ax@ax-radar:~/curated $ grep -l 'curated=true' sources/
33 srcsignal 72%cycle 04:32

curated · 2026-08-20

11 items · updated 3m ago
2026-08-20 · Thu
17:56
33d ago
AI HOT (Curated Pool)· aihot-apiZH17:56 · 08·20
Anthropic’s approach to teaching and learning AI
Anthropic published a blog post about how it teaches users to use AI effectively. The centerpiece is Claude Academy, a learning platform for users of all skill levels. The post does not disclose course details, user metrics, or effectiveness data—it reads more as a brand philosophy statement than a technical deep dive.
#Anthropic#Claude Academy
editor take
Anthropic launched Claude Academy, but the post is brand philosophy with zero course details or user metrics.
HKR breakdown
hook knowledge resonance
open source
39
SCORE
H0·K0·R0
16:52
33d ago
AI HOT (Curated Pool)· aihot-apiZH16:52 · 08·20
Liquid AI releases DSpark draft models for LFM2.5, up to 3.18× faster inference
Liquid AI released DSpark draft model checkpoints for three LFM2.5 sizes—1.2B, 2.6B, and 8B-A1B—on Hugging Face. DSpark is a speculative decoding method: a lightweight draft model proposes multiple candidate tokens in one forward pass, and the target model verifies them all at once, amortizing weight-loading cost. Throughput improves up to 3.18× on GPU and 2.87× on-device; function-calling latency drops 57% on average for the 2.6B model. Output quality stays the same with minimal memory overhead. llama.cpp and SGLang support it natively.
#Liquid AI#Hugging Face#llama.cpp#Open source
editor take
Liquid AI released speculative decoding draft models for LFM2.5, up to 3.18× GPU throughput with native llama.cpp and SGLang support.
HKR breakdown
hook knowledge resonance
open source
72
SCORE
H1·K1·R0
16:02
33d ago
AI HOT (Curated Pool)· aihot-apiZH16:02 · 08·20
Mistral launches Agentic Search for multi-step retrieval on complex documents
Mistral's new Agentic Search breaks queries into steps—decompose, retrieve, verify, synthesize—instead of a single search pass. The company claims higher accuracy on complex document queries but the post doesn't disclose benchmark numbers, baselines, pricing, or latency figures. Available via API.
#Mistral
editor take
Mistral's search now decomposes queries into steps, but no benchmarks or pricing yet—I'd hold off.
HKR breakdown
hook knowledge resonance
open source
68
SCORE
H1·K0·R0
16:00
33d ago
AI HOT (Curated Pool)· aihot-apiZH16:00 · 08·20
AlloyDB ScaNN scales vector search to 10 billion vectors with a four-level tree index
Google Cloud's AlloyDB now supports vector search up to 10 billion vectors by adding a four-level tree to its ScaNN index. The post doesn't disclose exact latency or recall improvements, only the index architecture change. For RAG or large-scale vector retrieval, this means AlloyDB can handle 10B embeddings natively without an external vector database.
#Google Cloud#AlloyDB#ScaNN
editor take
AlloyDB adds a fourth tree level to ScaNN, hitting 10B native vector search without a separate vector DB.
HKR breakdown
hook knowledge resonance
open source
39
SCORE
H0·K0·R0
13:26
33d ago
AI HOT (Curated Pool)· aihot-apiZH13:26 · 08·20
The Claude Code Guide For Startups
Anthropic publishes an official guide for startups using Claude Code, covering five key rules and founder insights. The post does not disclose the specific rules or case details, only confirming it's a practical guide for startup teams.
#Anthropic#Claude Code
editor take
Anthropic posted a Claude Code guide for startups, but the body doesn't list the five rules—only the title has substance.
HKR breakdown
hook knowledge resonance
open source
55
SCORE
H0·K0·R0
02:00
34d ago
AI HOT (Curated Pool)· aihot-apiZH02:00 · 08·20
Ling-3.0 tiny & flash Base Models Open-Sourced
Ling-3.0 tiny and flash base models are now open-sourced, but the article body is blocked by WeChat, so no details on parameters, training data, or benchmarks are available. Only the model names and the open-source release are confirmed.
#Ling#Open source
editor take
Ling-3.0 tiny & flash base models are open-sourced, but the article body is blocked by WeChat—no parameters or benchmarks yet.
HKR breakdown
hook knowledge resonance
open source
25
SCORE
H0·K0·R0
00:36
34d ago
AI HOT (Curated Pool)· aihot-apiZH00:36 · 08·20
MiniMax Design Launch: From Manipulating Pixels to Manipulating Semantics and Expressing Intent
MiniMax launched Design, shifting design from pixel tweaking to semantic and intent manipulation. The post is blocked by WeChat, so no details on features, pricing, or release date. The title suggests users describe design intent in natural language and the model generates UI. I'd take it with a grain of salt: only the title is confirmed, actual quality and learning curve are unknown.
#MiniMax
editor take
MiniMax launched Design, claiming designers describe intent instead of tweaking pixels. The post is blocked by WeChat, so zero details on features or quality. I'd hold off.
HKR breakdown
hook knowledge resonance
open source
39
SCORE
H0·K0·R0
00:00
34d ago
AI HOT (Curated Pool)· aihot-apiZH00:00 · 08·20
Claude Python SDK hits v1.0, swaps httpx for httpx2
Anthropic released Python SDK v1.0, switching the HTTP layer from httpx to httpx2, a maintained, API-compatible fork. You can build custom http_client, Timeout, and transport objects from httpx2; call httpx2.alias_httpx() at startup if you rely on tracing or mocking libraries that patch httpx. v1.0 requires Python 3.10+, removes the legacy Text Completions API, temperature/top_p/top_k params on Messages methods, and the client-side compaction_control. Async .with_raw_response results now need await response.parse(); AnthropicBedrock errors when no AWS region is configured instead of defaulting to us-east-1. A migration guide with before-and-after snippets is linked.
#Anthropic#Claude
editor take
Python SDK v1.0 swaps httpx for httpx2 — API-compatible fork, but tracing/mocking libs need httpx2.alias_httpx() at startup.
HKR breakdown
hook knowledge resonance
open source
25
SCORE
H0·K0·R0

more

feeds

admin