Anthony Platanios

VP of Research @ScaledCognition. Previously Semantic Machines (@Microsoft) and @mldcmu.

[Scaled Cognition @ScaledCognition]

Thanks @Nasdaq for the shoutout today 🥳

Read more about our $100M Series A: https://t.co/dnKyK8ZSNn

[Anthony Platanios @eaplatanios]

@dlwh Couldn’t agree more. I often see folks thinking that there’s this one big idea behind a powerful model but in reality, most of the time, it’s about getting many small things right rather than inventing a new big thing.

[Scaled Cognition @ScaledCognition]

"These frontier models are amazing—they're sort of like schizophrenic geniuses. They can create incredible answers, and then you can ask the same question a second time and get a completely different answer that might not even be correct." — @roth_dan, CEO of Scaled Cognition.

Thanks to @Steve_Rosenbush and @WSJ for sharing our story.

Read the Exclusive: https://t.co/velivMfIyU

[Dan Klein @profdanklein]

https://t.co/MFf2RCvEmy

[Anthony Platanios @eaplatanios]

Very excited to announce our Series A led by @vkhosla and @khoslaventures!

We've spent the last couple of years building foundational technology for improving reliability without sacrificing intelligence.

We're only getting started. If you're excited about our vision and want to be part of this journey, come join us!

We're expanding our research team to build the next generation of this technology and you can be part of it too! 🚀

[Dan Roth @roth_dan]

LLMs lie. We build models that tell the truth.
Today, we're excited to announce our $100M Series A led by @vkhosla and @KhoslaVentures.
@profdanklein and I founded @ScaledCognition to solve the key challenge in AI, reliability.

[Scaled Cognition @ScaledCognition]

Earlier this year we brought the Scaled Cognition team together for an offsite to talk about the future of what we're building.

The energy was something else.

Since then the team has grown a lot and we're not slowing down. We're building super reliable intelligence — AI with the fluency of the best LLMs and the reliability to handle the conversations enterprises can't afford to get wrong. Moving money, rebooking flights, making decisions that can't be undone.

🚀 We're hiring — come join an incredibly talented team of researchers and engineers.

🔗 https://t.co/m40S0wFfCJ

[Together AI @togethercompute]

@profdanklein spent 20 years studying how language forms intelligence. When LLMs exploded, he saw something everyone missed.

These systems were fluent, confident, and wrong.

And no one could tell the difference. 🧵

[Erfanzar @eraznafre]

Releasing

SpectraX is a JAX-native neural-network library built around true MPMD pipeline parallelism. Each physical rank compiles and runs its own XLA program — no shared shard_map HLO, no SPMD-same-shape constraint. Heterogeneous stages (eg, embed → blocks → head), nine pipeline schedules (GPipe, 1F1B, ZeroBubble, Interleaved, DualPipeV, …), and a unified https://t.co/vYOljO1K4k()/spx.jit() entry point that dispatches to SPMD or MPMD from the same training script.

https://t.co/GWPCsQVUwI

[Anthony Platanios @eaplatanios]

@ysu_nlp @ZhitingHu @NeoCognition Congrats Yu Su! Wishing you the best of luck and looking forward to what you build! :)

[Scaled Cognition @ScaledCognition]

Every other frontier model still can't solve for hallucinations. The first question people ask us: how can you guarantee what your flagship model APT-1 will and won't do? Here's what Dan Klein had to say at HumanX: https://t.co/cPGBy2PjUC

[Scaled Cognition @ScaledCognition]

Our CTO Dan Klein joined @jeremyutley and @werdelin on Beyond the Prompt AI to dig into one of enterprise AI's most underrated problems: systems that are confident, fluent, and wrong.

The good news: with APT-1, fluency and reliability aren't a tradeoff. You get both.

https://t.co/07GQ1H8Ya8

[Scaled Cognition @ScaledCognition]

Prompting and praying is not a deployment strategy.

Most AI agent demos look impressive. The production reality? Endless edge cases, policy conflicts, and a hill-climbing battle that never ends.

Last week at @HumanXCo 2026, our CTO Dan Klein announced AgentTwin.

We don't build agents anymore. AgentTwin does. Your transcripts, policies, knowledge bases — everything that captures how your business actually runs — AgentTwin ingests all of it and ships a production-ready agent in a day.

But speed is only half the story.

APT-1, our flagship model, cracks something the industry hasn't solved: the determinism to never hallucinate and take controlled actions — with the natural fluency of an LLM. Not rigid rule trees. Not prompts and prayers. Both, at once.

Watch the full talk — Dan Klein and @vkhosla on why super-reliable intelligence, powered by AgentTwin and APT-1, is the real unlock for enterprise AI adoption: https://t.co/6myvKE48eQ

[Scaled Cognition @ScaledCognition]

Our CTO Dan Klein had a great conversation with @vkhosla at @HumanXCo last week, moderated by @AnitaRamaswamy of @theinformation. The question that keeps coming up: not how smart is a model, but can you trust it to act? More thoughts from Dan about super-reliable intelligence here:
https://t.co/UGGNy2481h

[Scaled Cognition @ScaledCognition]

The AI industry has a problem it doesn't want to admit:

"The models people are mostly interacting with today are assembling plausible outputs on the fly. The bottom line is — we have built not truth engines, not reliability engines. We've built plausibility engines."

That's our CTO Dan Klein, speaking plainly about why most enterprise AI prototypes die within three months — and why the frontier models everyone is building on are the problem, not the solution.

He'd know. After five years at Microsoft watching brilliant demos get killed before they could ship, Dan co-founded Scaled Cognition to build something different — a model architected around reliability, not plausibility.

APT-1 isn't another plausibility engine. It's built from the ground up to take actions with structural guarantees — not guess at tokens and hope for the best.

Dan recently chatted with @ConorBronsdon of @chain_ofthought to get into all of it — the demo-to-production gap, why multi-model checking produces correlated errors instead of reliability, and why the current AI S-curve is leveling off faster than people think.

Worth a listen 🎧 https://t.co/lDDHg6Y95A

[Scaled Cognition @ScaledCognition]

The AI industry is racing toward ever-more-powerful models. But for enterprises, raw intelligence isn't the bottleneck — reliability is.

Tomorrow on the Main Stage at @HumanXC, our CTO Dan Klein joins @vkhosla for "Super-Reliability vs. Super Intelligence" —  exploring why the next frontier isn't smarter AI, it's AI you can actually trust at scale.

April 8, 4:20 PM: https://t.co/WIBYONdFXn

[Anthony Platanios @eaplatanios]

@percyliang @Diyi_Yang Congrats Percy! 🎉

[Scaled Cognition @ScaledCognition]

The failure mode is a property of the model, not the task. Every LLM wrapper calling itself an enterprise agent inherits these from the foundation up. An orchestration layer doesn't fix an architectural problem. It's why we built Super-Reliable Intelligence.

[Scaled Cognition @ScaledCognition]

Our CEO @roth_dan on the keynote stage at @Genesys Inspire 2026 alongside CPO Olivier Jouve, walking through how APT-1 is powering the new Genesys Cloud Agentic Virtual Agent — the industry's first virtual agent built on Large Action Models for enterprise CX.

The future of agentic CX is here. And we're just getting started.

[Scaled Cognition @ScaledCognition]

There's more AI being built for customer experience than ever before. The bar for what actually works in production without hallucinating is still remarkably low.

And there's a reason for that. A lot of the solutions out there are no more than a handful of API calls and prompts on top of a general-purpose LLM. They generate responses but can't reliably take action, and they fall apart the moment real customers are on the line.

At @ScaledCognition, we took a different approach. We built large action models from the ground up for enterprise CX. Purpose-built for reliability, transparency, and autonomous decision-making.

We're proud to be working alongside @Genesys, the global leader in CX, to power their agentic virtual agents with our large action models.

https://t.co/vOXMWhuuWk

[Scaled Cognition @ScaledCognition]

Most AI models speak language natively and learn action later.

That’s why so many enterprise AI deployments break in production.

Forbes featured our CEO @roth_dan today on how APT-1, our Large Action Model, is built differently—designed for verifiable execution, not just fluent conversation.

Proud to be building with @Genesys to bring the most reliable agentic CX to the world’s leading enterprises.

https://t.co/aEqEyJqhCI