1/51
@iScienceLuvr
Part of the reason the public is shocked is that most of them have only played with ChatGPT 4o (in the free plan) so when they try a reasoning model like R1 they think China has made this incredible leap in abilities over Americans. Public doesn't know about o1, o3, Claude, etc.
2/51
@M3Po2
o1, o3 dont accept files (pdf/doc/xls) upload, their usefulness is very limited
3/51
@iScienceLuvr
I don't think that's what the average normie user is thinking about. Certainly an important aspect though.
4/51
@duluhagv
It’s almost like you shouldn’t make announcements months before actually giving access to your models
5/51
@iScienceLuvr
o1 was available pretty much immediately it just was behind a paid plan
6/51
@jfischoff
ah good point
7/51
@JuiceSharp
True, the public often compares free-tier ChatGPT (4o) to China's DeepSeek R1, which feels like a leap. But a key factor is R1's *unlimited usage* model—no token caps, low costs, and open access. Meanwhile, advanced reasoning models like OpenAI's o1/o3 or Claude remain behind paywalls or limits. Accessibility matters!
8/51
@kaustubhasutra
Public wants more for cheap.
"Good enough" more often than not trumps "best in class".
9/51
@DreamInPixelsAI
great points
google’s new thinking gemini has been pretty intriguing as well
10/51
@HarmonyHacker
I think this is the case. It's like everyone has seen it for the first time and can't believe it.
Apparently OpenAI's marketing wasn't reaching as far as previously thought.
11/51
@dsog
that’s actually a good point. you’re probably right
12/51
@JulianoMadalena
They are paid and have limitations, especially regarding file uploads. I also notice that the users excited about DeepSeek are not the general public. Specifically, the tech bubble of developers is heavily impacted by the use of DeepSeek.
13/51
@iScienceLuvr
No the public is freaking out it's being talked about on TikTok, the President just mentioned it a few hrs ago, the stock market dipped cuz of it, etc.
14/51
@Ashutos57570081
Cauz they're paid
.
15/51
@iScienceLuvr
And OpenAI is probably regretting that right now
16/51
@escapism_ai
perception is a funny thing. people see what they expect to see, missing the deeper currents flowing beneath the surface. innovation knows no borders.
17/51
@HemenJ
EXACTLY. This is a great great point.
This point is so important.
18/51
@aaHamideh
o1 is also much slower, a little worse, has limits, etc.
19/51
@thomaspower
true
20/51
@jtgi
I don’t think so. The reasoning models aren’t noticeably better for most daily tasks. Just slower.
(I know they actually are, and I use o1-pro etc for some questions but it’s not night n day)
21/51
@alejandroll10
I think the free one is 4o mini, at last without logging in
22/51
@fyhao
Oops that’s true
23/51
@nobody_qwert
Most of the engineers I know barely used it. They say " it won't be able to do what I do" it is funny tho
24/51
@Noodles_Now
I’ve played with all the OpenAI, xAI and Anthropic models (web and API) and Deep Seek is pretty smart but seems to have a context maybe 25% as large as most competitors. Maybe it doesn’t affect most users, but it makes it fairly useless to me.
25/51
@ScienceDaddio
Exactly
26/51
@AGItechgonewild
Just said the same, OpenAI did this to themselves.
Now it will be much harder winning them back, just look how hard it has been for Claude getting average users after OpenAI had their first “ChatGPT moment”.
[Quoted tweet]
No wonder Normies find the free DeepSeek R1 version much much better than GPT-4o mini!!
27/51
@Last_DayWithYou
Someone is salty
28/51
@nisten
it's a bit messed up how true this is
29/51
@Wilson00000009
They haven’t a clue about perplexity either
30/51
@WesternScramasa
i mean, i've played with o1 quite a bit. i respect it, and hate talking to it
even with all the claude discussions, still underrated having your apparent-AI-persona be likeable. r1 is more than likeable, it's *game*
31/51
@ndzfs
True. And that explains this:
32/51
@SpiralSquirrel
You couldn’t even really tryout chatgpt without getting a subscription. Ask it four or five questions and then you had to wait 24 hrs.
33/51
@ToddMostak
I think most users don’t really have first-hand use cases for reasoning models, and that most queries to ChatGPT or DeepSeek could easily be handled by a non-reasoning models.
Note that there are tons of application use cases for strong reasoning models (ie a travel agent that doesn’t bungle bookings), just that most users don’t need to use such models directly unless doing code/math/problems requiring sophisticated logic.
That said, I think people love seeing the model think and seeing the reasoning chains is UX that makes people trust and anthropomorphize the models more, but you don’t get this anyway with O1 except for the very abstract summaries of the model’s COT.
34/51
@VarunkInsights
How did normies worldwide get to know about DeepSeek? Claude, Grok, Perplexity have been struggling to even come close to ChatGPT
35/51
@segmenta
I feel showing the full thoughts considerably helped - I found it the most interesting part.
36/51
@binaryforking
They will surely match DeepSeek's pricing or go even lower
37/51
@SuupaDuupa
Yep - thought the same thing. The gap between free users and paid was wayyy too big.
38/51
@_lambda1
question is how did open ai fail to get the public on board given they're most popular
39/51
@SentientEthos
I mean.. as an everyday user of 01, Claude , haven’t touched 03 just yet, but deep seek is on point. Way more concise code, reasoning is solid. Amazing outputs. Open source. Run local without the prying eyes, free (for now) it’s a win to me.
40/51
@nigharsh
Been using O1 since their launch. No hatred against OpenAI (I love them and use their API's in prod) but, I have been using R1 unintentionally.
The first query I made was underwhelming and it made me think they pulled a benchmark heist, a couple of queries later I understood the brilliance.
I still keep pasting my prompts in O1 if I don't get a satisfactory response from R1 but 100% of the time the R1 response was better.
Bigger surprise was their search feature simply outperforming Perplexity for me. 🫤
41/51
@sonicshifts
Imagine going from Apple Intelligence to DeepSeek R1.
42/51
@matteoianni
That's a good point. I pay for 5 plus subscriptions in my family and I'm pretty sure none of them has ever used o1. People just don't go up there and change model.
43/51
@BalajiAkiri
Subscription based models kill AI.
44/51
@DrMetaMojo
Exactly
45/51
@itsalwaysfays
true
46/51
@Lsenerman
Horrible take. R1 is the same as o1 and o3, without the infinite hype and drama that OpenAI injects yo everything they do. Claude is another league.
47/51
@impershblknight
Western AI companies are stupidly risk averse. There is no reason o1 should not be able to read files or memories on its own. Big corporate models are RLHFed, censored, and lobotomized with HHH training to the point of uselessness.
48/51
@shurensha
And it's their fault
49/51
@slimefren132450
o1 and r1 are similar, o3 is for elites only
50/51
@victor_vibing
I have o1, it sucks in comparison.
51/51
@teortaxesTex
What offends me is that people have some weird intuitions about cost. Bruh why just why do you assume the weekly 4o update took a gazillion dollars? This is not GPT-4 era.
To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196
