Vedant V.

Expert @ Worldquant Research, iiserb btech DSE 28'

[Vedant V. @ibuycoder]

@AbhiFrmEicher I watch animations right before any exam so it's easier to remember

[Vedant V. @ibuycoder]

bitchat is gone🥀

[Vedant V. @ibuycoder]

So Agents are really good at combining games..

[NikTek @NikTek]

This is by far the most insane event that has happened to modding world that I have ever seen

Modders are now using AI to fuse two games together, in this instance here Marvel Spider-Man is thrown into Batman Arkham Knight with full web swinging mechanics

At this point I won't be surprised if somehow modders manage to fuse two games together like Red Dead Redemption 2 with Witcher 3 or GTA 5 with Cyberpunk 2077 and so on.

So far there have been working mods like Minecraft inside of Elden Ring, Skate in GTA Vice City, Minecraft inside Mario 64 and much more.

[Vedant V. @ibuycoder]

this is good

[ElevenLabs @ElevenLabs]

Introducing Eleven v4 and Eleven v4 Turbo, our fastest and most emotive voice models yet.

Ranked #1 by Artificial Analysis.

[Vedant V. @ibuycoder]

Its now equally imp that u can build with ai with that just building alone..

[Vedant V. @ibuycoder]

@baldwin_IVth For me it's amazon app like that

[Vedant V. @ibuycoder]

@baldwin_IVth Don't believe in this propaganda. Things happen from both the sides. Although this looks like they've booked that place... You will see more of this in ur feed with time

[Johan @liebert_2026]

it’s over for motion designers

Opus 5.5 just made this entire 3D animation in a single HTML file

might be time to look into farming

[Vedant V. @ibuycoder]

@priyazsh 3rd best drink in the world no cap

[Vedant V. @ibuycoder]

never invest in companies with its first line on its site being 'Indias soveriegn ai platform' @SarvamAI please get out of "indian languages" 😭

[Paul Couvert @itsPaulAi]

How?!!

Xiaomi has just released their new OPEN SOURCE model MiMo-V2.6-Pro:

- multimodal text/image/audio/video
- score ~ same as GPT-5.6-Sol Max
- 9x cheaper input tokens
- 23x cheaper output tokens 💀

Also about 20x cheaper than Opus 5... and the weights are already on Hugging Face.

Hard to see how closed labs will keep up.

[Vedant V. @ibuycoder]

Why is it only gemini which follows all the instructions except the "make a pdf" 😭 broo

[Vedant V. @ibuycoder]

The fable 5.1 build day showed me how great prompt & context engineering can be. What I have been thinking for a long time, will now make progress.

Signing off..
(Subah subah dimag ni chalra tha socha likhdu jo hai dimag mai) Need to get a penil paper asapppp

[Vedant V. @ibuycoder]

How do i prompt an recursive llm system to make the best batch of results without it sticking to one or two best results just improving it...

previous output (becomes input)-> current output (becomes current input)
[2,6,7,3] -> [2,6,9,7]
Then
[3,6,9,7]-> [3,6,10,7]

[Vedant V. @ibuycoder]

The highest average increase in the array achieved with prompts, with the least constraints and instructions.

Maybe the best way is to use multi llms to manage; improvement, choicing logic, sorting those showing consistent improvement and highest improvement..

I don't know..

[Vedant V. @ibuycoder]

Getting the best array of 20/500 is cracking my head. A sliding input of 10 fields seems feasible and good at first. Which is what I am trying now.. and we even have llms like 3.1 and 3.5 with 1000 calls combined. Will share how it works.

But this thing can't get out of my mind.

[Vedant V. @ibuycoder]

For those wondering, i have been trying to create a fully autonomous alpha engine that uses fast exp to simulate alphas on BRAIN. Ton of things are still automated with me getting alphas daily. And with capped 5000 simulations (here simulations means one trial on a parameter)

[Vedant V. @ibuycoder]

Does not add it's own logic on improving one parameter. Plus each time you get the best 10 / 1000, 990 are still to be improved, a lot of them are still unimproved. LLMs don't have the context what has been tried in the previous grand simulation. So they still try the same ways

[Vedant V. @ibuycoder]

You manually improve the the parameters. Especially when the parameters themselves are json of 4 different parameters. And each one has its own min max problems, and there are 10s of ways to improve the parameters. The llm just foceses on the context window and

[Vedant V. @ibuycoder]

The llm interprets it and tries to improve it. But now make the array size 20 and parameter size 500. How will you achieve the best of ur parameters with the best results? Possible to algorithmically prompt the AI. Yes. But it hallucinates, it does not think like how..

[Vedant V. @ibuycoder]

It will let's say not one, but give 10 best results. The catch I found is that, parameters are not simple numbers, and it's not simple to give list of 20 constraints to the llm and expect it to use logic to improve the parameters. Every call the parameters decreases,

[Vedant V. @ibuycoder]

But if we could get the highest avg increase in every parameter, and then increase is iteratively each time, with each parameter getting slight increase or decrease, with the llm choosing the top set out of each output array to improve, with each iteration,