Will there be any large language model I newly use daily by July 1?
39
347
770
resolved Jul 1
Resolved
NO

This would include e.g. if I started using ChatGPT or Bing's chatbot every day. It doesn't include e.g. if some broader service I use comes to make use of a language model in a subtler way. e.g. if Gmail uses one for suggested autocompletes that doesn't count.

I won't bet on this market. I don't know of any large language model that I currently use daily.

ETA: by 'daily' I mean roughly every day, not necessarily literally every day.

Get Ṁ200 play money

🏅 Top traders

#NameTotal profit
1Ṁ44
2Ṁ28
3Ṁ22
4Ṁ20
5Ṁ20
Sort by:

ChatGPT is the top contender. Before checking, I decided to tentatively say yes if I used it on 10 out of last 14 days. I counted about three in the past 14 days though could be wrong. More confidently, ChatGPT says one in past 7 days.

bought Ṁ10 of YES

Added you to my Claude account. You can also try it on https://poe.com. Claude is better at reasoning through boundaries and emotions, flexible reasoning, and generally superior to ChatGPT in reasoning, though not superior in knowledge and coding. Try Claude instead of ChatGPT for a bit :)

@ampdot Are you talking about GPT-4 chatGPT? And claude+?

predicted YES

@firstuserhere Claude 1.3 ("Claude+") is superior to both ChatGPTsx in many domains because of the way it was aligned.

@ampdot Signed up to Claude but can't get it to do anything other than this

predicted YES

@KatjaGrace Just refresh a few times, sometimes the console has issues

@ampdot i guess i will give it a shot. I did compare gpt4 and Claude+ on a set of coding related tasks. Gpt4 was way superior imo. I also tried experiments with providing code and asking to generate in depth documentation for some functions, removing the functions, clearing the context, and asking to recreate those functions from description. Gpt4, for python, Scala, and cpp was able to produce working code in decreasing #of times but outperformed Claude+. But my sample size is small so..

Anyway, will try on those type of tasks you mentioned

predicted YES

@firstuserhere I already stated Claude+ wasn’t superior for coding.

By the way, Katja, you can get Claude+ on web by subscribing to Poe.com now.

@ampdot ah i see, glossed over that bit in your original comment lol.

@ampdot
In general I find:

For creative writing: GPT4 = Claude+ > Claude-instant > ChatGPT
For logic/general intelligence: GPT4 > ChatGPT = Claude+ > Claude-instant

bought Ṁ10 of YES

Fwiw I (and I believe much of the Manifold team) use Codex roughly daily

bought Ṁ10 of NO

"newly" -> "nearly" in the title, I assume.

@harfe I meant 'newly' actually—just to exclude any that I currently use daily without noticing. Though by 'daily' I also mean 'roughly daily' not literally every day.