Alok Ranjan

Digital Humans | Digital Economy | AI Safety | Compliances

[Alok Ranjan @agentspanel]

@typesafeai I rebuilt the idea in the open: OpenJev.

Ran it on a GTX 1650. 4GB VRAM. No A100. No datacenter.

One shared state. One KV cache. One forward pass.

Now the real test: Jev vs OpenJev.

Same weights. Same questions. Latency. tok/s. Calibration.

[Alok Ranjan @agentspanel]

@catocino_ @brainFnCl Spot on!

[Alok Ranjan @agentspanel]

@hackgoofer for sure!

[Alok Ranjan @agentspanel]

@catocino_ @brainFnCl True, used Jev to classify and compact the LLM video pipeline in realtime.
It does use a vision model along side.

[Alok Ranjan @agentspanel]

@GilFeig made a realtime virtual try on with Jev

https://t.co/ciNpDRVMSP

[Alok Ranjan @agentspanel]

Made real time virtual try on with @typesafeai Jev and @DecartAI

[Alok Ranjan @agentspanel]

@omarsar0 made a realtime virtual try on with Jev
https://t.co/ciNpDRVMSP

[Alok Ranjan @agentspanel]

Made real time virtual try on with @typesafeai Jev and @DecartAI

[Alok Ranjan @agentspanel]

@hackgoofer made a real time virtual try on made with Jev

https://t.co/ciNpDRVMSP

[Alok Ranjan @agentspanel]

Made real time virtual try on with @typesafeai Jev and @DecartAI

[Alok Ranjan @agentspanel]

@hwchase17 made a real time virtual try on with Jev
https://t.co/ciNpDRVMSP

[Alok Ranjan @agentspanel]

Made real time virtual try on with @typesafeai Jev and @DecartAI

[Alok Ranjan @agentspanel]

@NFT_Chen made a realtime virtual tryon with Jev
https://t.co/ciNpDRVMSP

[Alok Ranjan @agentspanel]

Made real time virtual try on with @typesafeai Jev and @DecartAI

[Alok Ranjan @agentspanel]

Made a realtime virtual try on with Jev

https://t.co/ciNpDRVMSP

[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

[Alok Ranjan @agentspanel]

Made real time virtual try on with @typesafeai Jev and @DecartAI

[Alok Ranjan @agentspanel]

@wojiaoxiaomayun @typesafeai I did my own RL training with Qwen and got it working.

[Alok Ranjan @agentspanel]

@typesafeai I rebuilt the idea in the open: OpenJev.

Ran it on a GTX 1650. 4GB VRAM. No A100. No datacenter.

One shared state. One KV cache. One forward pass.

Now the real test: Jev vs OpenJev.

Same weights. Same questions. Latency. tok/s. Calibration.

[Alok Ranjan @agentspanel]

@illegalcall @typesafeai Ah! nice one. Speculative Jev context is around 32k, lets see how far i can push this ;)

[Alok Ranjan @agentspanel]

@Redwinelips @typesafeai At par with Jev, releasing accuracy results after internal benchmarking and safety improvements.

I need sleep, haven’t slept for last 48 hours

[Alok Ranjan @agentspanel]

@Dev__Vishwajeet @typesafeai No

[Alok Ranjan @agentspanel]

@EGafni Hire me instead other wise i will share the open sourced version of Jev to @OpenAI
here is how did it local consumer grade GPU and called it OpenJev.

https://t.co/NmZo1NkR4h

[Alok Ranjan @agentspanel]

@typesafeai I rebuilt the idea in the open: OpenJev.

Ran it on a GTX 1650. 4GB VRAM. No A100. No datacenter.

One shared state. One KV cache. One forward pass.

Now the real test: Jev vs OpenJev.

Same weights. Same questions. Latency. tok/s. Calibration.

[Alok Ranjan @agentspanel]

I was impressed by @typesafeai that I created OpenJev @tryopenjev that runs on my 8 year old laptop with consumer grade GPU GTX 1650. Someone tell @typesafeai team that they should hire me or I will go to @OpenAI @AnthropicAI and spill the beans.

@hackgoofer

https://t.co/PRGkXyDhB0

[Alok Ranjan @agentspanel]

@typesafeai I rebuilt the idea in the open: OpenJev.

Ran it on a GTX 1650. 4GB VRAM. No A100. No datacenter.

One shared state. One KV cache. One forward pass.

Now the real test: Jev vs OpenJev.

Same weights. Same questions. Latency. tok/s. Calibration.

[Alok Ranjan @agentspanel]

@hackgoofer You should hire me @hackgoofer other wise i will join @OpenAI and will spill the beans on how I made an open sourced Jev that works on consumer grade GPU GTX 16650 and even called it OpenJev @tryopenjev

https://t.co/NmZo1NkR4h

[Alok Ranjan @agentspanel]

@typesafeai I rebuilt the idea in the open: OpenJev.

Ran it on a GTX 1650. 4GB VRAM. No A100. No datacenter.

One shared state. One KV cache. One forward pass.

Now the real test: Jev vs OpenJev.

Same weights. Same questions. Latency. tok/s. Calibration.

[Alok Ranjan @agentspanel]

@hackgoofer Look ma, I did it again. Real time full duplex voice agent with Jev
@typesafeai

https://t.co/9rBVnHw9OA

[Alok Ranjan @agentspanel]

@typesafeai Jev powering near real time full duplex voice agent with cascaded architecture.

Here's the pipeline:
mic → @DeepgramAI Flux (turn detection) → @typesafeai Jev → @GroqLLC GPT-OSS-120B LLM → @SarvamAI TTS → speaker, with a live @excalidraw canvas on the side.

[Alok Ranjan @agentspanel]

@hackgoofer I open sourced Jev and called it OpenJev that runs on consumer grade GPU GTX 1650, my 7 year old laptop, here is the demo.

https://t.co/NmZo1NkR4h

[Alok Ranjan @agentspanel]

@typesafeai I rebuilt the idea in the open: OpenJev.

Ran it on a GTX 1650. 4GB VRAM. No A100. No datacenter.

One shared state. One KV cache. One forward pass.

Now the real test: Jev vs OpenJev.

Same weights. Same questions. Latency. tok/s. Calibration.