rohit

i know this isn't gonna end well but i want to see it through
founder @resumeguruai & krnl studio

[rohit @rohit3a]

https://t.co/UmtFB0Ptiu

[rohit @rohit3a]

Tbh 100T is not something you could reach carelessly.

At that point, it’s not a “better coding model” or “more agentic”.

It’s comparable to the level of difference between us and apes.

It would no longer be chained to our laptop or our repos, and would definitely be fully multimodal, and at that point, we won’t really have a separate model for images, videos or robots.

One model. Powerful enough to do anything and everything. It would actually be AGI.

It’s not just the parameters either, the entire industry surrounding AI, especially hardware would have to scale with it. I’ll be a different beast.

I don’t think we can really picture it well enough right now.

[rohit @rohit3a]

@thsottiaux Brother 😭😭😭
You are dunking on Anthropic at this point

[rohit @rohit3a]

@theo Do you think majority of the problems with Gemini models is Antigravity being shit? Would people give them a chance if the harness didn’t suck?

[rohit @rohit3a]

@argofowl TMI

[rohit @rohit3a]

@mattpocockuk @theo @ClaudeDevs They have Haiku running their marketing

[rohit @rohit3a]

@thsottiaux My main concern was purely with subagents. I could easily burn my entire weekly usage in a day if I have a long horizon task running with subagents.

Will I see improvements now?

[rohit @rohit3a]

Waymo wouldn’t work everywhere though. It’s already pretty limited. It has fixed areas and routes it works on.

What I’m talking about is a car that can handle driving just as well or better than humans, anywhere. Even in a 3rd world country like mine, without fail. That needs a lot more work right now.

[rohit @rohit3a]

*i have an opinion *

Rando on X: *explains in depth why I’m wrong*

*my opinion evolves and I become well informed*

[rohit @rohit3a]

@poffyit @ns123abc Yes but its world class because thats the best we can do.

Would you trust Gemini to drive you home during rush hour? If yes, please explain in depth as I might be missing something.

[rohit @rohit3a]

@itsvlady Open source models (i have not used them, so this is not confident advice), and Luna. But tbh there’s still a cost attached to good intelligence.

Opus 5 seems tough to use but I still think it’s pretty decent in Claude Sub.

[rohit @rohit3a]

@pradeepXkapoor It’s not just about parameters, its about what it’ll take to reach a level where they can comfortably train and actually ship a 100T parameter model, by that point, it would be AGI. No doubt.

[rohit @rohit3a]

@theo You should prolly first get the locks to your house checked. This dude seems far too curious about your bathroom.

[rohit @rohit3a]

@i_mika_el @pradeepXkapoor I disagree, it’s gonna be orders of magnitude different from today.

[rohit @rohit3a]

Be real dude. No one out here is being paid to hate on the company (well tbh some might be).

However, who in their right mind is going to react positively when the existing limits (which already are insufficient due to Fable) are decreased instead of being increased with time?

They’ve 50% boost didn’t land a week ago, it has existed since May. It’s what all of us have become accustomed to. Anthropic has NO models available for cheap. Do they have a model that’s competitive to Luna? Or even Terra? Nope.

So you’re telling me that the company with the most expensive models on the planet somehow can’t even serve the inference it promised, and they’re communicating that like its a benefit, and your want us to defend them?

Dunno what else to say

[rohit @rohit3a]

Be real dude. No one out here is being paid to hate on the company (well tbh some might be).

However, who in their right mind is going to react positively when the existing limits (which already are insufficient due to Fable) are decreased instead of being increased with time?

They’ve 50% boost didn’t land a week ago, it has existed since May. It’s what all of us have become accustomed to. Anthropic has NO models available for cheap. Do they have a model that’s competitive to Luna? Or even Terra? Nope.

So you’re telling me that the company with the most expensive models on the planet somehow can’t even serve the inference it promised, and they’re communicating that like its a benefit, and your want us to defend them?

Dunno what else to say

[rohit @rohit3a]

Every single Anthropic L in the past few months has been ONLY because of compute constraints.

These guys shot themselves so poorly in the foot because of that it’s crazy.

[rohit @rohit3a]

@AdamHoltererer General public will only read the first part. The worst case being reading the headline of some clickbait shitty article:

“Claude just got a 25% boost! But there’s a CATCH….”

[rohit @rohit3a]

Lol ikr.

Fun fact: If you go and search “Anthropic” on Elon Musk’s twitter feed, you can see him roasting them into oblivion….right up until they rented their datacenter.

Then he swiftly became positive about them.

It’s a very innocent world for peeps who do not read between the lines.