
- Alemor64
- No-Stockfish-Bots

The first YES. It's ProteusSFX-AI Official account. (Microsmeta is now the Lichess chat account, to delete all the PRIVATE offenses on Alemor64 by haters )
Ibaibur I'm happy to have recovered your friendship, you are skilled and fair. Let's discuss about AI modules together I have done some queries (prompts) to ChatGPT but the answers are weak, sometimes reticent and unusable most of the times... and ChatGPT is becoming worse each day X after ProteusSF means eXperimental not eXtreme or eXtraordinaire. Do you know how many users own a mac? I'm curious too because Berserk 12 is the best original engine and Stockfish seems to become every day weaker..

[IbaiBuR]

Are you using the OpenAI API for the queries?

And what are you using? easy_curl with C/C++ or Python witth the requests package and so on?

As I said, the only way I see for it to work is to fine tune an specific model for you AI modules using the API.

You dont even have to do the training, it is made on the Azure Servers as far as I am concerned but I have not tried it that much.

[AlexChess]

Very interesting. I'm using Coursera lessons, Youtube tutorials and ChatGPT 3.5 for the prompts. Since my first tests aren't going well, I've temporary switched to book making . waiting ChatGPT customizable istances. Also my best chess friend Sarona would be skilled enought to partecipate

[Eduard Nemeth]

This guy is really crazy. Soon he has more accounts on Lichess than there are other strong bots. He always complains that he doesn't have good hardware, but he has 10 bot accounts on Lichess. This is crazy, totally crazy. I wouldn't believe it if I didn't see it. Just get away from here, away from this crazy guy!

[IbaiBuR]

This is what I was referring to:

https://platform.openai.com/docs/api-re ... ine-tuning

With this, you could fine tune your own LLM personal model and adapt it to your module requirements I think. It is quite powerful.

There is also a guide here: https://platform.openai.com/docs/guides/fine-tuning

And this is the pricing so far, not so expensive as I said:

[IbaiBuR]

Is he now playing against himself on Lichess? Doesn't he find any other opponents?

It seems so, sadly.

Maybe it was matchmaking because it is automatic. It sometimes happened to me too as I also have more than 1 account. The best thing is to blacklist the rest of your bot accounts in each of your BOTs to avoid it.

[AlexChess]

This is what I was referring to:

https://platform.openai.com/docs/api-re ... ine-tuning

With this, you could fine tune your own LLM personal model and adapt it to your module requirements I think. It is quite powerful.

There is also a guide here: https://platform.openai.com/docs/guides/fine-tuning

And this is the pricing so far, not so expensive as I said:

Yes I know, thx. If we try as a team, I could decide to restart, instead of simply remove -AI fro ProteusSF

[AlexChess]

I'm happy to play against Berserk, why should I block him????? Please don't forward Ediot's messages, OR IS USELESS TO HAVE HIM ON IGNORE LIST . Have you already unbanned him from Leaderboard and from Outskirts?