Who's Who of AI

Rafael Pinto

226 trust researcher @rcpinto.bsky.social · 4,701 followers
AI research
Why they matter

Researcher with public evidence across AI research.

AI signals
2
past 30d
Sources
2
distinct domains
Discusiones
8
past 30d
Latest signal
20d ago
View every signal from Rafael Pinto →
CS / AI / ML PhD / Professor. 🇧🇷 ♾️ Working on meta-learning, continual learning, control, evo, RL, game AI. Hobbyist game dev (@megamanbeyond.com) Local news and useless facts about my life in PT-BR. EN otherwise.
What they're sharing

Articles & links

www.anthropic.com/news/claude-...

Introducing Claude Sonnet 5 \ Anthropic anthropic.com
AI Weekly's analysis →
  • Anthropic released Claude Sonnet 5 on June 30, 2026, calling it 'the most agentic Sonnet model yet' and pitching it for autonomous browser and terminal use.
  • Through August 31, 2026 Sonnet 5 costs $2 per million input tokens and $10 per million output, then steps to standard rates of $3 and $15.
  • A new tokenizer means the same input can map to roughly 1.0 to 1.35 times more tokens than prior Anthropic models, partly offsetting the headline discount.
Read full analysis →
View on Bluesky · ♥ 7 ↻ 1 ↩ 1 · 8 from the directory shared this · 28d ago
↻ Rafael Pinto reposted
@unsloth.ai

Qwen3.6 now runs 2x faster with MTP GGUFs! Run locally on just 18GB RAM. ⚡️ MTP enables Qwen3.6 to generate ~1.4–2.2× faster with no accuracy change. Qwen3.6-27B MTP runs at 160 tokens/s. 35B-A3B reaches 240 t/s. GGUFs: huggingface.co/unsloth/Qwen... Guide: unsloth.ai/docs/mod…

unsloth/Qwen3.6-27B-MTP-GGUF · Hugging Face huggingface.co View on Bluesky →
↻ Rafael Pinto reposted
@unsloth.ai

DiffusionGemma can now run at 2000+ tokens/sec! ⚡ We made local DiffusionGemma inference 1.8× faster. Run it on 18GB RAM via Unsloth Studio. GitHub: github.com/unslothai/un... Guide: unsloth.ai/docs/models/...

DiffusionGemma - How to Run Locally | Unsloth Documentation unsloth.ai
AI Weekly's analysis →
  • DiffusionGemma 26B-A4B generates text in parallel using diffusion refinement rather than token-by-token decoding.
  • The 4-bit quantized model requires only 18 GB RAM and reaches 2,000+ tokens per second on an RTX 6000.
  • Speed comes at a cost: AIME 2026 accuracy drops from 88.3% on standard Gemma 4 to 69.1% on DiffusionGemma.
Read full analysis →
 View on Bluesky →
Their own posts

Recent commentary

GPT 5h limit removed for plus and business!

View on Bluesky · ♥ 4 ↻ 0 ↩ 1 · 16d ago

I asked my wife to stop me from subscribing to GPT Pro. Her answer was pretty convincing: "Is it going to improve your work so much that you won’t need to upgrade to a higher plan, kind of like that Black Mirror episode?"

View on Bluesky · ♥ 2 ↻ 0 ↩ 0 · 22d ago

"I hate AI because it steals our noble work. Oh, is it just stupid code? That's fine then."

View on Bluesky · ♥ 2 ↻ 0 ↩ 0 · 51d ago

Just heard this one: "Art is about expressing thought; the artist is a thinker before anything else. The technique comes only at the materialization step. That's what people who think AI Art is Art don't understand." Sure, nice that you admit the prompter is an artist.

View on Bluesky · ♥ 1 ↻ 0 ↩ 0 · 60d ago

You might think this is a transphobic comment, but it is about AI.

View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 44d ago

AI backlash is all about narcisistic injury and ego protection.

View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 64d ago
Their network

In Rafael Pinto's orbit

Center = Rafael Pinto. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.