Twitxy Saved

ScienceGuru

Verified account This account is verified. Learn more

@ScienceGuruAI

The AI OS for scientific discovery. We run the whole research lifecycle — ideas, experiments, IP, funding, publication. Built for world-class science.
scienceguru.ai Joined January 2025
29 Following 259 Followers 100 Posts

Latest posts 5

ScienceGuru Verified account @ScienceGuruAI
A quick start with ScienceGuru.

Start a chat. Keep your research, code, and data together in one project folder. Customize settings when you need to. Define the goal, materials, and output before work begins. Speak your task, review it, then send.

When it is time to move on, copy a working directory, conversation deeplink, or Markdown and pick up where you left off.

One desktop research workspace for organizing work, defining outcomes, and delivering results.

Download #ScienceGuru: scienceguru.ai
0:56
ScienceGuru Verified account @ScienceGuruAI
HF Trending #1 JEV-27B-VL is in ScienceGuru Desktop 3.2.5.

Visual decisions: routing, ranking, scoring, review. Guru-starred, with Guru Turbo 1.0/1.2.

October first login: 7 free days + 100M tokens. All users. No coupon.

No deployment—inside your research desktop.

scienceguru.ai

What would you have JEV decide?
ScienceGuru Verified account @ScienceGuruAI
Research agents increasingly have to work across papers, scientific software, simulations, and desktop interfaces. In each case, the question eventually becomes: what should happen next?

JEV-9B now adds visual System 1 decisions: screenshot or camera frame in, calibrated probabilities over the available next actions out.

The browser and robot-arm simulations are early demonstrations of a capability that matters for AI-native research workflows: observe, decide, act, then verify.

ScienceGuru is AutoTrust AI’s desktop research environment for Windows and macOS:
scienceguru.ai
AutoTrust Verified account @AutoTrustAI Oct 4
Robot-arm simulation + computer use: JEV-9B now sees.

Now multimodal: image-based System 1 decisions join its calibrated text decisions and System 2 reasoning.

New community-requested visual demos: shopping, settings, mail, and a MuJoCo robot-arm simulation.

AutoTrust leads open decision models.

Model ↓
huggingface.co/autotrust/JEV-…

For AutoTrust AI’s full research-agent workspace, download ScienceGuru for Windows or macOS:
scienceguru.ai
View quoted post on X
ScienceGuru Verified account @ScienceGuruAI
Why this matters for agents: they do not just need a model to describe what is on screen. They need a reliable way to rank the actions available right now.

JEV-27B-VL reads visual state and returns calibrated probabilities across candidate actions. That gives an agent a control signal: act when confidence is high; otherwise ask for another observation, call a stronger model, or hand off to a human.

That is the direction from game state to computer use and embodied interaction: see → decide → act.

Try the AutoTrust AI research workspace, ScienceGuru:
scienceguru.ai
AutoTrust Verified account @AutoTrustAI Oct 4
JEV-27B-VL System 1 sees state, then picks the next action.

RoboCasa: kitchen scenes + tasks. robosuite/OSC_POSE: arm + gripper.
MuJoCo: physics, collisions & rendering.

Drawer. Microwave. Cup to sink.
Then Space Invaders: 40/40, no damage.

huggingface.co/autotrust/JEV-…
View quoted post on X