Resolves as YES, if there’re more than 4 different companies providing general chat experience without major limitations with more than 10 million monthly active users in December 2023**. There should be a free tier allowing more than 10 chats, 50 questions a day, and available without the "waiting room" at least 16 hours a day on average.
**EDIT: We'll use "monthly visits". Yes, this is a major change to the conditions. But it seemed fair. I accept and will consider any objections. See https://manifold.markets/ValentinGolev/will-there-be-more-than-4-major-ana#OQV9s1bDjiMxj4zArZmp
I will use https://www.similarweb.com/ for traffic estimates.
Benchmark for “general chat experience” is ChatGPT as it currently is (it was 3.5 at the time of the market opening), example of “Major limitations” is Bing Chat’s “five questions per session” limitation.
Current tally/examples:
ChatGPT: generally available for free, counts
Poe: includes powerful models, generally available for free, counts (we're counting "companies providing access", so free gpt-3.5 through Poe is fine)
LLAMA & analogues: not powerful enough
Claude: included in Poe, but doesn't count separately as it's not generally available by itself
Character.ai: generally available. if someone shows me something that reasonably argues that it's close in power to ChatGPT-3.5, i'll consider it. doesn't count yet
Bard: I will continue to consider restriction to americans-only as a "major limitation", and I hope one day it'll go without saying. doesn't count
Bing AI: every time I try bing ai, it refuses to help and is annoying in many ways. this IMO is a major limitation. I was lucky to try bing chat when it was more helpful but it won't count anymore. (feel free to show me something that proves that it's changed). doesn't count
LLaMA 2: someone should show me that someone serves it for free for the general audience, or that the "powerful" version in question (70B?) can run at least on my very good laptop (M2 Pro) with the speed that still feels like a chat experience. otherwise it's a "no"
Pi.ai I think is a "yes", unless benchmarks prove otherwise, and if it survives as a free, generally available tool in december 2023, and gets some users. ( https://www.similarweb.com/website/pi.ai/#overview says 143k visits... too low)
so there's 2 actual, 1 questionable, and 2 potentially but not actually available
Related questions
the description said "more than 10 million monthly active users in December 2023" which I think was pretty wild on my part. I'll replace it by "10 million visits that month according to https://www.similarweb.com/website/pi.ai/#overview " if noone objects. there's a "94% chance" and I still hold a "NO" (obviously not because I'm smart), and that would be a move strongly against myself and for most of the holders, so I hope it's OK?
@ErickBall because i don't think anyone really noticed this and it feels like too much of a gotcha at this point?
one objection is enough for me though
the point of this question is economical. it's not about different technologies, but different business "models", so to say. basically "are there 4 different actors providing free and great LLM experience, at their own expense?" (see the first line of the description).
I will not consider "LLaMA 2" as a contender, unless either:
someone will show me that someone serves it for free for the general audience
the "powerful" version in question (70B?) can run at least on my very good laptop (M2 Pro) with the speed that still feels like a chat experience
@firstuserhere talking to it currently, asking it to code something, it keeps asking questions, like a clueless consultant over his head lmao https://pi.ai/s/FFuYbFb14YpJenrYRsqBe but OK i guess believe it. Pi counts if it's still around at the EOY 2023
Bard, bingAI, characterAI, Claude, etc. off the top of my head, of which the first 3 have >10mil monthly users iirc. Someone fill in what I'm missing
@firstuserhere please show me something to prove that character.ai is functionally analogous to chatgpt.
i don't have access to bard. i will continue to consider restriction to americans-only as a "major limitation" which for me always went without saying
every time i try bing ai, it refuses to help and is annoying in many ways. this IMO is a major limitation. i was lucky to try bing chat when it was more helpful but it won't count anymore. (feel free to show me something that proves that it's changed)
@ValentinGolev is bard still restricted to the US? I'm not in the states and have had bard for quite some time. Hahaha, i get what you mean about Bing being massively irritating, it straight up refuses half the requests and even then isn't so good anymore compared to other gpt4 implementations. Anyway, yeah there's half a year left for that to, for perplexity to gain a few more users, to share some stuff on character.ai etc, let's see.
@AnT yeah i know qualitatively its up there; thanks for sharing. Reminds me that our @Mira had created a character called "Pride" which was fun to talk to; @ValentinGolev you can do so here
@Feanor I'm in Germany, it's not on the list. Though obviously I wouldn't expect, say, Russia to be on this list even when it's released for almost everyone. I think it'll be clear when it's a honest "mostly available throughout the world"
@AnT thank you, OK. I wish there were, like benchmarks or something. when it's the only one missing from the list I'll try it myself for my typical uses
Everyone seems to forget character.ai, which I'm pretty sure meets all the criteria already.
@ValentinGolev yeah, it's especially good for conversations... probably less good at programming.
Why is the close date set to September, shouldn't it be the EOY?
What about open-assistant? What about https://huggingface.co/chat/ ?
How are we counting open source models? How about LLaMA derivatives?
Why is the close date set to September, shouldn't it be the EOY?
To avoid people buying the obvious answer. I want to stop trading when the stakes are still high.
What about open-assistant? What about https://huggingface.co/chat/ ?
Do open source models count?
How about LLaMA derivatives?
I don't think any of the small open-source models are analogous to GPT 3.5 in power, not the ones I saw so far. Will there ever be a bunch of open-source models? Probably. By EOY 2023? Not so sure.
How are we counting open source models?
Good question - we'll count base-models. LLAMA, Alpaca, and all of their siblings would count as 1, if it was even roughly as good as gpt 3.5.
