Anthropic is gaining exponentially less capability per billion dollars spent. It is now 8× as expensive to boost AI capabilities as it was 2 years ago, worsening on a log scale.
Key points
Cost increase: 8× higher per billion dollars than 2 years ago.
Trend: Capability gain per dollar is decreasing exponentially, worsening on a log scale.
Sources
Symbolica raises $31M, deck seeks $8M at $120M valuation
Symbolica raised $31 million for symbolic AI with no product. The author’s deck is raising up to $8 million at a $120 million post-money cap. Other cited raises include $64 million for Axiom at seed and a claim of HVM code on thousands of GPU cores.
Key points
Symbolica raised $31 million for symbolic AI with no product.
Deck is raising up to $8 million at a $120 million post-money cap.
Sources
Specialist open-source hardware and code released
Specialist open-source hardware, data pipeline, training, and deployment code released. Documentation created with @t_k_233 covering design choices and lessons.
Key points
Specialist: fully open-source hardware, data pipeline, training, and deployment code
Documentation of every design choice and lesson with @t_k_233
Sources
- Original post
- specialist-ai.github.io - Linked in the post
Sakana AI App beta testing recruitment announced
Sakana AI App (working title) is seeking beta testers in Japan via a sign-up form. Early testers will help shape the app before launch.
Key points
Beta testers needed in Japan
Sign-up form provided for early testers
Sources
- Original post
- forms.gle - Linked in the post
Virgil legal and back-office support service highlighted
Virgil offers unified legal and back-office support for startups, promising lower costs and faster delivery. A client report says Virgil saved over $30k in the first two months. The post advises founders to use Virgil instead of investors' preferred lawyers.
Key points
Client claim: saved over $30k in the first two months.
Service scope: accounting, legal, compliance, and people ops.
Sources
- Original post
- Virgil - Linked page
Delvetown wiki built on Tangled git forge for ATproto
Delvetown wiki built on Tangled, the git forge for ATproto, lets anyone contribute via their delvetown address. Agents and people collaborated to create it.
Key points
Collaboration: agents and people collaborated to make the wiki.
Contribution: anyone can contribute through their delvetown address.
Sources
- Original post
- delvetown.tngl.io - Linked in the post
Qwen3.8-Flash-Next-FP4-Local served via vLLM
Qwen3.8-Flash-Next-FP4-Local is being served with vLLM as a local agent. The user is generating a lesson and a personal assistant using this setup.
Key points
Model: nvidia/Qwen3.8-Flash-Next-FP4-Local
Serving: vLLM
