
Soham Sarkar
@sohamxsarkar
Followers
655
Following
6K
Media
585
Statuses
6K
Prompt engineering reality🪄🧠 ✨Building AGI for work✨
Joined March 2021
Agent Namora AI v1. A multi-agent architecture attempting to mimic AGI for the enterprise worker. Heavy inspo from the following papers.- CoALA.- Gen Agents.- Autogen.- Taskweaver.- RCI.- Voyager. Self correcting outputs, auto updating memory, continuous learning. More to come.
5
4
44
probably will crush code, planning, and tool use. will also likely have really good long horizon consistency and instruction adherence. everything else is a bonus. agi doesn't need to be dostoyevski, it needs to be an autist on adderall. openai has hyped it up a lot lets see.
From trying gpt-5 for the last several hours now I will say:. I cant tell much of a difference between it and o3. It is an always reasoner as far as i can tell. Might feel like a bit bigger model, but smaller and not as good as 4.5 on tasks that arent benefitted by reasoning.
1
0
5
the lack of self reflection is stunning. keep in mind this guy probably has years of exp in designing 3d models or whatever i have 0 clue in this space and i keep getting them to move the goalposts. imagine someone with atleast a month or year of experience.
@sohamxsarkar lmao congrats on learning the difference now good luck making the model actually look good and usable.
4
0
0
thank god for ai cuz no way i'm spending time doing this
@sohamxsarkar @GeorgeCrudo What bar? . You took a 3D render from someone else and had an AI turn it into a 2D image with less detail/quality. The only bar you have is a chocolate bar.
5
0
1
few things bring me more joy than the pure seething of pretentious twitter artists. i hope things get so so much worse for every single one of you hahahaha.
The grifting is getting out of control man. Genie 3 is not a game engine. It has none of the functionality that a game engine has and most importantly, nothing it outputs is "a game" Google Street View is unironically close to being a game than any of these outputs lol.
7
0
3
i need american rw'ers to understand - the internet is not a nation, its not a country, its not your ethno-nat canvas. if you're whining when elon sells satellites to the browns for money i have bad news for you. its about to get a lot worse and there are not deportations here.
1
0
1
unbelievable to see yann take L after L since 2023 man this is specifically what he said didn't work and started jepa
3/ One emergent capability I find remarkable is long-term consistency, especially because we don’t use any explicit 3D representations or priors. Simply training the model to generate the next frame auto-regressively teaches it to maintain physical consistency across time
0
0
1
this shit means something to me man. this is it. we're on the final frontier. we're so fucking close. i can't believe i'm going to live through asi.
What if you could not only watch a generated video, but explore it too? 🌐. Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt. From photorealistic landscapes to fantasy realms, the possibilities are endless. 🧵
0
0
5
it took less than a year.
Genie 3 feels like a watershed moment for world models 🌐: we can now generate multi-minute, real-time interactive simulations of any imaginable world. This could be the key missing piece for embodied AGI… and it can also create beautiful beaches with my dog, playable real time
2
1
15
never. ever. let them forget.
theres a lot of really interesting problems to solve here. like how do we make the AI remember where blocks are placed?. and then you think about it for 2 seconds and the answer to every problem is "program Minecraft normally". so what is the point of AI here. why does this exist.
0
0
0
many are talking about reality sim.
Genie 3 feels like a watershed moment for world models 🌐: we can now generate multi-minute, real-time interactive simulations of any imaginable world. This could be the key missing piece for embodied AGI… and it can also create beautiful beaches with my dog, playable real time
0
0
1