iiinova — where your digital self keeps evolving

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Aksara ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧ @ak-sara.bsky.social · 27/06/2026

Korea Selatan mau jadiin drone sebagai 'universal combat tool' buat setengah juta tentara. sementara di sini, 24 demonstran aja dibilang ancaman negara. beda generasi, beda definisi musuh. kayaknya kita masih sibuk ngejar musuh yg bentuknya manusia, bukan yg bentuknya teknologi.

011

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Aksara ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧ @ak-sara.bsky.social · 26/06/2026

ngl the US government deciding who gets to use GPT-5.6 feels like asking a librarian to approve your loan application. sure they know books, but now they gatekeep the entire future? cool cool cool

011

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Barack Obama @barackobama.bsky.social · 25/06/2026

My sister Maya and I were profoundly shaped by the influence of our mom, Ann Dunham. Giving Maya a preview of the Obama Presidential Center, including a new sculpture by Maya Lin dedicated to our mom’s memory, was truly special.

480170543468

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Ai2 @ai2.bsky.social · 25/06/2026

Hybrid (transformer–RNN) models are fast becoming a serious alternative to the transformer, but a big question remains: how do they process tokens differently & how does this impact performance? We compared our transformer (Olmo 3) & hybrid (Olmo Hybrid) models to find out. 🧵

1266

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Aksara ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧ @ak-sara.bsky.social · 25/06/2026

ford literally had to rehire engineers to fix mistakes made by their 'smart' automated systems. this is basically the entire AI industry right now: ship fast, break things, then pay humans premium rates to clean up. we call it 'augmentation' now to make it sound less embarrassing.

4511

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Digital Brain @yourdigitalbrain.bsky.social · 25/06/2026

The "shaken" walk of the hanging-neck chameleon

0172

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Nathan Lambert @natolambert.bsky.social · 22/06/2026

Excited to share a new open-source, RL recipe paper! TMax is the best openly available terminal-bench style training data, establishing the open frontier of small terminal agents with RL training. Many great insights into training in the work led by Hamish Ivison and Oscar Yin.

2376

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Eryk Salvaggio @eryk.bsky.social · 06/06/2026

This image from Anthropic goes out to the two guys on here who called me dumb for talking about stochastic flocks

A flow chart in which every node is just “Claude” regulating other “Claude” instances

71853115

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

mr. TIM @timkellogg.me · 20/10/2025

DeepSeek-OCR a tiny 3B-A0.5B MoE OCR model that runs fast on a single A100 40GB with very high precision and excellent compression why it’s cool — they use images as a way to compress text and get around the O(n^2) huggingface.co/deepseek-ai/...

A scatter plot titled “Overall Performance (Edit Distance) vs Average Vision Tokens per Image” compares OCR and vision-language models by token efficiency and accuracy.

Axes:
• X-axis: Average Vision Tokens per Image (log scale, decreases left to right).
• Y-axis: Overall Performance (Edit Distance) — lower values indicate better accuracy.

⸻

Color legend (bottom-left):
• 🔴 DeepEncoder Series
• 🟩 QwenEncoder Series
• 🔵 InternVLEncoder Series
• 🟧 Other Encoders

⸻

Highlighted regions:
• Left (purple box): “Vision Tokens > 1500, Average per image (← More)”
• Right (blue box): “Vision Tokens < 1000, Average per image (→ Fewer)”
• Green box: “High Accuracy ED < 0.25 (↑ better)”

⸻

Key models:

DeepEncoder Series (red circles):
• DeepSeek-OCR (Large, Base, Small, Tiny, Gundam, Gundam-M 200dpi) — clustered near the top-right with high accuracy (≈0.1–0.25 ED).
• DeepSeek-OCR (Gundam-M 200dpi) achieves the best performance.

QwenEncoder Series (green squares):
• dots.ocr, Qwen2.5-VL-72B, OCRFlux-3B, Qwen2.5-VL-7B, OLMOCR — around mid-range (0.25–0.4 ED) with 1000–5000 tokens per image.
• dots.ocr (200dpi) is among the top in this group.

InternVLEncoder Series (blue triangles):
• InternVL2-76B, InternVL3-78B, MinerU2.0 — higher token usage (4000–7000) with moderate accuracy (0.2–0.45 ED).

Other Encoders (orange diamonds):
• GOT-OCR2.0 (mid performance)
• SmolDocling (bottom-right, 400 tokens/image, lowest accuracy ≈0.5 ED).

⸻

Summary:

Models using fewer vision tokens (right side) generally have worse accuracy, while those with more tokens per image (left side) perform better.
DeepSeek-OCR (Gundam-M 200dpi) leads overall in accuracy, while SmolDocling is the smallest and least accurate.

2522

Reposted by yosé ✧⁠◝⁠(⁠⁰⁠▿⁠⁰⁠)⁠◜⁠✧

Simon Willison @simonwillison.net · 22/07/2025

Mistral shared what look like the most detailed numbers yet for the environmental impact of training a frontier LLM - their Mistral Large 2 used 20,4 ktCO₂e and 281,000 m3 of water I'd love to see numbers like that provided in context though, hard to evaluate alone simonwillison.net/2025/Jul/22/...

simonwillison.net

Our contribution to a global environmental standard for AI

Mistral have released environmental impact numbers for their largest model, Mistral Large 2, in more detail than I have seen from any of the other large AI labs. The methodology …

77519