Athan Roniou

[Athan Roniou @ARoniou]

@AlexanderKalian Even if AI can do your laundry, change your car's tyres, do the cooking, change a lightbulb and run your business it still doesn't prove AGI. But this is extremely extremely hard for people to comprehend

[Athan Roniou @ARoniou]

@bubbleboi The problem is probably you. It means your workflow is the same, your range is the same, your level of use case ambition is the same. It's all you. Not saying Astra is amazing but if you make the case that it's bad then o1/3 was horrendously bad. People forget

[Athan Roniou @ARoniou]

@iuditg What if you don't care about coding nor 3D modelling which model do you choose? Be honest (chinese models also viable answer just be honest)

[Athan Roniou @ARoniou]

@Da7_Tech Bro you'r way too pro Anthropic and it hurts your credibility.For the highest quality coding apps Fable is probably still king and thats fine. But some people just want an LLM to fill and submit their taxes or identify a bird from a photo,stop choosing teams like this is theNBA

[Athan Roniou @ARoniou]

@Whats_AI @OpenAI Another irrelevant benchmark

[Athan Roniou @ARoniou]

@elshayib_ You are forgetting that by the time they release Gemini 4 we will have Fable 6 and GPT 7, thus rendering Gemini 4 useless by then

[Athan Roniou @ARoniou]

@segyges It's funny but also utter nonsense. Francois always says he's making benchmarks that humans can do easily and AI not + you shouldn't need prior knowledge. The problem with all of you is that you all believe intelligence is a skill metric, which it isnt. But idiots dont change

[Athan Roniou @ARoniou]

@GaryMarcus Gary, you've been fear mongering for over a week now. OpenAI know that they need to focus more on safety now, they are explicitly slowing down further development to focus on safety. All this fear mongering will get us nowhere

[Athan Roniou @ARoniou]

@ravikiran_dev7 1. Google promises alot, delivers very little. We'll believe it, when we see it. They have to start releasing.
2. Google is so slow that by the time we get Google 4 we will get GPT 6.5 / Fable 5.5 but will probably be way ahead. We are routing for Google but its a sinking ship

[Athan Roniou @ARoniou]

@Yuchenj_UW Plot twist: AGI didn't arrive but everyone thought it had

[Athan Roniou @ARoniou]

@fchollet yes singularitarians are noisy but maybe arc agi 3's ability to test generalization efficiency is just too shallow if frontier systems can start saturating it in less than a year. Maybe it needs more depth or breadth or something different to test generalization efficiency

[Athan Roniou @ARoniou]

@theo Are you messing with us? You said its not good for coding and not good for design, the 2 things that matter and that you prefer Fable

[Athan Roniou @ARoniou]

@denon_opg @kimmonismus Fable 5 was 2 months ago. There's a release every 1-3 months. That's a decent gap. Problem is they all decided to release their model this week.

[Athan Roniou @ARoniou]

@haider1 Intelligence ≠ Capability
Intelligence = Generalization efficiency

You could come up with GPT 12 and Mythos 14, automate everything in the world, create cures for the all living things in the galaxy and still not have human level intelligence. Only a few people get this

[Athan Roniou @ARoniou]

@jkbkot @fchollet You are thinking to narrow. Stop thinking about anecdotal human examples thats very shallow thinking. Besides humans have almost identical compute level (noone walks around with a 50lb brain). Memory IS an evolutionary trick to improve efficiency thus improve intelligence index

[Athan Roniou @ARoniou]

@Josh_Criscenti7 @miniapeur Life isnt just about you. Theres others on this planet

[Athan Roniou @ARoniou]

@GutenLeinad @ziv_ravid We keep changing the definition of AGI because noone has come up with a formal scientific definition of AGI. On the contrary, people like you came up with some very shallow definitions initially and it makes dumb people think that we keep postponing the timeline

[Athan Roniou @ARoniou]

@ziv_ravid Haha someone said it

[Athan Roniou @ARoniou]

@kimmonismus First of all you didn't do shit. Second, every prediction about AI is wrong and you think this will be right? And third, we don't even know what AGI means, noone has come up with a formal scientific definition of AGI

[Athan Roniou @ARoniou]

@scaling01 If humans score far more than AI and helps you pinpoint what the bottlenecks are then whats the problem? Or are you so psychotic that you cant fathom humans being AI in anything?