Ankit Srivastava

GenAI Architect | Claude Certified | Open Source | Entrepreneur | Consultant

[Ankit Srivastava @genai_ankit]

One more observation, I was running Fable at max and it came up with some findings which I gave to grok 4.6 (medium) and to my surprise grok found gaps and some wrong assumptions/findings which I gave back to Fable and classic Anthropic style it said sorry about the miss and corrected itself.

[Ankit Srivastava @genai_ankit]

@kristaletz Lets see what “Originates”

[Ankit Srivastava @genai_ankit]

@cgtwts Must be a vibe-scientist built a time machine and got stuck in past trying to access chatgpt to ask build me wayback to my timeline part 😉

[Ankit Srivastava @genai_ankit]

@weswinder The next wave is going to be ACP powered Orchestrators which combines multiple agents for a more smarter work.

[Ankit Srivastava @genai_ankit]

@designwkarthick You will be running out of credits if you have a workload worth $200 plan of codex.

[Ankit Srivastava @genai_ankit]

@TTrimoreau Not entire product but I feel Antigravity has best Chrome Controls, it feels fast, smooth and correct.

[Ankit Srivastava @genai_ankit]

Gemini 3.7 Flash is a good model for lite workloads. I use it along with Grok 4.6 as Proposer, Fable (low) as reviewer and Gemini 3.7 Flash as executer for UI.

I have SuperGrok -> Runs entire week.
I have Google AI Pro -> Often reach the 5 hrs window quickly.

Google, why?

[Ankit Srivastava @genai_ankit]

@jun_song Leaving us to only use Fable

[Ankit Srivastava @genai_ankit]

@Im_IrushiK Grok.
When GPT Sol launched it was doing great but now I do not trust the facts it comes up with.

My experience with Opus 5 is almost same, can't trust it.

It is only Fable at High, XHigh and Ultra that I trust reliably but most of the times I use Grok.

[Ankit Srivastava @genai_ankit]

@bindureddy The benchmarks are meaningless. I have seen how for some of my use-cases a less intelligent model just gets it right and for some the different settings of a big model like Medium for Sol and Fable just gets it right.

[Ankit Srivastava @genai_ankit]

@iuditg Can you share more details about what actually did? Because I am a supergrok $30 user and I code all day long and generate images and 480p videos but even still my quotas are just fine.

[Ankit Srivastava @genai_ankit]

@bindureddy Let's see how the 4.6 comes out not completely tested yet.

[Ankit Srivastava @genai_ankit]

@bindureddy For me Grok 4.5 has worked at places where Opus 5 was simply hallucinating, I also used Sol but that model becomes too risky to trust as context grows. Most of the time when I point out mistakes or wrong conclusions they are like "You are right to q that.."

[Ankit Srivastava @genai_ankit]

@NovaXCode Really disappointing model, use fable low or grok if costs are concerned

[Ankit Srivastava @genai_ankit]

If you're really worried about code monitoring while working with AI, write a skill to generate a quiz after each session with a mandatory passing score for applying the patch

[Ankit Srivastava @genai_ankit]

@KellehEyad I have a setting to manually set the times they can go back anf forth but usually Round 2 is a safe spot with sometime Round 3 if grok reverts back with disagreement.

[Ankit Srivastava @genai_ankit]

@KellehEyad it's a WIP so ignore some here and there UI issues

[Ankit Srivastava @genai_ankit]

@KellehEyad I run my own Orchestrator, using ACP to combine Grok Build and Codex and it break downs tasks into tickets then back and forth reviews among Grok and Codex and once agreed moves forward

[Ankit Srivastava @genai_ankit]

@bindureddy Harness for coding tests?

[Ankit Srivastava @genai_ankit]

@elonmusk A new paradigm is what the world needs now. Hey @elonmusk can you setup a team for this?