AI, But Simple

The AI intelligence layer.

See what's emerging in AI before everyone else. PRO watches thousands of techniques, papers, code, jobs, and interview questions to rank what matters.

One subscription, four tracks.

Inside the intelligence layer.

Field

Where AI is moving.

1,000+ techniques and topics, new, rising, or established, measured daily across 17 fields.

AllLLMs & fine-tuningAgentsReasoningRetrieval / RAGDiffusionReinforcement learningRobotics NEW5OpenClawAgentsReActLLM-as-judge13LIBERORobotics & embodiedOpenVLA5Agent memoryRetrieval / RAGWATCHAgent memory systems RISING110LoRALLMs & fine-tuningADOPTLoRA fine-tuningQLoRA114GRPOReinforcement learningADOPTSupervised fine-tuningverl141AI agentsAgentsWATCHModel Context Protocol AllLLMs & fine-tuningAgentsReasoning RISING 110LoRALLMs & fine-tuningADOPT 114GRPOReinforcement learningADOPT 141AI agentsAgentsWATCH 107LLM post-trainingLLMs & fine-tuningWATCH NEW 5OpenClawAgents 13LIBERORobotics & embodied
Subject pages

One page per technique.

What it is, where it's moving, who's hiring for it, and the verdict, for your role.

Speculative decoding TECHNIQUE, INFERENCE & SERVING RELATED EAGLE-3DFlashMulti-Token Prediction steady, fewer than last week, in papers, code and jobs MENTIONS AUG 11 AUG 23 SEP 5 64 EVIDENCE 12 REPOS 9 JOBS ASKING 8 QUESTIONS FOR ENGINEERS ADOPT Drop-in for latency-bound serving when a draft model exists. COST  LOW    PAYOFF  2 TO 3X Speculative decoding TECHNIQUE, INFERENCE & SERVING RELATED EAGLE-3DFlashMulti-Token Prediction steady, fewer than last week, in papers and code MENTIONS AUG 11 AUG 23 SEP 5 FOR ENGINEERS ADOPT Drop-in for latency-bound serving when a draft model exists. COST  LOW    PAYOFF  2 TO 3X
Search

One query across everything.

Papers, code, news, jobs, and interview questions, with the subject on top.

speculative decoding Search Speculative decodingTECHNIQUE EAGLE-3DFlashMulti-Token Prediction 528 RESULTS Papers109 Code104 Pulse100 Jobs100 Interviews100 Questions14 PAPER Verification-Aware Training for Speculative Decoding CODE patchy631/time-to-first-token speculative decoding Search Speculative decodingTECHNIQUE EAGLE-3DFlashMulti-Token Prediction 528 RESULTS Papers109 Code104 Pulse100 Jobs100 Interviews100 Questions14
Interview tracker

The questions companies actually ask.

Sourced from over 100 companies with answers and working code, grouped by topic.

All1,148LLMs & fine-tuning148ML system design121ML fundamentals115Deep learning53 What's the trade-off between bias and variance? ML FUNDAMENTALS MEDIUM error = bias**2 + variance + noise REVIEW SOLID How do you combat the curse of dimensionality? Why do we scale attention by the square root of d_k? Compare batch norm and layer norm. All1,148LLMs & fine-tuning148ML system design121 What's the trade-off between bias and variance? ML FUNDAMENTALS MEDIUM error = bias**2 + variance + noise REVIEW SOLID How do you combat the curse of dimensionality? Why do we scale attention by root d_k? Compare batch norm and layer norm.
Job trends

Demand, quantified.

4,400+ live postings, ranked by role and company demand, re-scored daily.

DEMAND BY ROLE Sales & GTM 611+13 Solutions & forward-depl. 409-4 Infrastructure & platform 301+13 Agent roles 189+8 Product & design 151+3 Research engineer 81+3 Inference & serving 71+1 Safety & alignment 66-1 Training & pre-training 48+1 POSTINGS BY COMPANY Databricks500 OpenAI500 Anthropic433 Harvey352 Palantir281 xAI230 Sales & GTM 611 +13 Solutions & forward-deployed 409 -4 Infrastructure & platform 301 Agent roles 189 Product & design 160 Research engineer 142 Inference & serving 118 Safety & alignment 96
Pulse

The week's signal, grouped.

Releases, agents, benchmarks, funding: the news that matters, clustered by topic.

All topics100Models & releases19Agents6Coding & SWE5Reasoning4Funding3 +312 A 7B open model closes the gap on coding evals arxiv.org, 2h ago +198 Agent frameworks converge on a shared tool protocol github.com, 5h ago +154 Serving costs fall as quantized inference goes default news.ycombinator.com, 1d ago +139 Open-weight AI is having its Kubernetes moment tobi.knaup.me, 7h ago TOP SOURCES arxiv.org 28 github.com 21 huggingface.co 14 news.ycombinator.com 11 openai.com 7 All topics100Models & releases19Agents6 +312 A 7B open model closes the gap on coding evals arxiv.org, 2h ago +198 Agent frameworks converge on a tool protocol github.com, 5h ago +154 Serving costs fall as quantized inference wins news.ycombinator.com, 1d ago +139 Open-weight AI is having its Kubernetes moment tobi.knaup.me, 7h ago

One layer. Eight systems.

Topics & techniques

Search 1,000+ subjects across 17 fields, ranked by research momentum.

Paper tracker

The research that moved this week.

GitHub & model tracker

Trending repos and weights, attached to their papers.

+312A 7B open model closes the gap on coding evals

Pulse

The week's AI news clustered by topic.

Job trends

Live postings, ranked by role and by company demand.

Roadmaps & career paths

Fundamentals to offer, from the question bank.

Code vault

Every notebook from the newsletter.

Interview question tracker

Real questions from top companies, answered with code. Always current.

Simple, like the newsletter.

Free
$0
Free, forever.
  • The visual issue, weekly
  • Limited archive
  • Read on the web or by email
Subscribe free
PRO2 mo free
$10 / month, billed annually
Unlock AI research intelligence
  • Everything in Free
  • The AI intelligence layer
  • Signals: verdicts on topics and techniques
  • Code tutorial with every issue (Code vault)
  • Interview tracker and roadmaps
  • Full archive access
  • Ad-free issues
Upgrade to PRO

Questions, answered simply.

What stays free? +

The weekly visual newsletter stays free. PRO adds the intelligence layer: the field ranked daily, Signals on every technique, career intelligence, and the full archive.

How is this different from following arXiv or another newsletter? +

A feed shows you everything and ranks nothing. PRO watches the whole field, shows what actually moved, and connects the papers, repos, postings, and interview questions to each other, then narrows it to what matters for the work you do.

How often does the intelligence layer update? +

Continuously. The pipeline pulls from hundreds of sources every day and re-ranks on each run, so what you see reflects the current state of AI.

How is the data selected? +

We monitor hundreds of sources, then a proprietary ranking algorithm using 50+ factors decides what surfaces. Items are attributed, and your preferences decide what's displayed.

Get started with PRO.

Start now

or read a free issue first