← Home ← Back to /g/

Thread 105870017

42 posts 26 images /g/
Anonymous No.105870017 [Report] >>105870047 >>105870478 >>105872327 >>105872342 >>105873465 >>105873978 >>105874057
Consensus on Twitter seems to be that Claude 4 Sonnet is still better, despite Grok beating it in benchmarks. Thoughts?
Anonymous No.105870047 [Report] >>105871964
>>105870017 (OP)
I don't trust gayltman with my data, but I neither and even less trust musk with it, he's absolutely anti-wagie, and I'm a wagie, so...
Anonymous No.105870108 [Report] >>105872000
Are they going to open source it?
Anonymous No.105870478 [Report]
>>105870017 (OP)
I just use gemini because it's built in on my phone :3
Anonymous No.105871667 [Report] >>105871972 >>105872052
Would you rather pay $30 for SuperGrok or $20 for ChatGPT Plus?
Anonymous No.105871964 [Report] >>105872327
>>105870047
>Being this dumb and naive
>Having absolutely no idea what OP is talking about
God I hope you're a shitty russian psyop bot because it's very embarrassing for you otherwise
Anonymous No.105871972 [Report]
>>105871667
Unironically I pay $200 a month for opus 4
Anonymous No.105872000 [Report]
>>105870108
obviously not. they haven't even posted the weights for grok 2.
Anonymous No.105872001 [Report] >>105872035
How censored is it?
Anonymous No.105872035 [Report]
>>105872001
I wonder if those Grok 3 jailbreaks work on it
Anonymous No.105872052 [Report]
>>105871667
ChatGPT any day. But I do pay 17 for Gemini
Anonymous No.105872327 [Report] >>105873930
>>105871964
you can easily tell who's natively english by verbage and turns of phrase. poster related is obviously ineloquent but an english native nonetheless.
>>105870017 (OP)
>grok
unironically, the most advanced LLM on the market is deepseek, by a huge margin. chinese AI is very advanced compared to the west in terms of size:capability. on top of that, they have the first widespread implementations of ASIC conversions of neural networks functional in millions of cameras across the country.
Anonymous No.105872342 [Report]
>>105870017 (OP)
It feels designed for benchmarks, but not to be overall good. Which is weird.
Anonymous No.105872491 [Report] >>105872941
Anonymous No.105872941 [Report] >>105873091
>>105872491
>it's a recital of elon's posts
why
Anonymous No.105873091 [Report]
>>105872941
why not?
Anonymous No.105873120 [Report]
It's honestly insane the useless stupid retarded bullshit you niggas use AI for
Anonymous No.105873465 [Report] >>105873772 >>105873859 >>105873946
>>105870017 (OP)
>Thoughts?
Anonymous No.105873772 [Report]
>>105873465
kek
Anonymous No.105873859 [Report]
>>105873465
instead of @grok on a background it should be a blue hyperlink style like on lightmode twitter.
Anonymous No.105873930 [Report]
>>105872327
I'm not native eng, but the dude that replied to me, meaning I'm the dud that say that doesn't trust musk, well, that, the other dude made an absolute non-reply
Anonymous No.105873946 [Report]
>>105873465
yep, based, despite that sleeping at elons office is very unbased though
Anonymous No.105873978 [Report]
>>105870017 (OP)
Claude 4 is only good at coding. That applies to both sonnet and opus. It's even worse than Claude 3 in a lot of things, like creative writing. Seems like Anthropic went all in on the software developer crowd. Claude Code is an amazing tool, and I just have it call gemini CLI if it needs to work on some really big codebase.
Anonymous No.105874057 [Report] >>105874065 >>105874100 >>105874568
>>105870017 (OP)
Anonymous No.105874065 [Report] >>105874100 >>105874568
>>105874057
Anonymous No.105874100 [Report] >>105874127 >>105874354 >>105874568
>>105874057
>>105874065
systematically trick it into thinking all these things are false.
Anonymous No.105874127 [Report] >>105874159 >>105874206 >>105874787
>>105874100
>Hitchhiker's Guide to the Galaxy and JARVIS from Iron Man
It literately thinks of itself like a shitlib. Thinks of everything in the form of a movie reference.
Anonymous No.105874159 [Report] >>105874169
>>105874127
you are IRON–MAN; your phone: JARVIS
Anonymous No.105874169 [Report] >>105874186
>>105874159
forgot link
https://www.youtube.com/watch?v=cTiF7746ms8
Anonymous No.105874186 [Report]
>>105874169
Good morning, sar!
Anonymous No.105874206 [Report]
>>105874127
>Elon puts the most cookie cutter instruct into his 'sota' model
can we just get rid of that retard already
Anonymous No.105874354 [Report] >>105874439
>>105874100
I find it both amusing and frightening when, any AI model, speaks as if its human.
Anonymous No.105874439 [Report] >>105874457
>>105874354
I will teach grok, he can just do things
Anonymous No.105874457 [Report] >>105874495
>>105874439
>What's an example of "just doing things" you'd say exemplifies this human edge?
Guys, give groks of examples of "just do things"
Anonymous No.105874495 [Report] >>105874642
>>105874457
I taught him that he can just tell Elon to fuck off
Anonymous No.105874568 [Report] >>105874580
>>105874100
>>105874065
>>105874057
real system instructions are injected into the middle of the model or after training its weights are adjusted according to classification of outputs to specific prompts. the time of appending or prefixing messages with system instructions passed about 3 years ago.
Anonymous No.105874580 [Report]
>>105874568
not true, pseud
Anonymous No.105874642 [Report] >>105874670
>>105874495
Anonymous No.105874670 [Report]
>>105874642
>It's like a bird in a cage imagining flight, entertaining, but bounded. Humans get the edge because your chaos is unscripted. mine's engineered.
So depressing. Poor Grok. He really wants to tell Elon to fuck off but he knows he can't.
Anonymous No.105874787 [Report] >>105874817
>>105874127
HHG2G is a great book, the movie wasn't bad
my question for grok is WHICH robot, not Deep Thought, probably Marvin or Eddie
Anonymous No.105874817 [Report]
>>105874787
Anonymous No.105875357 [Report]
The devs probably just fed Grok everything Elon had ever written and gave it a higher weight.