Subraiz

Engineering @perceptroninc

[Codetaur @codetaur]

whew, off work! Im so excited to finally get to try fable. hope nothing comes up that prevents me from enjoying this!

[Kyle Gawley @kylegawley]

I cancelled my $10/mo Calendly subscription and vibe coded my own with Fable for $12,000

[Trev🙃r is hiring DM me! @whatdotcd]

LA: Have your agent talk to my agent.
SF: Have your agent talk to my agent.

[Denis Savenkov @DenXX]

The interesting part isn’t just open-vocabulary detection. It’s treating perception as an iterative process—looking closer when needed, building evidence, and adapting to the task instead of relying on a fixed detector.

[Eric Pence @PenceEJ]

Frontier perceptive models are finally cheap enough to call dozens of times on a single task. This means visual agents are now becoming feasible and new capabilities will emerge.

[Lee (Greater) @shortmagsmle]

Accidentally re-inventing “thinking” from first principles

[Armen Aghajanyan @ArmenAgha]

1/ Today we're shipping Perceptron Agentic Detection. Describe what you want in natural language, or show one example crop, and an agent grounds it in the image. No fine-tuning, no fixed class list.

[Akshat Shrivastava @AkshatS07]

Excited to announce our Agentic Detection API - an API that delivers frontier performance on dense, ambiguous detection tasks.

[Subraiz @Subraiz24]

Can finally settle this debate

[Subraiz @Subraiz24]

We built a new VLM detection endpoint for finding objects in images using natural language categories, examples, or the image itself.

Demo: https://t.co/Xp5z09270N

Docs: https://t.co/hKlEFHVuHq

[Perceptron AI @perceptroninc]

Today we're releasing Perceptron Agentic Detection: localize anything you can describe in natural language or show examples of.

[Subraiz @Subraiz24]

The holy grail of computer vision benchmarks

[Perceptron AI @perceptroninc]

Today we're releasing Perceptron Agentic Detection: localize anything you can describe in natural language or show examples of.

[Perceptron AI @perceptroninc]

Today we're releasing Perceptron Agentic Detection: localize anything you can describe in natural language or show examples of.

[Perceptron AI @perceptroninc]

We're hiring our first DevRel Engineer (aka Hacker-in-Residence) at Perceptron.

Demand for our model Mk1 is outpacing what our research-first team can serve. You'll help us close the loop between our models and the people building on them.

[The Hood Therapist @OhhMar24]

Jalen Brunson got NY folks stepping different for Eid

[Subraiz @Subraiz24]

@aribk24 Sick stuff dude!

[Degen CPA @DrewVento]

BREAKING: Victor Wembanyama has joined Anthropic.

[sophie @netcapgirl]

this is KD joining the warriors for people who know linear algebra

[How To Prompt @HowToPrompt__]

NVIDIA has done the impossible and nobody's talking about it.

They trained a 12 BILLION parameter LLM in 4-bit precision on 10 trillion tokens.

For years, the AI industry has been stuck.

If you wanted to train a world-class AI, you had to use 16-bit or 8-bit precision. Going lower to 4-bit, was a death sentence for the model. It would become unstable, "hallucinate" its own math, and eventually collapse.

But NVIDIA proved that "impossible" was just a math problem.

They used a new format called NVFP4.

Instead of a standard, rigid structure, NVFP4 uses "micro-scaling." It groups numbers into tiny blocks and applies individual scaling factors to each one. It’s like giving the AI a pair of high-definition glasses for its own data, allowing it to see fine details even with 75% less memory.

The result is a total paradigm shift:

- 2× to 3× faster arithmetic performance.
- 50% reduction in memory usage.
- Near-zero loss in intelligence.

The researchers compared the 4-bit model against a massive 8-bit baseline. The curves are identical. On MMLU, GSM8K, and coding benchmarks, the "tiny" 4-bit version performed within 0.1% of the more expensive model.

This is an economic earthquake.

Training a frontier model used to require tens of thousands of GPUs and months of time. NVIDIA just showed we can get the same results with half the hardware and a fraction of the electricity.

[Subraiz @Subraiz24]

@TimTrautman @kvickart @perceptroninc What exactly were you looking for in docs, perhaps I can link you directly to it or shoot you a dm. Right now only video frames. Audio coming very soon.

[Armen Aghajanyan @ArmenAgha]

We're throwing a release party in Bellevue, lined up with MLSys. Thursday May 21, 6 to 9pm; food, drinks, RSVP at https://t.co/kYtBNoiW2H