Who's Who of AI

Rohan Paul

180 trust @rohanpaul_ai · 154,262 followers
Why they matter

Tracked through public AI activity and peer connections inside the directory.

AI signals
31
past 30d
Sources
23
distinct domains
Discussões
0
past 30d
Latest signal
2h ago
View every signal from Rohan Paul →
What they're sharing

Articles & links

https://t.co/aHQsXKkYUt https://t.co/UvVzUtPytj

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 · ♥ 0 ↻ 0 ↩ 0 · 8 from the directory shared this · 41d ago

https://t.co/dT7kLdY8Hb https://t.co/8nZSnzI6OE

Start building with Nano Banana 2 Lite and Gemini Omni Flash blog.google
AI Weekly's analysis →
  • Nano Banana 2 Lite generates images in four seconds at $0.034 per 1,000 images, targeting high-volume drafting and iteration over premium creative output.
  • Gemini Omni Flash conversational editing covers character swaps, relighting, and style transfers at $0.10 per second for clips up to 10 seconds.
  • C2PA content credentials and SynthID watermarks ship on by default for both models, putting Google ahead on enterprise content provenance at launch.
Read full analysis →
View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 4 from the directory shared this · 41d ago

https://t.co/oZZqFEhiaX

ft.com
View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 5 from the directory shared this · 2d ago

https://t.co/dI43qO0rV9

Metis: Memory Foundation Model arxiv.org
AI Weekly's analysis →
  • Metis introduces a persistent, dynamically evolving memory state built into the model backbone rather than as an external retrieval module.
  • The authors train Metis at 4B, 9B, and 27B parameters on a Qwen3.5 backbone using 8 H100 GPUs and around 406M synthesized tokens.
  • Metis-27B reports 73.77% average on the paper's constructed test set versus 1.69% for the baseline Qwen3.5-27B without context.
Read full analysis →
View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 3 from the directory shared this · 1d ago

https://t.co/RrNuNv0zqT

From Brain Waves to Words: Brain2Qwerty Offers a New Path to Communication Without Surgery ai.meta.com
AI Weekly's analysis →
  • Brain2Qwerty v2 reaches 61% word accuracy decoding typed sentences from MEG recordings, up from roughly 8% for prior non-invasive methods.
  • The system was trained on around 22,000 sentences from nine volunteer participants, each wearing MEG for 10 hours while actively typing.
  • Meta is releasing full training code for v1 and v2, alongside a Basque Center dataset and a $5 million Digital Brain Project open-data fund.
Read full analysis →
View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 3 from the directory shared this · 41d ago