Chat GPT 4O getting closer

[Werewolf]

White: Chat GPT 40 with special instructions
Black: Novag Super VIP 1756 Elo (Schach)

[pgn]1. e4 d5 2. exd5 Nf6 3. d4 Nxd5 4. c4 Nb6 5. Nc3 Nc6 6. Be3 e5 7. d5 Ne7 8. Nf3 c6 9. dxc6 Nxc6 10. Qxd8+ Kxd8 11. O-O-O+ Nd7 12. Ng5 Ke8 13. Nb5 Ke7 14. Nc7 Rb8 15. Nd5+ Ke8 16. Nc7+ Ke7 17. Nd5+ Ke8 18. Nc7+ Ke7[/pgn] DRAW.

See below for enhanced instructions. I didn't bother with that but told Chat GPT to come up with improved tracking ideas itself and to impliment them, which it did. The result is the above game.

https://www.youtube.com/watch?v=ybAZ43La9xs&t=339s

With ChatGPT 5 expected within a few months, I expect to see 2000 Elo this year from the LLMs. Not bad considering they aren't built to play chess...

[glav]

White: Chat GPT 40 with special instructions
Black: Novag Super VIP 1756 Elo (Schach)

[pgn]1. e4 d5 2. exd5 Nf6 3. d4 Nxd5 4. c4 Nb6 5. Nc3 Nc6 6. Be3 e5 7. d5 Ne7 8. Nf3 c6 9. dxc6 Nxc6 10. Qxd8+ Kxd8 11. O-O-O+ Nd7 12. Ng5 Ke8 13. Nb5 Ke7 14. Nc7 Rb8 15. Nd5+ Ke8 16. Nc7+ Ke7 17. Nd5+ Ke8 18. Nc7+ Ke7[/pgn] DRAW.

See below for enhanced instructions. I didn't bother with that but told Chat GPT to come up with improved tracking ideas itself and to impliment them, which it did. The result is the above game.

https://www.youtube.com/watch?v=ybAZ43La9xs&t=339s

With ChatGPT 5 expected within a few months, I expect to see 2000 Elo this year from the LLMs. Not bad considering they aren't built to play chess...

I asked to ChatGPT how could it play chess, since, even with massive training, after a while it should be on its own. It acknowledged it and answered that then it uses an engine!

[tmokonen]

I see the ChatGPT bot has been scraping this site. Good luck getting valid info from us, sister. I'm OK with poisoning their well.

[smatovic]

I see the ChatGPT bot has been scraping this site. Good luck getting valid info from us, sister. I'm OK with poisoning their well.

Hehe, people are into the resistance these days

https://www.youtube.com/watch?v=E1CRSvn ... rt_radio=1

Want to mention, now we have the so called BS feedback loop, human datasets since 2022 are poisoned with generative AI, and feeding AI with AI content results in degeneration, they still need to invent some kind of zero approach IMO.

--
Srdja

[towforce]

...feeding AI with AI content results in degeneration...

I've heard this, and I think it's wrong:

1. There's enough human curation of content going on (e.g. each time you use a search engine, you train that search engine: if the answers that come back from the search aren't right, and you reword the question and search again, you may not realise it, but you've actually just trained the search engine!) to keep AI improving even when it's learning against AI generated content

2. Most trading in financial markets is now algo driven - but most of the time, the markets still behave properly (accepting the odd "flash crash")