Andy Marlow's banner

Andy Marlow

@AndyMarlowg • 106,387 subscribers

Product Lead for AI creator tools • Often prototyping at 2 AM with sand still on my feet from a late surf • @ iaproactiv partner

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

Most AI OCR tools can extract text from a document. But here's the real problem: How do you know if the extracted information is actually correct? OpenBMB just introduced something interesting with MiniCPM-OCR. It combines two specialized AI models: → MiniCPM-OCR 1.0: A compact 1.3B-parameter model that parses scanned documents, complex tables, formulas, and even tilted or curved pages. → MiniCPM-OCR Jev: A dedicated AI quality inspector that evaluates OCR outputs and helps identify potential errors before human review. And the official benchmarks are impressive: • 93.68 on Real5-OmniDocBench. • OpenBMB reports a 55% improvement in end-to-end processing efficiency with DSpark acceleration under its benchmark configuration. • Jev achieves an AUROC of 0.936 on JevOCRBench, approaching Qwen3.5-397B-A17B at 0.938. The interesting part isn't just extracting information faster. It's building a workflow where AI can parse documents and help determine which results need a second look. Jev doesn't automatically fix mistakes, but it can help make document-processing workflows more reliable. Open-source models, practical deployment options, and an interesting direction for document AI. Try the official global demo: GitHub: Hugging Face: -MiniCPM-OCR 1.0： -MiniCPM-OCR Jev： #MiniCPM #OCR #OpenSourceAI

Most AI OCR tools can extract text from a document. But here's the real problem: How do you know if the extracted information is actually correct? OpenBMB just introduced something interesting with MiniCPM-OCR. It combines two specialized AI models: → MiniCPM-OCR 1.0: A compact 1.3B-parameter model that parses scanned documents, complex tables, formulas, and even tilted or curved pages. → MiniCPM-OCR Jev: A dedicated AI quality inspector that evaluates OCR outputs and helps identify potential errors before human review. And the official benchmarks are impressive: • 93.68 on Real5-OmniDocBench. • OpenBMB reports a 55% improvement in end-to-end processing efficiency with DSpark acceleration under its benchmark configuration. • Jev achieves an AUROC of 0.936 on JevOCRBench, approaching Qwen3.5-397B-A17B at 0.938. The interesting part isn't just extracting information faster. It's building a workflow where AI can parse documents and help determine which results need a second look. Jev doesn't automatically fix mistakes, but it can help make document-processing workflows more reliable. Open-source models, practical deployment options, and an interesting direction for document AI. Try the official global demo: GitHub: Hugging Face: -MiniCPM-OCR 1.0： -MiniCPM-OCR Jev： #MiniCPM #OCR #OpenSourceAI

26,329 次观看 • 5 天前

To test a game idea, a studio builds the level first. The team spends days modeling and coding. Sometimes it drops the idea after the first playtest. Nobody built the level in this video. The model generates every frame in real time from the player's keystrokes. ⬇️

To test a game idea, a studio builds the level first. The team spends days modeling and coding. Sometimes it drops the idea after the first playtest. Nobody built the level in this video. The model generates every frame in real time from the player's keystrokes. ⬇️

22,624 次观看 • 23 天前

Tried Soniox's Compare tool and this is honestly way more useful than just looking at STT benchmarks. I put the exact same recording through Soniox + 3 other providers and Soniox was just quicker to transcribe and more consistent overall in my test. Then when you factor in how cost effective it is compared to a lot of the alternatives, especially if you're processing audio at scale, it starts to make a lot of sense. Best part is you don't have to take anyone's word for it - just throw your own audio in and compare them side by side:

Tried Soniox's Compare tool and this is honestly way more useful than just looking at STT benchmarks. I put the exact same recording through Soniox + 3 other providers and Soniox was just quicker to transcribe and more consistent overall in my test. Then when you factor in how cost effective it is compared to a lot of the alternatives, especially if you're processing audio at scale, it starts to make a lot of sense. Best part is you don't have to take anyone's word for it - just throw your own audio in and compare them side by side:

25,593 次观看 • 1 个月前

Walking CES 2026, the “Smarter AI for All” theme finally starts to make sense in flesh and hardware.Humanoid robots feel less like demos and more like early coworkers—focused on balance, dexterity, and reliability. #CES2026

Walking CES 2026, the “Smarter AI for All” theme finally starts to make sense in flesh and hardware.Humanoid robots feel less like demos and more like early coworkers—focused on balance, dexterity, and reliability. #CES2026

74,075 次观看 • 9 个月前

Walking through CES 2025, I realized I was filtering products very differently than before. Coming from generative media, I’m used to fast output;what stood out instead was how AI glasses, robots, and hardware behaved in real space, over time, with people around them. #ces

Walking through CES 2025, I realized I was filtering products very differently than before. Coming from generative media, I’m used to fast output;what stood out instead was how AI glasses, robots, and hardware behaved in real space, over time, with people around them. #ces

45,243 次观看 • 10 个月前

Unleash the full potential of next-gen communication where Say It. Type It. Show It. See It Happen merges effortlessly with Conversation, Without the Wait for a truly immersive experience.

Unleash the full potential of next-gen communication where Say It. Type It. Show It. See It Happen merges effortlessly with Conversation, Without the Wait for a truly immersive experience.

13,424 次观看 • 3 个月前

没有更多内容可加载