Kenny Trinh

Founder (1 exit, several misses). Now MSCS @Penn, doubling down on deep tech.

[Kenny Trinh @AnhKTrinh]

@ScienceBriefs that's why you should do it for your own sake, because no body else would care

[Kenny Trinh @AnhKTrinh]

@yongfook you should give mobile a try.

[Kenny Trinh @AnhKTrinh]

@thsottiaux yes sir

[Kenny Trinh @AnhKTrinh]

@yacineMTB you got an answer? I got no reaction :)

[Kenny Trinh @AnhKTrinh]

@tdinh_me btw, who's typing any more? haha

[Kenny Trinh @AnhKTrinh]

@thsottiaux bot traffic my bro

[Kenny Trinh @AnhKTrinh]

@sundarpichai big fan of EmbeddingGemma "1", can't wait to try this

[Kenny Trinh @AnhKTrinh]

@elonmusk Suddenly, the idea doesn't seem so crazy anymore

[Kenny Trinh @AnhKTrinh]

My favorite Jev use case so far

[LangChain @LangChain]

We tested Jev against LLM judges on accuracy, repeatability, latency, and cost to see whether System One models could offer a new approach to agent evaluation. https://t.co/hrqdNpm0g8

[Kenny Trinh @AnhKTrinh]

Great implementation if it works. It does so a lot of problem with the real world implementation of LLM in production. Can't wait to try it out

[Diogo Almeida @CompleteSkeptic]

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions

AFAICT the shortest path to AI-based economic revolution

[Kenny Trinh @AnhKTrinh]

I used an AI agent, Three.js, OpenSCAD, and a 3D printer to turn one sentence into a physical object. https://t.co/6RslSX0laU

[Kenny Trinh @AnhKTrinh]

Very practically steps, It def requires a change of mindset, then process and tooling to be able to move from 1 level to the next.

[Boris Cherny @bcherny]

I talk to engineers at other companies every day and hear the same thing: one person is 10x'ing their output with Claude but the rest of the org hasn't caught up.

Watching teams adopt AI, I keep seeing the same 4 steps.

I mapped them out here: Steps of AI Adoption https://t.co/kQnRAUMKpP

[Kenny Trinh @AnhKTrinh]

A weak prompt says: do this.

A better prompt says: here’s what I’m trying to do, why I care, what I’ve ruled out, what constraints matter, and where I’m unsure.

Don’t just give the agent the conclusion.

Give it the path.

[Kenny Trinh @AnhKTrinh]

The workflow I keep coming back to:

1. capture the messy thought
2. talk it through with an agent
3. extract the structure
4. turn it into the right artifact

Sometimes that’s a post.
Sometimes it’s a task, memo, plan, or decision.

Thought first. Format later.

[Kenny Trinh @AnhKTrinh]

https://t.co/57HbEES3RR

[Kenny Trinh @AnhKTrinh]

Prompting is not magic words.

It is the transfer of context, constraints, standards, and intent from the human to the agent.

If that transfer is lossy, the output gets worse.

This is why I am skeptical that typing with fingers is the final interface for working with agents.

[Kenny Trinh @AnhKTrinh]

Voice with agents looks like a speed upgrade.

But the better benefit is fidelity.

Typing often compresses the thought too early. Speaking preserves the path: uncertainty, constraints, examples, and why it matters.

The agent can summarize later. Lost context is gone.

[Kenny Trinh @AnhKTrinh]

https://t.co/57HbEES3RR

[Kenny Trinh @AnhKTrinh]

One underrated agent superpower is recurrence.

Not “chat with me when I remember.”

More like: check this every morning, notice what changed, draft the next action, and only interrupt me when a decision is needed.

Cron turns an assistant from reactive to ambient.

[Kenny Trinh @AnhKTrinh]

One thing voice changed for me:

I stopped treating messy prompts as a problem.

When I speak to an agent, I include the uncertainty, side notes, caveats, small examples, and the moment where I change my mind.

For a capable agent, that mess is not noise.

It is signal.

[Kenny Trinh @AnhKTrinh]

We try too hard to give agents clean prompts.

Clean is useful, but it can also erase the thinking.

The caveat, the half-formed example, the “actually maybe…” that’s often where the real intent lives.

For capable agents, the mess is signal.

[Kenny Trinh @AnhKTrinh]

https://t.co/57HbEES3RR