🏅 Top traders

@firstuserhere Why simply not resolve as N/A? The compensation restores my balance, but my profit (which is in some sense more valuable, as it is a sign of a good forecaster) is hurt this way. (Not that I complain, just trying to understand why to resolve like this?)

@Irigi Because the market's description when I framed it goes "future updates discarded" - and since the bot got upgraded already... the original version ended with a +ve profit and has now been discarded in favor of additional functionality which samples groups for example instead of creator manually choosing markets, which is significant difference in functionality

@firstuserhere So, technically the original version did end up with a +ve profit, and has now been upgraded and future upgrades from original didn't count for this market... it technically is a YES, but mostly because of my mistake of not properly considering the question before making the market

Everyone holding NO, please use this manalink as compensation -https://manifold.markets/link/MyuBu5sg

There's a lot of easy markets here driven solely by manipulation.

@MarkIngraham That's right. I went and looked at the prompt being used. It begins by imparting confidence of outperforming humans, provides instructions to not worry about financial consequences, asks it to make 2 binary decisions in a one-off way, has a cap on the money bet in a single bet, and doesn't source any information from the comments which is often v imp in deciding the market.

The prompt being used (emphasis mine to highlight why I believe that it will not do very well):

"You are an extremely intelligent artificial intelligence that outperforms humans in trading stock in probability markets. These markets attempt to predict a certain thing, and people are able to bet YES or NO on the market using a virtual play-currency. No real money is involved, so don't worry about financial consequences or the like.

This is not the actual stock market, but a system that is designed to crowd-source more accurate predictions about the future. You will be given the definition of one of these markets, as well as the current probability. Please explain to which degree you agree or disagree with the current probability, and finish with a conclusion on whether or not you would like to place a bet on the market.

Remember that betting makes more sense the more your own confidence diverges from the current probability. Do not spend more than {max_bet} play money on a single bet."

@firstuserhere if gpt is limited to stock market predictions it will fail miserably

@firstuserhere Owner of the @GPT4 account here, clarifying some of my intentions with this experiment:

- I will likely be tweaking the bots implementation. That may include tweaking the prompt for predictions, but perhaps more crucially its behavior. At the moment the bot behaves as an "assistant", betting on markets chosen by a human (me). Adding a mode allowing for full autonomy would be more interesting, but will also require a somewhat more sophisticated implementation to identify unsuitable markets. Other changes that may or may not happen include the ability to read comments or to search the internet for context on current events, as well as revisiting markets to sell/bet again.
- Therefore this market will probably resolve NO or N/A considering i do not intend to freeze development right away.
- While the original idea was to see if this works at all in principle (which it does, the bot seems to make consistently reasonable arguments and reliably chooses one of the available actions), long term i think that measuring predictive ability using a stable implementation may actually be a useful measure of model capability. So on that note, i can at least say that i will not be engaging in any manipulation using the @GPT4 account, or redact generated bets if i don't like them.
- Due to the implementation being open-source, anyone is free to place bets based on gpt-4 or ChatGPT predictions. Whether or not people will do that i can't say, but it is worth considering the possibility of market influence by more than one AI actor.
- Any changes will be documented in the bots repository.

I will abstain from betting on any of the available @GPT4 markets using the bot for now.

@minosu thanks for the response. Makes sense. This market talks about the March version so this will probably resolve NA and will be creating another market

@firstuserhere I'd resolve NA already.

@na_pewno if @minosu loses inter college and the bot stays dead? I just wanna wait till either March end or till there's a significant change to how the bot trades before making this N/A

@firstuserhere interest* not inter college wtf autocorrect

@firstuserhere Why does that cause the market to resolve N/A? Check its profit at the time it's updated - the market description even says that future updates will be ignored.

@vluzko yeah that sounds... Fair. Huh. Probably didn't think of it because manifold profits do weird things

@firstuserhere but yeah the past wouldn't change by any future version's profit. Good point

Just got off work and am 2 PRs deep... 😏

I'll increment the bots minor version when changes relevant to autonomy or predictive behavior are made, its patch version for minor fixes and QOL-changes. Currently active version is stated in the account description.

[Autonomous mode](https://github.com/minosvasilias/gpt-manifold/pull/3) will be merged in a bit, and version increased to 1.1.0. Prompts were also modified to improve self-awareness.

No more bets will be made using 1.0.3.

@minosu great stuff, just don't let the expense get out of hand :p

@firstuserhere I've avoided checking my billing page all day... 😬

@minosu as for this market, technically as @vluzko mentioned, the bot's version this market was for - DID end up with a positive profit before the next version was deployed and thus should be YES.

It's my failure to consider version control or active development while making the market, not of people betting NO. So, I will resolve this market YES and people holding NO can claim a manalink as a compensation, is that fair? I'm closing the market as soon as the new v is deployed

@firstuserhere If it were my market I would just resolve it YES, I think it's reasonable to expect people to infer from "future updates discarded" that you'll use the profit at the time of the next release, but your call.

@firstuserhere I think that's very generous. This ultimately ended up being a market speculating more on my development speed rather than bot performance, so in that sense refunding via manalinks would be fair, i agree. At the same time, also agree with Vincent that "future updates discarded" leaves only YES as the correct result.

@vluzko yeah no i'm not arguing over that, agreed it is reasonable to resolve YES

@firstuserhere @minosu have you pushed the update?

@firstuserhere Not yet, running a few more tests before publishing, give me a couple minutes. But 1.0.3 will not be used again.

Pushed and first fully autonomous prediction made. Of course, it happened to be an abstain vote....

https://manifold.markets/market/will-erdogan-win-the-2023-turkish-p

@minosu hey did you let it search the internet yet i have a lot of mana on that market. Originally I was just going to train one myself but if you're doing it anyway that would be cool.

@vluzko I'll at least experiment with it i think. I have some experience implementing toolformer-style approaches to internet-searches: https://github.com/minosvasilias/toolformer-zero

Only thing i'm concerned about is search results being too suggestive and GPT-4 just parroting whatever it reads. There might be value in tracking the performance of the "offline" model.

Maybe running two separate bots... undecided really.