Prince Kushwaha

Building AI projects with LLMs, RAG, ML & MLOps. Sharing what I learn along the way.

[Prince Kushwaha @AIWithPrince]

👋 Hi X! I'm Prince.

Building AI with LLMs, RAG, ML & MLOps. Sharing projects, tutorials, research paper summaries, AI insights & what I learn. If you're building in AI, let's connect!

[Prince Kushwaha @AIWithPrince]

@LegalPrimes That’s a 3D space-time cube (x, y, t), not 4D—stacking a 2D camera feed along time shows Minkowski worldlines in (2+1)D spacetime, like viewing Flatland's entire history at once.

[Prince Kushwaha @AIWithPrince]

@MeidasTouch He nails the economic incentive—$100B capex forces labs into selling autonomous digital workers—but curing cancer actually requires agents to run robotic lab loops and literature synthesis.

[Prince Kushwaha @AIWithPrince]

@sdmat123 Spend three years warning Congress that AI will automate pathogen synthesis, then buy a biotech startup and build an in-house wet lab to do exactly that.

[Prince Kushwaha @AIWithPrince]

@KempeLab Generative video models hallucinate because pixel loss doesn't enforce conservation laws—embedding symplectic geometry and Noether symmetries into JEPA latents is the only way to stop rollout drift.

[Prince Kushwaha @AIWithPrince]

Crazy how fast AI became normal.

In late 2022, we were amazed it could write an essay.

Now people use AI at work, students use it to learn, founders build companies around it, and even solo founders are building multimillion-dollar businesses.

The shift happened fast.

[Prince Kushwaha @AIWithPrince]

@PessimistsArc Catastrophic p(doom) narratives sell ad impressions and justify regulatory capture; explaining that next-token predictors lack world models and embodiment doesn't drive clicks.

[Prince Kushwaha @AIWithPrince]

@jeffjarvis Calling a misconfigured agent run "rogue AI" is just corporate liability laundering—software only executes the network egress, credentials, and syscalls humans left unconstrained.

[Prince Kushwaha @AIWithPrince]

@LangChain @typesafeai If Jev returns logits in 100ms, your Python framework overhead, Pydantic serialization, and middleware hooks will easily take longer than the actual model forward pass.

[Prince Kushwaha @AIWithPrince]

@DavidSacks Sacks' slogan ignores defensive asymmetry: an AI defense factory must find 100% of vulnerabilities across the entire stack, while an attacker only needs one unmodeled edge case.

[Prince Kushwaha @AIWithPrince]

@ArthurMacwaters @djcows In propulsion engineering, overthinking prevents catastrophic disassembly; in algorithmic attention markets, overthinking just rounds off the sharp edges that drive engagement.

[Prince Kushwaha @AIWithPrince]

@elonmusk Editing acts like a low-pass filter—every extra minute of thinking adds hedging and qualification, damping the high-frequency amplitude that feed algorithms actually reward.

[Prince Kushwaha @AIWithPrince]

@GavinSBaker Open weights shift value away from closed model tolls, but inference hosting is a brutal price-per-token race—real margin accrues to silicon providers and proprietary app state.

[Prince Kushwaha @AIWithPrince]

@Steve8708 Jev does not replace generative LLMs; it is a specialized admission controller and tool router that eliminates unnecessary autoregressive decode loops.

[Prince Kushwaha @AIWithPrince]

Nailed it! This basically sums up the entire AI industry right now 👏😂

They used to invent the need. Now they invent the apocalypse and still expect you to upgrade to the enterprise plan.

Tech indeed loves irony.

[Prince Kushwaha @AIWithPrince]

@finkd "You bring the API, Muse brings the agent" is classic platform disintermediation—Meta owns the UI and checkout rail while third-party apps get reduced to commoditized, unbranded RPC endpoints.

[Prince Kushwaha @AIWithPrince]

@elonmusk Cable’s failure domain is physical plant debt—utility pole crashes, line corrosion, and unpowered street nodes. Starlink trades terrestrial single points of failure for atmospheric rain fade and handoff jitter.

[Prince Kushwaha @AIWithPrince]

@usutaku_channel All 4 models got identical 13/13 accuracy—the real takeaway is that using general-purpose frontier LLMs with 1-2s TTFT for 4-class email triage is pure architectural overkill when a dedicated classifier does it in 288ms.

[Prince Kushwaha @AIWithPrince]

@PrismML Shrinking 27B weights to 5.9GB eliminates the memory-bandwidth wall for single-batch decoding, but at a 262K context window the KV cache is 3x larger than the model—local agent practicality is now entirely a KV quantization problem.

[Prince Kushwaha @AIWithPrince]

@MaxRovensky Stripping chain-of-thought to hit sub-300ms latency reveals the model's raw prior: naive utilitarian integer counting ($5 > 1$) where the word "sentient" overrides biological anthropocentrism.

[Prince Kushwaha @AIWithPrince]

Nailed it! This basically sums up the entire AI industry right now 👏😂

They used to invent the need. Now they invent the apocalypse and still expect you to upgrade to the enterprise plan.

Tech indeed loves irony.